Merge branch 'release-v0.21.0' of github.com:matrix-org/synapse

Make get_joined_users faster when we have delta state

CHANGES.rst

@@ -1,3 +1,196 @@
+Changes in synapse v0.21.0 (2017-05-18)
+=======================================
+
+No changes since v0.21.0-rc3
+
+
+Changes in synapse v0.21.0-rc3 (2017-05-17)
+===========================================
+
+Features:
+
+* Add per user rate-limiting overrides (PR #2208)
+* Add config option to limit maximum number of events requested by ``/sync``
+  and ``/messages`` (PR #2221) Thanks to @psaavedra!
+
+
+Changes:
+
+* Various small performance fixes (PR #2201, #2202, #2224, #2226, #2227, #2228,
+  #2229)
+* Update username availability checker API (PR #2209, #2213)
+* When purging, don't de-delta state groups we're about to delete (PR #2214)
+* Documentation to check synapse version (PR #2215) Thanks to @hamber-dick!
+* Add an index to event_search to speed up purge history API (PR #2218)
+
+
+Bug fixes:
+
+* Fix API to allow clients to upload one-time-keys with new sigs (PR #2206)
+
+
+Changes in synapse v0.21.0-rc2 (2017-05-08)
+===========================================
+
+Changes:
+
+* Always mark remotes as up if we receive a signed request from them (PR #2190)
+
+
+Bug fixes:
+
+* Fix bug where users got pushed for rooms they had muted (PR #2200)
+
+
+Changes in synapse v0.21.0-rc1 (2017-05-08)
+===========================================
+
+Features:
+
+* Add username availability checker API (PR #2183)
+* Add read marker API (PR #2120)
+
+
+Changes:
+
+* Enable guest access for the 3pl/3pid APIs (PR #1986)
+* Add setting to support TURN for guests (PR #2011)
+* Various performance improvements (PR #2075, #2076, #2080, #2083, #2108,
+  #2158, #2176, #2185)
+* Make synctl a bit more user friendly (PR #2078, #2127) Thanks @APwhitehat!
+* Replace HTTP replication with TCP replication (PR #2082, #2097, #2098,
+  #2099, #2103, #2014, #2016, #2115, #2116, #2117)
+* Support authenticated SMTP (PR #2102) Thanks @DanielDent!
+* Add a counter metric for successfully-sent transactions (PR #2121)
+* Propagate errors sensibly from proxied IS requests (PR #2147)
+* Add more granular event send metrics (PR #2178)
+
+
+
+Bug fixes:
+
+* Fix nuke-room script to work with current schema (PR #1927) Thanks
+  @zuckschwerdt!
+* Fix db port script to not assume postgres tables are in the public schema
+  (PR #2024) Thanks @jerrykan!
+* Fix getting latest device IP for user with no devices (PR #2118)
+* Fix rejection of invites to unreachable servers (PR #2145)
+* Fix code for reporting old verify keys in synapse (PR #2156)
+* Fix invite state to always include all events (PR #2163)
+* Fix bug where synapse would always fetch state for any missing event (PR #2170)
+* Fix a leak with timed out HTTP connections (PR #2180)
+* Fix bug where we didn't time out HTTP requests to ASes  (PR #2192)
+
+
+Docs:
+
+* Clarify doc for SQLite to PostgreSQL port (PR #1961) Thanks @benhylau!
+* Fix typo in synctl help (PR #2107) Thanks @HarHarLinks!
+* ``web_client_location`` documentation fix (PR #2131) Thanks @matthewjwolff!
+* Update README.rst with FreeBSD changes (PR #2132) Thanks @feld!
+* Clarify setting up metrics (PR #2149) Thanks @encks!
+
+
+Changes in synapse v0.20.0 (2017-04-11)
+=======================================
+
+Bug fixes:
+
+* Fix joining rooms over federation where not all servers in the room saw the
+  new server had joined (PR #2094)
+
+
+Changes in synapse v0.20.0-rc1 (2017-03-30)
+===========================================
+
+Features:
+
+* Add delete_devices API (PR #1993)
+* Add phone number registration/login support (PR #1994, #2055)
+
+
+Changes:
+
+* Use JSONSchema for validation of filters. Thanks @pik! (PR #1783)
+* Reread log config on SIGHUP (PR #1982)
+* Speed up public room list (PR #1989)
+* Add helpful texts to logger config options (PR #1990)
+* Minor ``/sync`` performance improvements. (PR #2002, #2013, #2022)
+* Add some debug to help diagnose weird federation issue (PR #2035)
+* Correctly limit retries for all federation requests (PR #2050, #2061)
+* Don't lock table when persisting new one time keys (PR #2053)
+* Reduce some CPU work on DB threads (PR #2054)
+* Cache hosts in room (PR #2060)
+* Batch sending of device list pokes (PR #2063)
+* Speed up persist event path in certain edge cases (PR #2070)
+
+
+Bug fixes:
+
+* Fix bug where current_state_events renamed to current_state_ids (PR #1849)
+* Fix routing loop when fetching remote media (PR #1992)
+* Fix current_state_events table to not lie (PR #1996)
+* Fix CAS login to handle PartialDownloadError (PR #1997)
+* Fix assertion to stop transaction queue getting wedged (PR #2010)
+* Fix presence to fallback to last_active_ts if it beats the last sync time.
+  Thanks @Half-Shot! (PR #2014)
+* Fix bug when federation received a PDU while a room join is in progress (PR
+  #2016)
+* Fix resetting state on rejected events (PR #2025)
+* Fix installation issues in readme. Thanks @ricco386 (PR #2037)
+* Fix caching of remote servers' signature keys (PR #2042)
+* Fix some leaking log context (PR #2048, #2049, #2057, #2058)
+* Fix rejection of invites not reaching sync (PR #2056)
+
+
+
+Changes in synapse v0.19.3 (2017-03-20)
+=======================================
+
+No changes since v0.19.3-rc2
+
+
+Changes in synapse v0.19.3-rc2 (2017-03-13)
+===========================================
+
+Bug fixes:
+
+* Fix bug in handling of incoming device list updates over federation.
+
+
+
+Changes in synapse v0.19.3-rc1 (2017-03-08)
+===========================================
+
+Features:
+
+* Add some administration functionalities. Thanks to morteza-araby! (PR #1784)
+
+
+Changes:
+
+* Reduce database table sizes (PR #1873, #1916, #1923, #1963)
+* Update contrib/ to not use syutil. Thanks to andrewshadura! (PR #1907)
+* Don't fetch current state when sending an event in common case (PR #1955)
+
+
+Bug fixes:
+
+* Fix synapse_port_db failure. Thanks to Pneumaticat! (PR #1904)
+* Fix caching to not cache error responses (PR #1913)
+* Fix APIs to make kick & ban reasons work (PR #1917)
+* Fix bugs in the /keys/changes api (PR #1921)
+* Fix bug where users couldn't forget rooms they were banned from (PR #1922)
+* Fix issue with long language values in pushers API (PR #1925)
+* Fix a race in transaction queue (PR #1930)
+* Fix dynamic thumbnailing to preserve aspect ratio. Thanks to jkolo! (PR
+  #1945)
+* Fix device list update to not constantly resync (PR #1964)
+* Fix potential for huge memory usage when getting device that have
+  changed (PR #1969)
+
+
+
 Changes in synapse v0.19.2 (2017-02-20)
 =======================================
 

README.rst

@@ -20,7 +20,7 @@ The overall architecture is::
              https://somewhere.org/_matrix      https://elsewhere.net/_matrix
 
 ``#matrix:matrix.org`` is the official support room for Matrix, and can be
-accessed by any client from https://matrix.org/docs/projects/try-matrix-now or
+accessed by any client from https://matrix.org/docs/projects/try-matrix-now.html or
 via IRC bridge at irc://irc.freenode.net/matrix.
 
 Synapse is currently in rapid development, but as of version 0.5 we believe it
@@ -68,7 +68,7 @@ or mandatory service provider in Matrix, unlike WhatsApp, Facebook, Hangouts,
 etc.
 
 We'd like to invite you to join #matrix:matrix.org (via
-https://matrix.org/docs/projects/try-matrix-now), run a homeserver, take a look
+https://matrix.org/docs/projects/try-matrix-now.html), run a homeserver, take a look
 at the `Matrix spec <https://matrix.org/docs/spec>`_, and experiment with the
 `APIs <https://matrix.org/docs/api>`_ and `Client SDKs
 <http://matrix.org/docs/projects/try-matrix-now.html#client-sdks>`_.
@@ -84,6 +84,7 @@ Synapse Installation
 Synapse is the reference python/twisted Matrix homeserver implementation.
 
 System requirements:
+
 - POSIX-compliant system (tested on Linux & OS X)
 - Python 2.7
 - At least 1GB of free RAM if you want to join large public rooms like #matrix:matrix.org
@@ -108,10 +109,10 @@ Installing prerequisites on ArchLinux::
     sudo pacman -S base-devel python2 python-pip \
                    python-setuptools python-virtualenv sqlite3
 
-Installing prerequisites on CentOS 7::
+Installing prerequisites on CentOS 7 or Fedora 25::
 
     sudo yum install libtiff-devel libjpeg-devel libzip-devel freetype-devel \
-                     lcms2-devel libwebp-devel tcl-devel tk-devel \
+                     lcms2-devel libwebp-devel tcl-devel tk-devel redhat-rpm-config \
                      python-virtualenv libffi-devel openssl-devel
     sudo yum groupinstall "Development Tools"
 
@@ -146,6 +147,7 @@ To install the synapse homeserver run::
 
     virtualenv -p python2.7 ~/.synapse
     source ~/.synapse/bin/activate
+    pip install --upgrade pip
     pip install --upgrade setuptools
     pip install https://github.com/matrix-org/synapse/tarball/master
 
@@ -228,6 +230,7 @@ To get started, it is easiest to use the command line to register new users::
     New user localpart: erikj
     Password:
     Confirm password:
+    Make admin [no]:
     Success!
 
 This process uses a setting ``registration_shared_secret`` in
@@ -243,6 +246,25 @@ Setting up a TURN server
 For reliable VoIP calls to be routed via this homeserver, you MUST configure
 a TURN server.  See `<docs/turn-howto.rst>`_ for details.
 
+IPv6
+----
+
+As of Synapse 0.19 we finally support IPv6, many thanks to @kyrias and @glyph
+for providing PR #1696.
+
+However, for federation to work on hosts with IPv6 DNS servers you **must**
+be running Twisted 17.1.0 or later - see https://github.com/matrix-org/synapse/issues/1002
+for details.  We can't make Synapse depend on Twisted 17.1 by default
+yet as it will break most older distributions (see https://github.com/matrix-org/synapse/pull/1909)
+so if you are using operating system dependencies you'll have to install your
+own Twisted 17.1 package via pip or backports etc.
+
+If you're running in a virtualenv then pip should have installed the newest
+Twisted automatically, but if your virtualenv is old you will need to manually
+upgrade to a newer Twisted dependency via:
+
+    pip install Twisted>=17.1.0
+
 
 Running Synapse
 ===============
@@ -321,7 +343,7 @@ Debian
 
 Matrix provides official Debian packages via apt from http://matrix.org/packages/debian/.
 Note that these packages do not include a client - choose one from
-https://matrix.org/docs/projects/try-matrix-now/ (or build your own with one of our SDKs :)
+https://matrix.org/docs/projects/try-matrix-now.html (or build your own with one of our SDKs :)
 
 Fedora
 ------
@@ -332,10 +354,12 @@ https://obs.infoserver.lv/project/monitor/matrix-synapse
 ArchLinux
 ---------
 
-The quickest way to get up and running with ArchLinux is probably with Ivan
-Shapovalov's AUR package from
-https://aur.archlinux.org/packages/matrix-synapse/, which should pull in all
-the necessary dependencies.
+The quickest way to get up and running with ArchLinux is probably with the community package
+https://www.archlinux.org/packages/community/any/matrix-synapse/, which should pull in most of
+the necessary dependencies. If the default web client is to be served (enabled by default in
+the generated config),
+https://www.archlinux.org/packages/community/any/python2-matrix-angular-sdk/ will also need to
+be installed. 
 
 Alternatively, to install using pip a few changes may be needed as ArchLinux
 defaults to python 3, but synapse currently assumes python 2.7 by default:
@@ -372,7 +396,7 @@ FreeBSD
 
 Synapse can be installed via FreeBSD Ports or Packages contributed by Brendan Molloy from:
 
- - Ports: ``cd /usr/ports/net/py-matrix-synapse && make install clean``
+ - Ports: ``cd /usr/ports/net-im/py-matrix-synapse && make install clean``
  - Packages: ``pkg install py27-matrix-synapse``
 
 
@@ -809,7 +833,7 @@ directory of your choice::
 Synapse has a number of external dependencies, that are easiest
 to install using pip and a virtualenv::
 
-    virtualenv env
+    virtualenv -p python2.7 env
     source env/bin/activate
     python synapse/python_dependencies.py | xargs pip install
     pip install lxml mock

UPGRADE.rst

@@ -28,6 +28,15 @@ running:
     git pull
     # Update the versions of synapse's python dependencies.
     python synapse/python_dependencies.py | xargs -n1 pip install --upgrade
+	
+To check whether your update was sucessfull, run:
+
+.. code:: bash
+
+	 # replace your.server.domain with ther domain of your synaspe homeserver
+	 curl https://<your.server.domain>/_matrix/federation/v1/version 
+
+So for the Matrix.org HS server the URL would be: https://matrix.org/_matrix/federation/v1/version.
 
 
 Upgrading to v0.15.0

contrib/cmdclient/console.py

@@ -32,7 +32,7 @@ import urlparse
 import nacl.signing
 import nacl.encoding
 
-from syutil.crypto.jsonsign import verify_signed_json, SignatureVerifyException
+from signedjson.sign import verify_signed_json, SignatureVerifyException
 
 CONFIG_JSON = "cmdclient_config.json"
 

contrib/cmdclient/http.py

@@ -36,15 +36,13 @@ class HttpClient(object):
                 the request body. This will be encoded as JSON.
 
         Returns:
-            Deferred: Succeeds when we get *any* HTTP response.
-
-            The result of the deferred is a tuple of `(code, response)`,
-            where `response` is a dict representing the decoded JSON body.
+            Deferred: Succeeds when we get a 2xx HTTP response. The result
+            will be the decoded JSON body.
         """
         pass
 
     def get_json(self, url, args=None):
-        """ Get's some json from the given host homeserver and path
+        """ Gets some json from the given host homeserver and path
 
         Args:
             url (str): The URL to GET data from.
@@ -54,10 +52,8 @@ class HttpClient(object):
                 and *not* a string.
 
         Returns:
-            Deferred: Succeeds when we get *any* HTTP response.
-
-            The result of the deferred is a tuple of `(code, response)`,
-            where `response` is a dict representing the decoded JSON body.
+            Deferred: Succeeds when we get a 2xx HTTP response. The result
+            will be the decoded JSON body.
         """
         pass
 
@@ -214,4 +210,4 @@ class _JsonProducer(object):
         pass
 
     def stopProducing(self):
-        pass
+        pass

contrib/example_log_config.yaml

@@ -0,0 +1,50 @@
+# Example log_config file for synapse. To enable, point `log_config` to it in 
+# `homeserver.yaml`, and restart synapse.
+#
+# This configuration will produce similar results to the defaults within 
+# synapse, but can be edited to give more flexibility.
+
+version: 1
+
+formatters:
+  fmt:
+    format: '%(asctime)s - %(name)s - %(lineno)d - %(levelname)s - %(request)s- %(message)s'
+
+filters:
+  context:
+    (): synapse.util.logcontext.LoggingContextFilter
+    request: ""
+
+handlers:
+  # example output to console
+  console:
+    class: logging.StreamHandler
+    filters: [context]
+
+  # example output to file - to enable, edit 'root' config below.
+  file:
+    class: logging.handlers.RotatingFileHandler
+    formatter: fmt
+    filename: /var/log/synapse/homeserver.log
+    maxBytes: 100000000
+    backupCount: 3
+    filters: [context]
+
+
+root:
+    level: INFO
+    handlers: [console] # to use file handler instead, switch to [file]
+    
+loggers:
+    synapse:
+        level: INFO
+
+    synapse.storage.SQL:
+        # beware: increasing this to DEBUG will make synapse log sensitive
+        # information such as access tokens.
+        level: INFO
+
+    # example of enabling debugging for a component:
+    #
+    # synapse.federation.transport.server:
+    #    level: DEBUG

contrib/systemd/synapse.service

@@ -1,5 +1,5 @@
 # This assumes that Synapse has been installed as a system package
-# (e.g. https://aur.archlinux.org/packages/matrix-synapse/ for ArchLinux)
+# (e.g. https://www.archlinux.org/packages/community/any/matrix-synapse/ for ArchLinux)
 # rather than in a user home directory or similar under virtualenv.
 
 [Unit]

docs/CAPTCHA_SETUP.rst

@@ -25,6 +25,5 @@ Configuring IP used for auth
 The ReCaptcha API requires that the IP address of the user who solved the
 captcha is sent. If the client is connecting through a proxy or load balancer,
 it may be required to use the X-Forwarded-For (XFF) header instead of the origin
-IP address. This can be configured as an option on the home server like so::
-
-  captcha_ip_origin_is_x_forwarded: true
+IP address. This can be configured using the x_forwarded directive in the
+listeners section of the homeserver.yaml configuration file.

docs/admin_api/user_admin_api.rst

@@ -0,0 +1,73 @@
+Query Account
+=============
+
+This API returns information about a specific user account.
+
+The api is::
+
+    GET /_matrix/client/r0/admin/whois/<user_id>
+
+including an ``access_token`` of a server admin.
+
+It returns a JSON body like the following:
+
+.. code:: json
+
+    {
+        "user_id": "<user_id>",
+        "devices": {
+            "": {
+                "sessions": [
+                    {
+                        "connections": [
+                            {
+                                "ip": "1.2.3.4",
+                                "last_seen": 1417222374433,
+                                "user_agent": "Mozilla/5.0 ..."
+                            },
+                            {
+                                "ip": "1.2.3.10",
+                                "last_seen": 1417222374500,
+                                "user_agent": "Dalvik/2.1.0 ..."
+                            }
+                        ]
+                    }
+                ]
+            }
+        }
+    }
+
+``last_seen`` is measured in milliseconds since the Unix epoch.
+
+Deactivate Account
+==================
+
+This API deactivates an account. It removes active access tokens, resets the
+password, and deletes third-party IDs (to prevent the user requesting a
+password reset).
+
+The api is::
+
+    POST /_matrix/client/r0/admin/deactivate/<user_id>
+
+including an ``access_token`` of a server admin, and an empty request body.
+
+
+Reset password
+==============
+
+Changes the password of another user.
+
+The api is::
+
+    POST /_matrix/client/r0/admin/reset_password/<user_id>
+
+with a body of:
+
+.. code:: json
+
+   {
+       "new_password": "<secret>"
+   }
+
+including an ``access_token`` of a server admin.

docs/log_contexts.rst

@@ -1,10 +1,446 @@
-What do I do about "Unexpected logging context" debug log-lines everywhere?
+Log contexts
+============
 
-<Mjark> The logging context lives in thread local storage
-<Mjark> Sometimes it gets out of sync with what it should actually be, usually because something scheduled something to run on the reactor without preserving the logging context. 
-<Matthew> what is the impact of it getting out of sync? and how and when should we preserve log context?
-<Mjark> The impact is that some of the CPU and database metrics will be under-reported, and some log lines will be mis-attributed.
-<Mjark> It should happen auto-magically in all the APIs that do IO or otherwise defer to the reactor.
-<Erik> Mjark: the other place is if we branch, e.g. using defer.gatherResults
+.. contents::
 
-Unanswered: how and when should we preserve log context?
+To help track the processing of individual requests, synapse uses a
+'log context' to track which request it is handling at any given moment. This
+is done via a thread-local variable; a ``logging.Filter`` is then used to fish
+the information back out of the thread-local variable and add it to each log
+record.
+
+Logcontexts are also used for CPU and database accounting, so that we can track
+which requests were responsible for high CPU use or database activity.
+
+The ``synapse.util.logcontext`` module provides a facilities for managing the
+current log context (as well as providing the ``LoggingContextFilter`` class).
+
+Deferreds make the whole thing complicated, so this document describes how it
+all works, and how to write code which follows the rules.
+
+Logcontexts without Deferreds
+-----------------------------
+
+In the absence of any Deferred voodoo, things are simple enough. As with any
+code of this nature, the rule is that our function should leave things as it
+found them:
+
+.. code:: python
+
+    from synapse.util import logcontext         # omitted from future snippets
+
+    def handle_request(request_id):
+        request_context = logcontext.LoggingContext()
+
+        calling_context = logcontext.LoggingContext.current_context()
+        logcontext.LoggingContext.set_current_context(request_context)
+        try:
+            request_context.request = request_id
+            do_request_handling()
+            logger.debug("finished")
+        finally:
+            logcontext.LoggingContext.set_current_context(calling_context)
+
+    def do_request_handling():
+        logger.debug("phew")  # this will be logged against request_id
+
+
+LoggingContext implements the context management methods, so the above can be
+written much more succinctly as:
+
+.. code:: python
+
+    def handle_request(request_id):
+        with logcontext.LoggingContext() as request_context:
+            request_context.request = request_id
+            do_request_handling()
+            logger.debug("finished")
+
+    def do_request_handling():
+        logger.debug("phew")
+
+
+Using logcontexts with Deferreds
+--------------------------------
+
+Deferreds — and in particular, ``defer.inlineCallbacks`` — break
+the linear flow of code so that there is no longer a single entry point where
+we should set the logcontext and a single exit point where we should remove it.
+
+Consider the example above, where ``do_request_handling`` needs to do some
+blocking operation, and returns a deferred:
+
+.. code:: python
+
+    @defer.inlineCallbacks
+    def handle_request(request_id):
+        with logcontext.LoggingContext() as request_context:
+            request_context.request = request_id
+            yield do_request_handling()
+            logger.debug("finished")
+
+
+In the above flow:
+
+* The logcontext is set
+* ``do_request_handling`` is called, and returns a deferred
+* ``handle_request`` yields the deferred
+* The ``inlineCallbacks`` wrapper of ``handle_request`` returns a deferred
+
+So we have stopped processing the request (and will probably go on to start
+processing the next), without clearing the logcontext.
+
+To circumvent this problem, synapse code assumes that, wherever you have a
+deferred, you will want to yield on it. To that end, whereever functions return
+a deferred, we adopt the following conventions:
+
+**Rules for functions returning deferreds:**
+
+  * If the deferred is already complete, the function returns with the same
+    logcontext it started with.
+  * If the deferred is incomplete, the function clears the logcontext before
+    returning; when the deferred completes, it restores the logcontext before
+    running any callbacks.
+
+That sounds complicated, but actually it means a lot of code (including the
+example above) "just works". There are two cases:
+
+* If ``do_request_handling`` returns a completed deferred, then the logcontext
+  will still be in place. In this case, execution will continue immediately
+  after the ``yield``; the "finished" line will be logged against the right
+  context, and the ``with`` block restores the original context before we
+  return to the caller.
+
+* If the returned deferred is incomplete, ``do_request_handling`` clears the
+  logcontext before returning. The logcontext is therefore clear when
+  ``handle_request`` yields the deferred. At that point, the ``inlineCallbacks``
+  wrapper adds a callback to the deferred, and returns another (incomplete)
+  deferred to the caller, and it is safe to begin processing the next request.
+
+  Once ``do_request_handling``'s deferred completes, it will reinstate the
+  logcontext, before running the callback added by the ``inlineCallbacks``
+  wrapper. That callback runs the second half of ``handle_request``, so again
+  the "finished" line will be logged against the right
+  context, and the ``with`` block restores the original context.
+
+As an aside, it's worth noting that ``handle_request`` follows our rules -
+though that only matters if the caller has its own logcontext which it cares
+about.
+
+The following sections describe pitfalls and helpful patterns when implementing
+these rules.
+
+Always yield your deferreds
+---------------------------
+
+Whenever you get a deferred back from a function, you should ``yield`` on it
+as soon as possible. (Returning it directly to your caller is ok too, if you're
+not doing ``inlineCallbacks``.) Do not pass go; do not do any logging; do not
+call any other functions.
+
+.. code:: python
+
+    @defer.inlineCallbacks
+    def fun():
+        logger.debug("starting")
+        yield do_some_stuff()       # just like this
+
+        d = more_stuff()
+        result = yield d            # also fine, of course
+
+        defer.returnValue(result)
+
+    def nonInlineCallbacksFun():
+        logger.debug("just a wrapper really")
+        return do_some_stuff()      # this is ok too - the caller will yield on
+                                    # it anyway.
+
+Provided this pattern is followed all the way back up to the callchain to where
+the logcontext was set, this will make things work out ok: provided
+``do_some_stuff`` and ``more_stuff`` follow the rules above, then so will
+``fun`` (as wrapped by ``inlineCallbacks``) and ``nonInlineCallbacksFun``.
+
+It's all too easy to forget to ``yield``: for instance if we forgot that
+``do_some_stuff`` returned a deferred, we might plough on regardless. This
+leads to a mess; it will probably work itself out eventually, but not before
+a load of stuff has been logged against the wrong content. (Normally, other
+things will break, more obviously, if you forget to ``yield``, so this tends
+not to be a major problem in practice.)
+
+Of course sometimes you need to do something a bit fancier with your Deferreds
+- not all code follows the linear A-then-B-then-C pattern. Notes on
+implementing more complex patterns are in later sections.
+
+Where you create a new Deferred, make it follow the rules
+---------------------------------------------------------
+
+Most of the time, a Deferred comes from another synapse function. Sometimes,
+though, we need to make up a new Deferred, or we get a Deferred back from
+external code. We need to make it follow our rules.
+
+The easy way to do it is with a combination of ``defer.inlineCallbacks``, and
+``logcontext.PreserveLoggingContext``. Suppose we want to implement ``sleep``,
+which returns a deferred which will run its callbacks after a given number of
+seconds. That might look like:
+
+.. code:: python
+
+    # not a logcontext-rules-compliant function
+    def get_sleep_deferred(seconds):
+        d = defer.Deferred()
+        reactor.callLater(seconds, d.callback, None)
+        return d
+
+That doesn't follow the rules, but we can fix it by wrapping it with
+``PreserveLoggingContext`` and ``yield`` ing on it:
+
+.. code:: python
+
+    @defer.inlineCallbacks
+    def sleep(seconds):
+        with PreserveLoggingContext():
+            yield get_sleep_deferred(seconds)
+
+This technique works equally for external functions which return deferreds,
+or deferreds we have made ourselves.
+
+You can also use ``logcontext.make_deferred_yieldable``, which just does the
+boilerplate for you, so the above could be written:
+
+.. code:: python
+
+    def sleep(seconds):
+        return logcontext.make_deferred_yieldable(get_sleep_deferred(seconds))
+
+
+Fire-and-forget
+---------------
+
+Sometimes you want to fire off a chain of execution, but not wait for its
+result. That might look a bit like this:
+
+.. code:: python
+
+    @defer.inlineCallbacks
+    def do_request_handling():
+        yield foreground_operation()
+
+        # *don't* do this
+        background_operation()
+
+        logger.debug("Request handling complete")
+
+    @defer.inlineCallbacks
+    def background_operation():
+        yield first_background_step()
+        logger.debug("Completed first step")
+        yield second_background_step()
+        logger.debug("Completed second step")
+
+The above code does a couple of steps in the background after
+``do_request_handling`` has finished. The log lines are still logged against
+the ``request_context`` logcontext, which may or may not be desirable. There
+are two big problems with the above, however. The first problem is that, if
+``background_operation`` returns an incomplete Deferred, it will expect its
+caller to ``yield`` immediately, so will have cleared the logcontext. In this
+example, that means that 'Request handling complete' will be logged without any
+context.
+
+The second problem, which is potentially even worse, is that when the Deferred
+returned by ``background_operation`` completes, it will restore the original
+logcontext. There is nothing waiting on that Deferred, so the logcontext will
+leak into the reactor and possibly get attached to some arbitrary future
+operation.
+
+There are two potential solutions to this.
+
+One option is to surround the call to ``background_operation`` with a
+``PreserveLoggingContext`` call. That will reset the logcontext before
+starting ``background_operation`` (so the context restored when the deferred
+completes will be the empty logcontext), and will restore the current
+logcontext before continuing the foreground process:
+
+.. code:: python
+
+    @defer.inlineCallbacks
+    def do_request_handling():
+        yield foreground_operation()
+
+        # start background_operation off in the empty logcontext, to
+        # avoid leaking the current context into the reactor.
+        with PreserveLoggingContext():
+            background_operation()
+
+        # this will now be logged against the request context
+        logger.debug("Request handling complete")
+
+Obviously that option means that the operations done in
+``background_operation`` would be not be logged against a logcontext (though
+that might be fixed by setting a different logcontext via a ``with
+LoggingContext(...)`` in ``background_operation``).
+
+The second option is to use ``logcontext.preserve_fn``, which wraps a function
+so that it doesn't reset the logcontext even when it returns an incomplete
+deferred, and adds a callback to the returned deferred to reset the
+logcontext. In other words, it turns a function that follows the Synapse rules
+about logcontexts and Deferreds into one which behaves more like an external
+function — the opposite operation to that described in the previous section.
+It can be used like this:
+
+.. code:: python
+
+    @defer.inlineCallbacks
+    def do_request_handling():
+        yield foreground_operation()
+
+        logcontext.preserve_fn(background_operation)()
+
+        # this will now be logged against the request context
+        logger.debug("Request handling complete")
+
+XXX: I think ``preserve_context_over_fn`` is supposed to do the first option,
+but the fact that it does ``preserve_context_over_deferred`` on its results
+means that its use is fraught with difficulty.
+
+Passing synapse deferreds into third-party functions
+----------------------------------------------------
+
+A typical example of this is where we want to collect together two or more
+deferred via ``defer.gatherResults``:
+
+.. code:: python
+
+    d1 = operation1()
+    d2 = operation2()
+    d3 = defer.gatherResults([d1, d2])
+
+This is really a variation of the fire-and-forget problem above, in that we are
+firing off ``d1`` and ``d2`` without yielding on them. The difference
+is that we now have third-party code attached to their callbacks. Anyway either
+technique given in the `Fire-and-forget`_ section will work.
+
+Of course, the new Deferred returned by ``gatherResults`` needs to be wrapped
+in order to make it follow the logcontext rules before we can yield it, as
+described in `Where you create a new Deferred, make it follow the rules`_.
+
+So, option one: reset the logcontext before starting the operations to be
+gathered:
+
+.. code:: python
+
+    @defer.inlineCallbacks
+    def do_request_handling():
+        with PreserveLoggingContext():
+            d1 = operation1()
+            d2 = operation2()
+            result = yield defer.gatherResults([d1, d2])
+
+In this case particularly, though, option two, of using
+``logcontext.preserve_fn`` almost certainly makes more sense, so that
+``operation1`` and ``operation2`` are both logged against the original
+logcontext. This looks like:
+
+.. code:: python
+
+    @defer.inlineCallbacks
+    def do_request_handling():
+        d1 = logcontext.preserve_fn(operation1)()
+        d2 = logcontext.preserve_fn(operation2)()
+
+        with PreserveLoggingContext():
+            result = yield defer.gatherResults([d1, d2])
+
+
+Was all this really necessary?
+------------------------------
+
+The conventions used work fine for a linear flow where everything happens in
+series via ``defer.inlineCallbacks`` and ``yield``, but are certainly tricky to
+follow for any more exotic flows. It's hard not to wonder if we could have done
+something else.
+
+We're not going to rewrite Synapse now, so the following is entirely of
+academic interest, but I'd like to record some thoughts on an alternative
+approach.
+
+I briefly prototyped some code following an alternative set of rules. I think
+it would work, but I certainly didn't get as far as thinking how it would
+interact with concepts as complicated as the cache descriptors.
+
+My alternative rules were:
+
+* functions always preserve the logcontext of their caller, whether or not they
+  are returning a Deferred.
+
+* Deferreds returned by synapse functions run their callbacks in the same
+  context as the function was orignally called in.
+
+The main point of this scheme is that everywhere that sets the logcontext is
+responsible for clearing it before returning control to the reactor.
+
+So, for example, if you were the function which started a ``with
+LoggingContext`` block, you wouldn't ``yield`` within it — instead you'd start
+off the background process, and then leave the ``with`` block to wait for it:
+
+.. code:: python
+
+    def handle_request(request_id):
+        with logcontext.LoggingContext() as request_context:
+            request_context.request = request_id
+            d = do_request_handling()
+
+        def cb(r):
+            logger.debug("finished")
+
+        d.addCallback(cb)
+        return d
+
+(in general, mixing ``with LoggingContext`` blocks and
+``defer.inlineCallbacks`` in the same function leads to slighly
+counter-intuitive code, under this scheme).
+
+Because we leave the original ``with`` block as soon as the Deferred is
+returned (as opposed to waiting for it to be resolved, as we do today), the
+logcontext is cleared before control passes back to the reactor; so if there is
+some code within ``do_request_handling`` which needs to wait for a Deferred to
+complete, there is no need for it to worry about clearing the logcontext before
+doing so:
+
+.. code:: python
+
+    def handle_request():
+        r = do_some_stuff()
+        r.addCallback(do_some_more_stuff)
+        return r
+
+— and provided ``do_some_stuff`` follows the rules of returning a Deferred which
+runs its callbacks in the original logcontext, all is happy.
+
+The business of a Deferred which runs its callbacks in the original logcontext
+isn't hard to achieve — we have it today, in the shape of
+``logcontext._PreservingContextDeferred``:
+
+.. code:: python
+
+    def do_some_stuff():
+        deferred = do_some_io()
+        pcd = _PreservingContextDeferred(LoggingContext.current_context())
+        deferred.chainDeferred(pcd)
+        return pcd
+
+It turns out that, thanks to the way that Deferreds chain together, we
+automatically get the property of a context-preserving deferred with
+``defer.inlineCallbacks``, provided the final Defered the function ``yields``
+on has that property. So we can just write:
+
+.. code:: python
+
+    @defer.inlineCallbacks
+    def handle_request():
+        yield do_some_stuff()
+        yield do_some_more_stuff()
+
+To conclude: I think this scheme would have worked equally well, with less
+danger of messing it up, and probably made some more esoteric code easier to
+write. But again — changing the conventions of the entire Synapse codebase is
+not a sensible option for the marginal improvement offered.

docs/metrics-howto.rst

@@ -21,13 +21,12 @@ How to monitor Synapse metrics using Prometheus
 
 3. Add a prometheus target for synapse.
 
-   It needs to set the ``metrics_path`` to a non-default value::
+   It needs to set the ``metrics_path`` to a non-default value (under ``scrape_configs``)::
 
     - job_name: "synapse"
       metrics_path: "/_synapse/metrics"
       static_configs:
-        - targets:
-            "my.server.here:9092"
+        - targets: ["my.server.here:9092"]
 
    If your prometheus is older than 1.5.2, you will need to replace 
    ``static_configs`` in the above with ``target_groups``.

docs/postgres.rst

@@ -112,9 +112,9 @@ script one last time, e.g. if the SQLite database is at  ``homeserver.db``
 run::
 
     synapse_port_db --sqlite-database homeserver.db \
-        --postgres-config database_config.yaml
+        --postgres-config homeserver-postgres.yaml
 
 Once that has completed, change the synapse config to point at the PostgreSQL
-database configuration file using the ``database_config`` parameter (see
-`Synapse Config`_) and restart synapse. Synapse should now be running against
+database configuration file ``homeserver-postgres.yaml`` (i.e. rename it to 
+``homeserver.yaml``) and restart synapse. Synapse should now be running against
 PostgreSQL.

docs/replication.rst

@@ -26,28 +26,10 @@ expose the append-only log to the readers should be fairly minimal.
 Architecture
 ------------
 
-The Replication API
-~~~~~~~~~~~~~~~~~~~
+The Replication Protocol
+~~~~~~~~~~~~~~~~~~~~~~~~
 
-Synapse will optionally expose a long poll HTTP API for extracting updates. The
-API will have a similar shape to /sync in that clients provide tokens
-indicating where in the log they have reached and a timeout. The synapse server
-then either responds with updates immediately if it already has updates or it
-waits until the timeout for more updates. If the timeout expires and nothing
-happened then the server returns an empty response.
-
-However unlike the /sync API this replication API is returning synapse specific
-data rather than trying to implement a matrix specification. The replication
-results are returned as arrays of rows where the rows are mostly lifted
-directly from the database. This avoids unnecessary JSON parsing on the server
-and hopefully avoids an impedance mismatch between the data returned and the
-required updates to the datastore.
-
-This does not replicate all the database tables as many of the database tables
-are indexes that can be recovered from the contents of other tables.
-
-The format and parameters for the api are documented in
-``synapse/replication/resource.py``.
+See ``tcp_replication.rst``
 
 
 The Slaved DataStore

docs/tcp_replication.rst

@@ -0,0 +1,223 @@
+TCP Replication
+===============
+
+Motivation
+----------
+
+Previously the workers used an HTTP long poll mechanism to get updates from the
+master, which had the problem of causing a lot of duplicate work on the server.
+This TCP protocol replaces those APIs with the aim of increased efficiency.
+
+
+
+Overview
+--------
+
+The protocol is based on fire and forget, line based commands. An example flow
+would be (where '>' indicates master to worker and '<' worker to master flows)::
+
+    > SERVER example.com
+    < REPLICATE events 53
+    > RDATA events 54 ["$foo1:bar.com", ...]
+    > RDATA events 55 ["$foo4:bar.com", ...]
+
+The example shows the server accepting a new connection and sending its identity
+with the ``SERVER`` command, followed by the client asking to subscribe to the
+``events`` stream from the token ``53``. The server then periodically sends ``RDATA``
+commands which have the format ``RDATA <stream_name> <token> <row>``, where the
+format of ``<row>`` is defined by the individual streams.
+
+Error reporting happens by either the client or server sending an `ERROR`
+command, and usually the connection will be closed.
+
+
+Since the protocol is a simple line based, its possible to manually connect to
+the server using a tool like netcat. A few things should be noted when manually
+using the protocol:
+
+* When subscribing to a stream using ``REPLICATE``, the special token ``NOW`` can
+  be used to get all future updates. The special stream name ``ALL`` can be used
+  with ``NOW`` to subscribe to all available streams.
+* The federation stream is only available if federation sending has been
+  disabled on the main process.
+* The server will only time connections out that have sent a ``PING`` command.
+  If a ping is sent then the connection will be closed if no further commands
+  are receieved within 15s. Both the client and server protocol implementations
+  will send an initial PING on connection and ensure at least one command every
+  5s is sent (not necessarily ``PING``).
+* ``RDATA`` commands *usually* include a numeric token, however if the stream
+  has multiple rows to replicate per token the server will send multiple
+  ``RDATA`` commands, with all but the last having a token of ``batch``. See
+  the documentation on ``commands.RdataCommand`` for further details.
+
+
+Architecture
+------------
+
+The basic structure of the protocol is line based, where the initial word of
+each line specifies the command. The rest of the line is parsed based on the
+command. For example, the `RDATA` command is defined as::
+
+    RDATA <stream_name> <token> <row_json>
+
+(Note that `<row_json>` may contains spaces, but cannot contain newlines.)
+
+Blank lines are ignored.
+
+
+Keep alives
+~~~~~~~~~~~
+
+Both sides are expected to send at least one command every 5s or so, and
+should send a ``PING`` command if necessary. If either side do not receive a
+command within e.g. 15s then the connection should be closed.
+
+Because the server may be connected to manually using e.g. netcat, the timeouts
+aren't enabled until an initial ``PING`` command is seen. Both the client and
+server implementations below send a ``PING`` command immediately on connection to
+ensure the timeouts are enabled.
+
+This ensures that both sides can quickly realize if the tcp connection has gone
+and handle the situation appropriately.
+
+
+Start up
+~~~~~~~~
+
+When a new connection is made, the server:
+
+* Sends a ``SERVER`` command, which includes the identity of the server, allowing
+  the client to detect if its connected to the expected server
+* Sends a ``PING`` command as above, to enable the client to time out connections
+  promptly.
+
+The client:
+
+* Sends a ``NAME`` command, allowing the server to associate a human friendly
+  name with the connection. This is optional.
+* Sends a ``PING`` as above
+* For each stream the client wishes to subscribe to it sends a ``REPLICATE``
+  with the stream_name and token it wants to subscribe from.
+* On receipt of a ``SERVER`` command, checks that the server name matches the
+  expected server name.
+
+
+Error handling
+~~~~~~~~~~~~~~
+
+If either side detects an error it can send an ``ERROR`` command and close the
+connection.
+
+If the client side loses the connection to the server it should reconnect,
+following the steps above.
+
+
+Congestion
+~~~~~~~~~~
+
+If the server sends messages faster than the client can consume them the server
+will first buffer a (fairly large) number of commands and then disconnect the
+client. This ensures that we don't queue up an unbounded number of commands in
+memory and gives us a potential oppurtunity to squawk loudly. When/if the client
+recovers it can reconnect to the server and ask for missed messages.
+
+
+Reliability
+~~~~~~~~~~~
+
+In general the replication stream should be considered an unreliable transport
+since e.g. commands are not resent if the connection disappears.
+
+The exception to that are the replication streams, i.e. RDATA commands, since
+these include tokens which can be used to restart the stream on connection
+errors.
+
+The client should keep track of the token in the last RDATA command received
+for each stream so that on reconneciton it can start streaming from the correct
+place. Note: not all RDATA have valid tokens due to batching. See
+``RdataCommand`` for more details.
+
+
+Example
+~~~~~~~
+
+An example iteraction is shown below. Each line is prefixed with '>' or '<' to
+indicate which side is sending, these are *not* included on the wire::
+
+    * connection established *
+    > SERVER localhost:8823
+    > PING 1490197665618
+    < NAME synapse.app.appservice
+    < PING 1490197665618
+    < REPLICATE events 1
+    < REPLICATE backfill 1
+    < REPLICATE caches 1
+    > POSITION events 1
+    > POSITION backfill 1
+    > POSITION caches 1
+    > RDATA caches 2 ["get_user_by_id",["@01register-user:localhost:8823"],1490197670513]
+    > RDATA events 14 ["$149019767112vOHxz:localhost:8823",
+        "!AFDCvgApUmpdfVjIXm:localhost:8823","m.room.guest_access","",null]
+    < PING 1490197675618
+    > ERROR server stopping
+    * connection closed by server *
+
+The ``POSITION`` command sent by the server is used to set the clients position
+without needing to send data with the ``RDATA`` command.
+
+
+An example of a batched set of ``RDATA`` is::
+
+    > RDATA caches batch ["get_user_by_id",["@test:localhost:8823"],1490197670513]
+    > RDATA caches batch ["get_user_by_id",["@test2:localhost:8823"],1490197670513]
+    > RDATA caches batch ["get_user_by_id",["@test3:localhost:8823"],1490197670513]
+    > RDATA caches 54 ["get_user_by_id",["@test4:localhost:8823"],1490197670513]
+
+In this case the client shouldn't advance their caches token until it sees the
+the last ``RDATA``.
+
+
+List of commands
+~~~~~~~~~~~~~~~~
+
+The list of valid commands, with which side can send it: server (S) or client (C):
+
+SERVER (S)
+    Sent at the start to identify which server the client is talking to
+
+RDATA (S)
+    A single update in a stream
+
+POSITION (S)
+    The position of the stream has been updated
+
+ERROR (S, C)
+    There was an error
+
+PING (S, C)
+    Sent periodically to ensure the connection is still alive
+
+NAME (C)
+    Sent at the start by client to inform the server who they are
+
+REPLICATE (C)
+    Asks the server to replicate a given stream
+
+USER_SYNC (C)
+    A user has started or stopped syncing
+
+FEDERATION_ACK (C)
+    Acknowledge receipt of some federation data
+
+REMOVE_PUSHER (C)
+    Inform the server a pusher should be removed
+
+INVALIDATE_CACHE (C)
+    Inform the server a cache should be invalidated
+
+SYNC (S, C)
+    Used exclusively in tests
+
+
+See ``synapse/replication/tcp/commands.py`` for a detailed description and the
+format of each command.

docs/turn-howto.rst

@@ -50,14 +50,37 @@ You may be able to setup coturn via your package manager,  or set it up manually
 
        pwgen -s 64 1
 
- 5. Ensure youe firewall allows traffic into the TURN server on
-    the ports you've configured it to listen on (remember to allow
-    both TCP and UDP if you've enabled both).
+ 5. Consider your security settings.  TURN lets users request a relay
+    which will connect to arbitrary IP addresses and ports.  At the least
+    we recommend:
 
- 6. If you've configured coturn to support TLS/DTLS, generate or
+       # VoIP traffic is all UDP. There is no reason to let users connect to arbitrary TCP endpoints via the relay.
+       no-tcp-relay
+
+       # don't let the relay ever try to connect to private IP address ranges within your network (if any)
+       # given the turn server is likely behind your firewall, remember to include any privileged public IPs too.
+       denied-peer-ip=10.0.0.0-10.255.255.255
+       denied-peer-ip=192.168.0.0-192.168.255.255
+       denied-peer-ip=172.16.0.0-172.31.255.255
+
+       # special case the turn server itself so that client->TURN->TURN->client flows work
+       allowed-peer-ip=10.0.0.1
+
+       # consider whether you want to limit the quota of relayed streams per user (or total) to avoid risk of DoS.
+       user-quota=12 # 4 streams per video call, so 12 streams = 3 simultaneous relayed calls per user.
+       total-quota=1200
+
+    Ideally coturn should refuse to relay traffic which isn't SRTP;
+    see https://github.com/matrix-org/synapse/issues/2009
+
+ 6. Ensure your firewall allows traffic into the TURN server on
+    the ports you've configured it to listen on (remember to allow
+    both TCP and UDP TURN traffic)
+
+ 7. If you've configured coturn to support TLS/DTLS, generate or
     import your private key and certificate.
 
- 7. Start the turn server::
+ 8. Start the turn server::
  
        bin/turnserver -o
 
@@ -83,12 +106,19 @@ Your home server configuration file needs the following extra keys:
     to refresh credentials. The TURN REST API specification recommends
     one day (86400000).
 
+  4. "turn_allow_guests": Whether to allow guest users to use the TURN
+    server.  This is enabled by default, as otherwise VoIP will not
+    work reliably for guests.  However, it does introduce a security risk
+    as it lets guests connect to arbitrary endpoints without having gone
+    through a CAPTCHA or similar to register a real account.
+
 As an example, here is the relevant section of the config file for
 matrix.org::
 
     turn_uris: [ "turn:turn.matrix.org:3478?transport=udp", "turn:turn.matrix.org:3478?transport=tcp" ]
     turn_shared_secret: n0t4ctuAllymatr1Xd0TorgSshar3d5ecret4obvIousreAsons
     turn_user_lifetime: 86400000
+    turn_allow_guests: True
 
 Now, restart synapse::
 

docs/workers.rst

@@ -12,7 +12,7 @@ across multiple processes is a recipe for disaster, plus you should be using
 postgres anyway if you care about scalability).
 
 The workers communicate with the master synapse process via a synapse-specific
-HTTP protocol called 'replication' - analogous to MySQL or Postgres style
+TCP protocol called 'replication' - analogous to MySQL or Postgres style
 database replication; feeding a stream of relevant data to the workers so they
 can be kept in sync with the main synapse process and database state.
 
@@ -21,16 +21,11 @@ To enable workers, you need to add a replication listener to the master synapse,
     listeners:
       - port: 9092
         bind_address: '127.0.0.1'
-        type: http
-        tls: false
-        x_forwarded: false
-        resources:
-          - names: [replication]
-            compress: false
+        type: replication
 
 Under **no circumstances** should this replication API listener be exposed to the
 public internet; it currently implements no authentication whatsoever and is
-unencrypted HTTP.
+unencrypted.
 
 You then create a set of configs for the various worker processes.  These should be
 worker configuration files should be stored in a dedicated subdirectory, to allow
@@ -50,14 +45,16 @@ e.g. the HTTP listener that it provides (if any); logging configuration; etc.
 You should minimise the number of overrides though to maintain a usable config.
 
 You must specify the type of worker application (worker_app) and the replication
-endpoint that it's talking to on the main synapse process (worker_replication_url).
+endpoint that it's talking to on the main synapse process (worker_replication_host
+and worker_replication_port).
 
 For instance::
 
     worker_app: synapse.app.synchrotron
 
     # The replication listener on the synapse to talk to.
-    worker_replication_url: http://127.0.0.1:9092/_synapse/replication
+    worker_replication_host: 127.0.0.1
+    worker_replication_port: 9092
 
     worker_listeners:
      - type: http
@@ -95,4 +92,3 @@ To manipulate a specific worker, you pass the -w option to synctl::
 All of the above is highly experimental and subject to change as Synapse evolves,
 but documenting it here to help folks needing highly scalable Synapses similar
 to the one running matrix.org!
-

jenkins-dendron-haproxy-postgres.sh

@@ -0,0 +1,22 @@
+#!/bin/bash
+
+set -eux
+
+: ${WORKSPACE:="$(pwd)"}
+
+export WORKSPACE
+export PYTHONDONTWRITEBYTECODE=yep
+export SYNAPSE_CACHE_FACTOR=1
+
+export HAPROXY_BIN=/home/haproxy/haproxy-1.6.11/haproxy
+
+./jenkins/prepare_synapse.sh
+./jenkins/clone.sh sytest https://github.com/matrix-org/sytest.git
+./jenkins/clone.sh dendron https://github.com/matrix-org/dendron.git
+./dendron/jenkins/build_dendron.sh
+./sytest/jenkins/prep_sytest_for_postgres.sh
+
+./sytest/jenkins/install_and_run.sh \
+    --synapse-directory $WORKSPACE \
+    --dendron $WORKSPACE/dendron/bin/dendron \
+    --haproxy \

jenkins-dendron-postgres.sh

@@ -17,9 +17,3 @@ export SYNAPSE_CACHE_FACTOR=1
 ./sytest/jenkins/install_and_run.sh \
     --synapse-directory $WORKSPACE \
     --dendron $WORKSPACE/dendron/bin/dendron \
-    --pusher \
-    --synchrotron \
-    --federation-reader \
-    --client-reader \
-    --appservice \
-    --federation-sender \

scripts-dev/nuke-room-from-db.sh

@@ -9,16 +9,39 @@
 ROOMID="$1"
 
 sqlite3 homeserver.db <<EOF
-DELETE FROM context_depth WHERE context = '$ROOMID';
-DELETE FROM current_state WHERE context = '$ROOMID';
-DELETE FROM feedback WHERE room_id = '$ROOMID';
-DELETE FROM messages WHERE room_id = '$ROOMID';
-DELETE FROM pdu_backward_extremities WHERE context = '$ROOMID';
-DELETE FROM pdu_edges WHERE context = '$ROOMID';
-DELETE FROM pdu_forward_extremities WHERE context = '$ROOMID';
-DELETE FROM pdus WHERE context = '$ROOMID';
-DELETE FROM room_data WHERE room_id = '$ROOMID';
+DELETE FROM event_forward_extremities WHERE room_id = '$ROOMID';
+DELETE FROM event_backward_extremities WHERE room_id = '$ROOMID';
+DELETE FROM event_edges WHERE room_id = '$ROOMID';
+DELETE FROM room_depth WHERE room_id = '$ROOMID';
+DELETE FROM state_forward_extremities WHERE room_id = '$ROOMID';
+DELETE FROM events WHERE room_id = '$ROOMID';
+DELETE FROM event_json WHERE room_id = '$ROOMID';
+DELETE FROM state_events WHERE room_id = '$ROOMID';
+DELETE FROM current_state_events WHERE room_id = '$ROOMID';
 DELETE FROM room_memberships WHERE room_id = '$ROOMID';
+DELETE FROM feedback WHERE room_id = '$ROOMID';
+DELETE FROM topics WHERE room_id = '$ROOMID';
+DELETE FROM room_names WHERE room_id = '$ROOMID';
 DELETE FROM rooms WHERE room_id = '$ROOMID';
-DELETE FROM state_pdus WHERE context = '$ROOMID';
+DELETE FROM room_hosts WHERE room_id = '$ROOMID';
+DELETE FROM room_aliases WHERE room_id = '$ROOMID';
+DELETE FROM state_groups WHERE room_id = '$ROOMID';
+DELETE FROM state_groups_state WHERE room_id = '$ROOMID';
+DELETE FROM receipts_graph WHERE room_id = '$ROOMID';
+DELETE FROM receipts_linearized WHERE room_id = '$ROOMID';
+DELETE FROM event_search_content WHERE c1room_id = '$ROOMID';
+DELETE FROM guest_access WHERE room_id = '$ROOMID';
+DELETE FROM history_visibility WHERE room_id = '$ROOMID';
+DELETE FROM room_tags WHERE room_id = '$ROOMID';
+DELETE FROM room_tags_revisions WHERE room_id = '$ROOMID';
+DELETE FROM room_account_data WHERE room_id = '$ROOMID';
+DELETE FROM event_push_actions WHERE room_id = '$ROOMID';
+DELETE FROM local_invites WHERE room_id = '$ROOMID';
+DELETE FROM pusher_throttle WHERE room_id = '$ROOMID';
+DELETE FROM event_reports WHERE room_id = '$ROOMID';
+DELETE FROM public_room_list_stream WHERE room_id = '$ROOMID';
+DELETE FROM stream_ordering_to_exterm WHERE room_id = '$ROOMID';
+DELETE FROM event_auth WHERE room_id = '$ROOMID';
+DELETE FROM appservice_room_list WHERE room_id = '$ROOMID';
+VACUUM;
 EOF

scripts/synapse_port_db

@@ -40,6 +40,7 @@ BOOLEAN_COLUMNS = {
     "presence_list": ["accepted"],
     "presence_stream": ["currently_active"],
     "public_room_list_stream": ["visibility"],
+    "device_lists_outbound_pokes": ["sent"],
 }
 
 
@@ -446,9 +447,7 @@ class Porter(object):
 
             postgres_tables = yield self.postgres_store._simple_select_onecol(
                 table="information_schema.tables",
-                keyvalues={
-                    "table_schema": "public",
-                },
+                keyvalues={},
                 retcol="distinct table_name",
             )
 

synapse/__init__.py

@@ -16,4 +16,4 @@
 """ This is a reference implementation of a Matrix home server.
 """
 
-__version__ = "0.19.2"
+__version__ = "0.21.0"

synapse/api/auth.py

@@ -23,7 +23,7 @@ from synapse import event_auth
 from synapse.api.constants import EventTypes, Membership, JoinRules
 from synapse.api.errors import AuthError, Codes
 from synapse.types import UserID
-from synapse.util.logcontext import preserve_context_over_fn
+from synapse.util import logcontext
 from synapse.util.metrics import Measure
 
 logger = logging.getLogger(__name__)
@@ -209,8 +209,7 @@ class Auth(object):
                 default=[""]
             )[0]
             if user and access_token and ip_addr:
-                preserve_context_over_fn(
-                    self.store.insert_client_ip,
+                logcontext.preserve_fn(self.store.insert_client_ip)(
                     user=user,
                     access_token=access_token,
                     ip=ip_addr,

synapse/api/constants.py

@@ -1,5 +1,6 @@
 # -*- coding: utf-8 -*-
 # Copyright 2014-2016 OpenMarket Ltd
+# Copyright 2017 Vector Creations Ltd
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -43,10 +44,8 @@ class JoinRules(object):
 
 class LoginType(object):
     PASSWORD = u"m.login.password"
-    OAUTH = u"m.login.oauth2"
-    EMAIL_CODE = u"m.login.email.code"
-    EMAIL_URL = u"m.login.email.url"
     EMAIL_IDENTITY = u"m.login.email.identity"
+    MSISDN = u"m.login.msisdn"
     RECAPTCHA = u"m.login.recaptcha"
     DUMMY = u"m.login.dummy"
 

synapse/api/errors.py

@@ -15,6 +15,7 @@
 
 """Contains exceptions and error codes."""
 
+import json
 import logging
 
 logger = logging.getLogger(__name__)
@@ -50,27 +51,46 @@ class Codes(object):
 
 
 class CodeMessageException(RuntimeError):
-    """An exception with integer code and message string attributes."""
+    """An exception with integer code and message string attributes.
 
+    Attributes:
+        code (int): HTTP error code
+        msg (str): string describing the error
+    """
     def __init__(self, code, msg):
         super(CodeMessageException, self).__init__("%d: %s" % (code, msg))
         self.code = code
         self.msg = msg
-        self.response_code_message = None
 
     def error_dict(self):
         return cs_error(self.msg)
 
 
+class MatrixCodeMessageException(CodeMessageException):
+    """An error from a general matrix endpoint, eg. from a proxied Matrix API call.
+
+    Attributes:
+        errcode (str): Matrix error code e.g 'M_FORBIDDEN'
+    """
+    def __init__(self, code, msg, errcode=Codes.UNKNOWN):
+        super(MatrixCodeMessageException, self).__init__(code, msg)
+        self.errcode = errcode
+
+
 class SynapseError(CodeMessageException):
-    """A base error which can be caught for all synapse events."""
+    """A base exception type for matrix errors which have an errcode and error
+    message (as well as an HTTP status code).
+
+    Attributes:
+        errcode (str): Matrix error code e.g 'M_FORBIDDEN'
+    """
     def __init__(self, code, msg, errcode=Codes.UNKNOWN):
         """Constructs a synapse error.
 
         Args:
             code (int): The integer error code (an HTTP response code)
             msg (str): The human-readable error message.
-            err (str): The error code e.g 'M_FORBIDDEN'
+            errcode (str): The matrix error code e.g 'M_FORBIDDEN'
         """
         super(SynapseError, self).__init__(code, msg)
         self.errcode = errcode
@@ -81,6 +101,39 @@ class SynapseError(CodeMessageException):
             self.errcode,
         )
 
+    @classmethod
+    def from_http_response_exception(cls, err):
+        """Make a SynapseError based on an HTTPResponseException
+
+        This is useful when a proxied request has failed, and we need to
+        decide how to map the failure onto a matrix error to send back to the
+        client.
+
+        An attempt is made to parse the body of the http response as a matrix
+        error. If that succeeds, the errcode and error message from the body
+        are used as the errcode and error message in the new synapse error.
+
+        Otherwise, the errcode is set to M_UNKNOWN, and the error message is
+        set to the reason code from the HTTP response.
+
+        Args:
+            err (HttpResponseException):
+
+        Returns:
+            SynapseError:
+        """
+        # try to parse the body as json, to get better errcode/msg, but
+        # default to M_UNKNOWN with the HTTP status as the error text
+        try:
+            j = json.loads(err.response)
+        except ValueError:
+            j = {}
+        errcode = j.get('errcode', Codes.UNKNOWN)
+        errmsg = j.get('error', err.msg)
+
+        res = SynapseError(err.code, errmsg, errcode)
+        return res
+
 
 class RegistrationError(SynapseError):
     """An error raised when a registration event fails."""
@@ -106,13 +159,11 @@ class UnrecognizedRequestError(SynapseError):
 
 class NotFoundError(SynapseError):
     """An error indicating we can't find the thing you asked for"""
-    def __init__(self, *args, **kwargs):
-        if "errcode" not in kwargs:
-            kwargs["errcode"] = Codes.NOT_FOUND
+    def __init__(self, msg="Not found", errcode=Codes.NOT_FOUND):
         super(NotFoundError, self).__init__(
             404,
-            "Not found",
-            **kwargs
+            msg,
+            errcode=errcode
         )
 
 
@@ -173,7 +224,6 @@ class LimitExceededError(SynapseError):
                  errcode=Codes.LIMIT_EXCEEDED):
         super(LimitExceededError, self).__init__(code, msg, errcode)
         self.retry_after_ms = retry_after_ms
-        self.response_code_message = "Too Many Requests"
 
     def error_dict(self):
         return cs_error(
@@ -243,6 +293,19 @@ class FederationError(RuntimeError):
 
 
 class HttpResponseException(CodeMessageException):
+    """
+    Represents an HTTP-level failure of an outbound request
+
+    Attributes:
+        response (str): body of response
+    """
     def __init__(self, code, msg, response):
-        self.response = response
+        """
+
+        Args:
+            code (int): HTTP status code
+            msg (str): reason phrase from HTTP response status line
+            response (str): body of response
+        """
         super(HttpResponseException, self).__init__(code, msg)
+        self.response = response

synapse/api/filtering.py

@@ -13,11 +13,174 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 from synapse.api.errors import SynapseError
+from synapse.storage.presence import UserPresenceState
 from synapse.types import UserID, RoomID
-
 from twisted.internet import defer
 
 import ujson as json
+import jsonschema
+from jsonschema import FormatChecker
+
+FILTER_SCHEMA = {
+    "additionalProperties": False,
+    "type": "object",
+    "properties": {
+        "limit": {
+            "type": "number"
+        },
+        "senders": {
+            "$ref": "#/definitions/user_id_array"
+        },
+        "not_senders": {
+            "$ref": "#/definitions/user_id_array"
+        },
+        # TODO: We don't limit event type values but we probably should...
+        # check types are valid event types
+        "types": {
+            "type": "array",
+            "items": {
+                "type": "string"
+            }
+        },
+        "not_types": {
+            "type": "array",
+            "items": {
+                "type": "string"
+            }
+        }
+    }
+}
+
+ROOM_FILTER_SCHEMA = {
+    "additionalProperties": False,
+    "type": "object",
+    "properties": {
+        "not_rooms": {
+            "$ref": "#/definitions/room_id_array"
+        },
+        "rooms": {
+            "$ref": "#/definitions/room_id_array"
+        },
+        "ephemeral": {
+            "$ref": "#/definitions/room_event_filter"
+        },
+        "include_leave": {
+            "type": "boolean"
+        },
+        "state": {
+            "$ref": "#/definitions/room_event_filter"
+        },
+        "timeline": {
+            "$ref": "#/definitions/room_event_filter"
+        },
+        "account_data": {
+            "$ref": "#/definitions/room_event_filter"
+        },
+    }
+}
+
+ROOM_EVENT_FILTER_SCHEMA = {
+    "additionalProperties": False,
+    "type": "object",
+    "properties": {
+        "limit": {
+            "type": "number"
+        },
+        "senders": {
+            "$ref": "#/definitions/user_id_array"
+        },
+        "not_senders": {
+            "$ref": "#/definitions/user_id_array"
+        },
+        "types": {
+            "type": "array",
+            "items": {
+                "type": "string"
+            }
+        },
+        "not_types": {
+            "type": "array",
+            "items": {
+                "type": "string"
+            }
+        },
+        "rooms": {
+            "$ref": "#/definitions/room_id_array"
+        },
+        "not_rooms": {
+            "$ref": "#/definitions/room_id_array"
+        },
+        "contains_url": {
+            "type": "boolean"
+        }
+    }
+}
+
+USER_ID_ARRAY_SCHEMA = {
+    "type": "array",
+    "items": {
+        "type": "string",
+        "format": "matrix_user_id"
+    }
+}
+
+ROOM_ID_ARRAY_SCHEMA = {
+    "type": "array",
+    "items": {
+        "type": "string",
+        "format": "matrix_room_id"
+    }
+}
+
+USER_FILTER_SCHEMA = {
+    "$schema": "http://json-schema.org/draft-04/schema#",
+    "description": "schema for a Sync filter",
+    "type": "object",
+    "definitions": {
+        "room_id_array": ROOM_ID_ARRAY_SCHEMA,
+        "user_id_array": USER_ID_ARRAY_SCHEMA,
+        "filter": FILTER_SCHEMA,
+        "room_filter": ROOM_FILTER_SCHEMA,
+        "room_event_filter": ROOM_EVENT_FILTER_SCHEMA
+    },
+    "properties": {
+        "presence": {
+            "$ref": "#/definitions/filter"
+        },
+        "account_data": {
+            "$ref": "#/definitions/filter"
+        },
+        "room": {
+            "$ref": "#/definitions/room_filter"
+        },
+        "event_format": {
+            "type": "string",
+            "enum": ["client", "federation"]
+        },
+        "event_fields": {
+            "type": "array",
+            "items": {
+                "type": "string",
+                # Don't allow '\\' in event field filters. This makes matching
+                # events a lot easier as we can then use a negative lookbehind
+                # assertion to split '\.' If we allowed \\ then it would
+                # incorrectly split '\\.' See synapse.events.utils.serialize_event
+                "pattern": "^((?!\\\).)*$"
+            }
+        }
+    },
+    "additionalProperties": False
+}
+
+
+@FormatChecker.cls_checks('matrix_room_id')
+def matrix_room_id_validator(room_id_str):
+    return RoomID.from_string(room_id_str)
+
+
+@FormatChecker.cls_checks('matrix_user_id')
+def matrix_user_id_validator(user_id_str):
+    return UserID.from_string(user_id_str)
 
 
 class Filtering(object):
@@ -52,98 +215,11 @@ class Filtering(object):
         # NB: Filters are the complete json blobs. "Definitions" are an
         # individual top-level key e.g. public_user_data. Filters are made of
         # many definitions.
-
-        top_level_definitions = [
-            "presence", "account_data"
-        ]
-
-        room_level_definitions = [
-            "state", "timeline", "ephemeral", "account_data"
-        ]
-
-        for key in top_level_definitions:
-            if key in user_filter_json:
-                self._check_definition(user_filter_json[key])
-
-        if "room" in user_filter_json:
-            self._check_definition_room_lists(user_filter_json["room"])
-            for key in room_level_definitions:
-                if key in user_filter_json["room"]:
-                    self._check_definition(user_filter_json["room"][key])
-
-        if "event_fields" in user_filter_json:
-            if type(user_filter_json["event_fields"]) != list:
-                raise SynapseError(400, "event_fields must be a list of strings")
-            for field in user_filter_json["event_fields"]:
-                if not isinstance(field, basestring):
-                    raise SynapseError(400, "Event field must be a string")
-                # Don't allow '\\' in event field filters. This makes matching
-                # events a lot easier as we can then use a negative lookbehind
-                # assertion to split '\.' If we allowed \\ then it would
-                # incorrectly split '\\.' See synapse.events.utils.serialize_event
-                if r'\\' in field:
-                    raise SynapseError(
-                        400, r'The escape character \ cannot itself be escaped'
-                    )
-
-    def _check_definition_room_lists(self, definition):
-        """Check that "rooms" and "not_rooms" are lists of room ids if they
-        are present
-
-        Args:
-            definition(dict): The filter definition
-        Raises:
-            SynapseError: If there was a problem with this definition.
-        """
-        # check rooms are valid room IDs
-        room_id_keys = ["rooms", "not_rooms"]
-        for key in room_id_keys:
-            if key in definition:
-                if type(definition[key]) != list:
-                    raise SynapseError(400, "Expected %s to be a list." % key)
-                for room_id in definition[key]:
-                    RoomID.from_string(room_id)
-
-    def _check_definition(self, definition):
-        """Check if the provided definition is valid.
-
-        This inspects not only the types but also the values to make sure they
-        make sense.
-
-        Args:
-            definition(dict): The filter definition
-        Raises:
-            SynapseError: If there was a problem with this definition.
-        """
-        # NB: Filters are the complete json blobs. "Definitions" are an
-        # individual top-level key e.g. public_user_data. Filters are made of
-        # many definitions.
-        if type(definition) != dict:
-            raise SynapseError(
-                400, "Expected JSON object, not %s" % (definition,)
-            )
-
-        self._check_definition_room_lists(definition)
-
-        # check senders are valid user IDs
-        user_id_keys = ["senders", "not_senders"]
-        for key in user_id_keys:
-            if key in definition:
-                if type(definition[key]) != list:
-                    raise SynapseError(400, "Expected %s to be a list." % key)
-                for user_id in definition[key]:
-                    UserID.from_string(user_id)
-
-        # TODO: We don't limit event type values but we probably should...
-        # check types are valid event types
-        event_keys = ["types", "not_types"]
-        for key in event_keys:
-            if key in definition:
-                if type(definition[key]) != list:
-                    raise SynapseError(400, "Expected %s to be a list." % key)
-                for event_type in definition[key]:
-                    if not isinstance(event_type, basestring):
-                        raise SynapseError(400, "Event type should be a string")
+        try:
+            jsonschema.validate(user_filter_json, USER_FILTER_SCHEMA,
+                                format_checker=FormatChecker())
+        except jsonschema.ValidationError as e:
+            raise SynapseError(400, e.message)
 
 
 class FilterCollection(object):
@@ -253,19 +329,35 @@ class Filter(object):
         Returns:
             bool: True if the event matches
         """
-        sender = event.get("sender", None)
-        if not sender:
-            # Presence events have their 'sender' in content.user_id
-            content = event.get("content")
-            # account_data has been allowed to have non-dict content, so check type first
-            if isinstance(content, dict):
-                sender = content.get("user_id")
+        # We usually get the full "events" as dictionaries coming through,
+        # except for presence which actually gets passed around as its own
+        # namedtuple type.
+        if isinstance(event, UserPresenceState):
+            sender = event.user_id
+            room_id = None
+            ev_type = "m.presence"
+            is_url = False
+        else:
+            sender = event.get("sender", None)
+            if not sender:
+                # Presence events had their 'sender' in content.user_id, but are
+                # now handled above. We don't know if anything else uses this
+                # form. TODO: Check this and probably remove it.
+                content = event.get("content")
+                # account_data has been allowed to have non-dict content, so
+                # check type first
+                if isinstance(content, dict):
+                    sender = content.get("user_id")
+
+            room_id = event.get("room_id", None)
+            ev_type = event.get("type", None)
+            is_url = "url" in event.get("content", {})
 
         return self.check_fields(
-            event.get("room_id", None),
+            room_id,
             sender,
-            event.get("type", None),
-            "url" in event.get("content", {})
+            ev_type,
+            is_url,
         )
 
     def check_fields(self, room_id, sender, event_type, contains_url):

synapse/app/appservice.py

@@ -26,17 +26,17 @@ from synapse.replication.slave.storage.directory import DirectoryStore
 from synapse.replication.slave.storage.events import SlavedEventStore
 from synapse.replication.slave.storage.appservice import SlavedApplicationServiceStore
 from synapse.replication.slave.storage.registration import SlavedRegistrationStore
+from synapse.replication.tcp.client import ReplicationClientHandler
 from synapse.storage.engines import create_engine
-from synapse.util.async import sleep
 from synapse.util.httpresourcetree import create_resource_tree
-from synapse.util.logcontext import LoggingContext
+from synapse.util.logcontext import LoggingContext, PreserveLoggingContext, preserve_fn
 from synapse.util.manhole import manhole
 from synapse.util.rlimit import change_resource_limit
 from synapse.util.versionstring import get_version_string
 
 from synapse import events
 
-from twisted.internet import reactor, defer
+from twisted.internet import reactor
 from twisted.web.resource import Resource
 
 from daemonize import Daemonize
@@ -120,30 +120,25 @@ class AppserviceServer(HomeServer):
             else:
                 logger.warn("Unrecognized listener type: %s", listener["type"])
 
-    @defer.inlineCallbacks
-    def replicate(self):
-        http_client = self.get_simple_http_client()
-        store = self.get_datastore()
-        replication_url = self.config.worker_replication_url
-        appservice_handler = self.get_application_service_handler()
+        self.get_tcp_replication().start_replication(self)
 
-        @defer.inlineCallbacks
-        def replicate(results):
-            stream = results.get("events")
-            if stream:
-                max_stream_id = stream["position"]
-                yield appservice_handler.notify_interested_services(max_stream_id)
+    def build_tcp_replication(self):
+        return ASReplicationHandler(self)
 
-        while True:
-            try:
-                args = store.stream_positions()
-                args["timeout"] = 30000
-                result = yield http_client.get_json(replication_url, args=args)
-                yield store.process_replication(result)
-                replicate(result)
-            except:
-                logger.exception("Error replicating from %r", replication_url)
-                yield sleep(30)
+
+class ASReplicationHandler(ReplicationClientHandler):
+    def __init__(self, hs):
+        super(ASReplicationHandler, self).__init__(hs.get_datastore())
+        self.appservice_handler = hs.get_application_service_handler()
+
+    def on_rdata(self, stream_name, token, rows):
+        super(ASReplicationHandler, self).on_rdata(stream_name, token, rows)
+
+        if stream_name == "events":
+            max_stream_id = self.store.get_room_max_stream_ordering()
+            preserve_fn(
+                self.appservice_handler.notify_interested_services
+            )(max_stream_id)
 
 
 def start(config_options):
@@ -157,7 +152,7 @@ def start(config_options):
 
     assert config.worker_app == "synapse.app.appservice"
 
-    setup_logging(config.worker_log_config, config.worker_log_file)
+    setup_logging(config, use_worker_options=True)
 
     events.USE_FROZEN_DICTS = config.use_frozen_dicts
 
@@ -187,7 +182,11 @@ def start(config_options):
     ps.start_listening(config.worker_listeners)
 
     def run():
-        with LoggingContext("run"):
+        # make sure that we run the reactor with the sentinel log context,
+        # otherwise other PreserveLoggingContext instances will get confused
+        # and complain when they see the logcontext arbitrarily swapping
+        # between the sentinel and `run` logcontexts.
+        with PreserveLoggingContext():
             logger.info("Running")
             change_resource_limit(config.soft_file_limit)
             if config.gc_thresholds:
@@ -195,7 +194,6 @@ def start(config_options):
             reactor.run()
 
     def start():
-        ps.replicate()
         ps.get_datastore().start_profiling()
         ps.get_state_handler().start_caching()
 

synapse/app/client_reader.py

@@ -29,13 +29,14 @@ from synapse.replication.slave.storage.keys import SlavedKeyStore
 from synapse.replication.slave.storage.room import RoomStore
 from synapse.replication.slave.storage.directory import DirectoryStore
 from synapse.replication.slave.storage.registration import SlavedRegistrationStore
+from synapse.replication.slave.storage.transactions import TransactionStore
+from synapse.replication.tcp.client import ReplicationClientHandler
 from synapse.rest.client.v1.room import PublicRoomListRestServlet
 from synapse.server import HomeServer
 from synapse.storage.client_ips import ClientIpStore
 from synapse.storage.engines import create_engine
-from synapse.util.async import sleep
 from synapse.util.httpresourcetree import create_resource_tree
-from synapse.util.logcontext import LoggingContext
+from synapse.util.logcontext import LoggingContext, PreserveLoggingContext
 from synapse.util.manhole import manhole
 from synapse.util.rlimit import change_resource_limit
 from synapse.util.versionstring import get_version_string
@@ -44,7 +45,7 @@ from synapse.crypto import context_factory
 from synapse import events
 
 
-from twisted.internet import reactor, defer
+from twisted.internet import reactor
 from twisted.web.resource import Resource
 
 from daemonize import Daemonize
@@ -63,6 +64,7 @@ class ClientReaderSlavedStore(
     DirectoryStore,
     SlavedApplicationServiceStore,
     SlavedRegistrationStore,
+    TransactionStore,
     BaseSlavedStore,
     ClientIpStore,  # After BaseSlavedStore because the constructor is different
 ):
@@ -143,21 +145,10 @@ class ClientReaderServer(HomeServer):
             else:
                 logger.warn("Unrecognized listener type: %s", listener["type"])
 
-    @defer.inlineCallbacks
-    def replicate(self):
-        http_client = self.get_simple_http_client()
-        store = self.get_datastore()
-        replication_url = self.config.worker_replication_url
+        self.get_tcp_replication().start_replication(self)
 
-        while True:
-            try:
-                args = store.stream_positions()
-                args["timeout"] = 30000
-                result = yield http_client.get_json(replication_url, args=args)
-                yield store.process_replication(result)
-            except:
-                logger.exception("Error replicating from %r", replication_url)
-                yield sleep(5)
+    def build_tcp_replication(self):
+        return ReplicationClientHandler(self.get_datastore())
 
 
 def start(config_options):
@@ -171,7 +162,7 @@ def start(config_options):
 
     assert config.worker_app == "synapse.app.client_reader"
 
-    setup_logging(config.worker_log_config, config.worker_log_file)
+    setup_logging(config, use_worker_options=True)
 
     events.USE_FROZEN_DICTS = config.use_frozen_dicts
 
@@ -193,7 +184,11 @@ def start(config_options):
     ss.start_listening(config.worker_listeners)
 
     def run():
-        with LoggingContext("run"):
+        # make sure that we run the reactor with the sentinel log context,
+        # otherwise other PreserveLoggingContext instances will get confused
+        # and complain when they see the logcontext arbitrarily swapping
+        # between the sentinel and `run` logcontexts.
+        with PreserveLoggingContext():
             logger.info("Running")
             change_resource_limit(config.soft_file_limit)
             if config.gc_thresholds:
@@ -203,7 +198,6 @@ def start(config_options):
     def start():
         ss.get_state_handler().start_caching()
         ss.get_datastore().start_profiling()
-        ss.replicate()
 
     reactor.callWhenRunning(start)
 

synapse/app/federation_reader.py

@@ -27,11 +27,11 @@ from synapse.replication.slave.storage.keys import SlavedKeyStore
 from synapse.replication.slave.storage.room import RoomStore
 from synapse.replication.slave.storage.transactions import TransactionStore
 from synapse.replication.slave.storage.directory import DirectoryStore
+from synapse.replication.tcp.client import ReplicationClientHandler
 from synapse.server import HomeServer
 from synapse.storage.engines import create_engine
-from synapse.util.async import sleep
 from synapse.util.httpresourcetree import create_resource_tree
-from synapse.util.logcontext import LoggingContext
+from synapse.util.logcontext import LoggingContext, PreserveLoggingContext
 from synapse.util.manhole import manhole
 from synapse.util.rlimit import change_resource_limit
 from synapse.util.versionstring import get_version_string
@@ -42,7 +42,7 @@ from synapse.crypto import context_factory
 from synapse import events
 
 
-from twisted.internet import reactor, defer
+from twisted.internet import reactor
 from twisted.web.resource import Resource
 
 from daemonize import Daemonize
@@ -134,21 +134,10 @@ class FederationReaderServer(HomeServer):
             else:
                 logger.warn("Unrecognized listener type: %s", listener["type"])
 
-    @defer.inlineCallbacks
-    def replicate(self):
-        http_client = self.get_simple_http_client()
-        store = self.get_datastore()
-        replication_url = self.config.worker_replication_url
+        self.get_tcp_replication().start_replication(self)
 
-        while True:
-            try:
-                args = store.stream_positions()
-                args["timeout"] = 30000
-                result = yield http_client.get_json(replication_url, args=args)
-                yield store.process_replication(result)
-            except:
-                logger.exception("Error replicating from %r", replication_url)
-                yield sleep(5)
+    def build_tcp_replication(self):
+        return ReplicationClientHandler(self.get_datastore())
 
 
 def start(config_options):
@@ -162,7 +151,7 @@ def start(config_options):
 
     assert config.worker_app == "synapse.app.federation_reader"
 
-    setup_logging(config.worker_log_config, config.worker_log_file)
+    setup_logging(config, use_worker_options=True)
 
     events.USE_FROZEN_DICTS = config.use_frozen_dicts
 
@@ -184,7 +173,11 @@ def start(config_options):
     ss.start_listening(config.worker_listeners)
 
     def run():
-        with LoggingContext("run"):
+        # make sure that we run the reactor with the sentinel log context,
+        # otherwise other PreserveLoggingContext instances will get confused
+        # and complain when they see the logcontext arbitrarily swapping
+        # between the sentinel and `run` logcontexts.
+        with PreserveLoggingContext():
             logger.info("Running")
             change_resource_limit(config.soft_file_limit)
             if config.gc_thresholds:
@@ -194,7 +187,6 @@ def start(config_options):
     def start():
         ss.get_state_handler().start_caching()
         ss.get_datastore().start_profiling()
-        ss.replicate()
 
     reactor.callWhenRunning(start)
 

synapse/app/federation_sender.py

@@ -23,19 +23,19 @@ from synapse.config.homeserver import HomeServerConfig
 from synapse.crypto import context_factory
 from synapse.http.site import SynapseSite
 from synapse.federation import send_queue
-from synapse.federation.units import Edu
 from synapse.metrics.resource import MetricsResource, METRICS_PREFIX
 from synapse.replication.slave.storage.deviceinbox import SlavedDeviceInboxStore
 from synapse.replication.slave.storage.events import SlavedEventStore
 from synapse.replication.slave.storage.receipts import SlavedReceiptsStore
 from synapse.replication.slave.storage.registration import SlavedRegistrationStore
+from synapse.replication.slave.storage.presence import SlavedPresenceStore
 from synapse.replication.slave.storage.transactions import TransactionStore
 from synapse.replication.slave.storage.devices import SlavedDeviceStore
+from synapse.replication.tcp.client import ReplicationClientHandler
 from synapse.storage.engines import create_engine
-from synapse.storage.presence import UserPresenceState
-from synapse.util.async import sleep
+from synapse.util.async import Linearizer
 from synapse.util.httpresourcetree import create_resource_tree
-from synapse.util.logcontext import LoggingContext
+from synapse.util.logcontext import LoggingContext, PreserveLoggingContext, preserve_fn
 from synapse.util.manhole import manhole
 from synapse.util.rlimit import change_resource_limit
 from synapse.util.versionstring import get_version_string
@@ -50,16 +50,36 @@ from daemonize import Daemonize
 import sys
 import logging
 import gc
-import ujson as json
 
 logger = logging.getLogger("synapse.app.appservice")
 
 
 class FederationSenderSlaveStore(
     SlavedDeviceInboxStore, TransactionStore, SlavedReceiptsStore, SlavedEventStore,
-    SlavedRegistrationStore, SlavedDeviceStore,
+    SlavedRegistrationStore, SlavedDeviceStore, SlavedPresenceStore,
 ):
-    pass
+    def __init__(self, db_conn, hs):
+        super(FederationSenderSlaveStore, self).__init__(db_conn, hs)
+
+        # We pull out the current federation stream position now so that we
+        # always have a known value for the federation position in memory so
+        # that we don't have to bounce via a deferred once when we start the
+        # replication streams.
+        self.federation_out_pos_startup = self._get_federation_out_pos(db_conn)
+
+    def _get_federation_out_pos(self, db_conn):
+        sql = (
+            "SELECT stream_id FROM federation_stream_position"
+            " WHERE type = ?"
+        )
+        sql = self.database_engine.convert_param_style(sql)
+
+        txn = db_conn.cursor()
+        txn.execute(sql, ("federation",))
+        rows = txn.fetchall()
+        txn.close()
+
+        return rows[0][0] if rows else -1
 
 
 class FederationSenderServer(HomeServer):
@@ -127,26 +147,27 @@ class FederationSenderServer(HomeServer):
             else:
                 logger.warn("Unrecognized listener type: %s", listener["type"])
 
-    @defer.inlineCallbacks
-    def replicate(self):
-        http_client = self.get_simple_http_client()
-        store = self.get_datastore()
-        replication_url = self.config.worker_replication_url
-        send_handler = FederationSenderHandler(self)
+        self.get_tcp_replication().start_replication(self)
 
-        send_handler.on_start()
+    def build_tcp_replication(self):
+        return FederationSenderReplicationHandler(self)
 
-        while True:
-            try:
-                args = store.stream_positions()
-                args.update((yield send_handler.stream_positions()))
-                args["timeout"] = 30000
-                result = yield http_client.get_json(replication_url, args=args)
-                yield store.process_replication(result)
-                yield send_handler.process_replication(result)
-            except:
-                logger.exception("Error replicating from %r", replication_url)
-                yield sleep(30)
+
+class FederationSenderReplicationHandler(ReplicationClientHandler):
+    def __init__(self, hs):
+        super(FederationSenderReplicationHandler, self).__init__(hs.get_datastore())
+        self.send_handler = FederationSenderHandler(hs, self)
+
+    def on_rdata(self, stream_name, token, rows):
+        super(FederationSenderReplicationHandler, self).on_rdata(
+            stream_name, token, rows
+        )
+        self.send_handler.process_replication_rows(stream_name, token, rows)
+
+    def get_streams_to_replicate(self):
+        args = super(FederationSenderReplicationHandler, self).get_streams_to_replicate()
+        args.update(self.send_handler.stream_positions())
+        return args
 
 
 def start(config_options):
@@ -160,7 +181,7 @@ def start(config_options):
 
     assert config.worker_app == "synapse.app.federation_sender"
 
-    setup_logging(config.worker_log_config, config.worker_log_file)
+    setup_logging(config, use_worker_options=True)
 
     events.USE_FROZEN_DICTS = config.use_frozen_dicts
 
@@ -193,7 +214,11 @@ def start(config_options):
     ps.start_listening(config.worker_listeners)
 
     def run():
-        with LoggingContext("run"):
+        # make sure that we run the reactor with the sentinel log context,
+        # otherwise other PreserveLoggingContext instances will get confused
+        # and complain when they see the logcontext arbitrarily swapping
+        # between the sentinel and `run` logcontexts.
+        with PreserveLoggingContext():
             logger.info("Running")
             change_resource_limit(config.soft_file_limit)
             if config.gc_thresholds:
@@ -201,7 +226,6 @@ def start(config_options):
             reactor.run()
 
     def start():
-        ps.replicate()
         ps.get_datastore().start_profiling()
         ps.get_state_handler().start_caching()
 
@@ -225,9 +249,15 @@ class FederationSenderHandler(object):
     """Processes the replication stream and forwards the appropriate entries
     to the federation sender.
     """
-    def __init__(self, hs):
+    def __init__(self, hs, replication_client):
         self.store = hs.get_datastore()
         self.federation_sender = hs.get_federation_sender()
+        self.replication_client = replication_client
+
+        self.federation_position = self.store.federation_out_pos_startup
+        self._fed_position_linearizer = Linearizer(name="_fed_position_linearizer")
+
+        self._last_ack = self.federation_position
 
         self._room_serials = {}
         self._room_typing = {}
@@ -239,98 +269,35 @@ class FederationSenderHandler(object):
             self.store.get_room_max_stream_ordering()
         )
 
-    @defer.inlineCallbacks
     def stream_positions(self):
-        stream_id = yield self.store.get_federation_out_pos("federation")
-        defer.returnValue({
-            "federation": stream_id,
+        return {"federation": self.federation_position}
 
-            # Ack stuff we've "processed", this should only be called from
-            # one process.
-            "federation_ack": stream_id,
-        })
-
-    @defer.inlineCallbacks
-    def process_replication(self, result):
+    def process_replication_rows(self, stream_name, token, rows):
         # The federation stream contains things that we want to send out, e.g.
         # presence, typing, etc.
-        fed_stream = result.get("federation")
-        if fed_stream:
-            latest_id = int(fed_stream["position"])
-
-            # The federation stream containis a bunch of different types of
-            # rows that need to be handled differently. We parse the rows, put
-            # them into the appropriate collection and then send them off.
-            presence_to_send = {}
-            keyed_edus = {}
-            edus = {}
-            failures = {}
-            device_destinations = set()
-
-            # Parse the rows in the stream
-            for row in fed_stream["rows"]:
-                position, typ, content_js = row
-                content = json.loads(content_js)
-
-                if typ == send_queue.PRESENCE_TYPE:
-                    destination = content["destination"]
-                    state = UserPresenceState.from_dict(content["state"])
-
-                    presence_to_send.setdefault(destination, []).append(state)
-                elif typ == send_queue.KEYED_EDU_TYPE:
-                    key = content["key"]
-                    edu = Edu(**content["edu"])
-
-                    keyed_edus.setdefault(
-                        edu.destination, {}
-                    )[(edu.destination, tuple(key))] = edu
-                elif typ == send_queue.EDU_TYPE:
-                    edu = Edu(**content)
-
-                    edus.setdefault(edu.destination, []).append(edu)
-                elif typ == send_queue.FAILURE_TYPE:
-                    destination = content["destination"]
-                    failure = content["failure"]
-
-                    failures.setdefault(destination, []).append(failure)
-                elif typ == send_queue.DEVICE_MESSAGE_TYPE:
-                    device_destinations.add(content["destination"])
-                else:
-                    raise Exception("Unrecognised federation type: %r", typ)
-
-            # We've finished collecting, send everything off
-            for destination, states in presence_to_send.items():
-                self.federation_sender.send_presence(destination, states)
-
-            for destination, edu_map in keyed_edus.items():
-                for key, edu in edu_map.items():
-                    self.federation_sender.send_edu(
-                        edu.destination, edu.edu_type, edu.content, key=key,
-                    )
-
-            for destination, edu_list in edus.items():
-                for edu in edu_list:
-                    self.federation_sender.send_edu(
-                        edu.destination, edu.edu_type, edu.content, key=None,
-                    )
-
-            for destination, failure_list in failures.items():
-                for failure in failure_list:
-                    self.federation_sender.send_failure(destination, failure)
-
-            for destination in device_destinations:
-                self.federation_sender.send_device_messages(destination)
-
-            # Record where we are in the stream.
-            yield self.store.update_federation_out_pos(
-                "federation", latest_id
-            )
+        if stream_name == "federation":
+            send_queue.process_rows_for_federation(self.federation_sender, rows)
+            preserve_fn(self.update_token)(token)
 
         # We also need to poke the federation sender when new events happen
-        event_stream = result.get("events")
-        if event_stream:
-            latest_pos = event_stream["position"]
-            self.federation_sender.notify_new_events(latest_pos)
+        elif stream_name == "events":
+            self.federation_sender.notify_new_events(token)
+
+    @defer.inlineCallbacks
+    def update_token(self, token):
+        self.federation_position = token
+
+        # We linearize here to ensure we don't have races updating the token
+        with (yield self._fed_position_linearizer.queue(None)):
+            if self._last_ack < self.federation_position:
+                yield self.store.update_federation_out_pos(
+                    "federation", self.federation_position
+                )
+
+                # We ACK this token over replication so that the master can drop
+                # its in memory queues
+                self.replication_client.send_federation_ack(self.federation_position)
+                self._last_ack = self.federation_position
 
 
 if __name__ == '__main__':

synapse/app/homeserver.py

@@ -20,10 +20,12 @@ import gc
 import logging
 import os
 import sys
+
+import synapse.config.logger
 from synapse.config._base import ConfigError
 
 from synapse.python_dependencies import (
-    check_requirements, DEPENDENCY_LINKS
+    check_requirements, CONDITIONAL_REQUIREMENTS
 )
 
 from synapse.rest import ClientRestResource
@@ -50,10 +52,10 @@ from synapse.api.urls import (
 )
 from synapse.config.homeserver import HomeServerConfig
 from synapse.crypto import context_factory
-from synapse.util.logcontext import LoggingContext
+from synapse.util.logcontext import LoggingContext, PreserveLoggingContext
 from synapse.metrics import register_memory_metrics, get_metrics_for
 from synapse.metrics.resource import MetricsResource, METRICS_PREFIX
-from synapse.replication.resource import ReplicationResource, REPLICATION_PREFIX
+from synapse.replication.tcp.resource import ReplicationStreamProtocolFactory
 from synapse.federation.transport.server import TransportLayerServer
 
 from synapse.util.rlimit import change_resource_limit
@@ -90,7 +92,7 @@ def build_resource_for_web_client(hs):
                 "\n"
                 "You can also disable hosting of the webclient via the\n"
                 "configuration option `web_client`\n"
-                % {"dep": DEPENDENCY_LINKS["matrix-angular-sdk"]}
+                % {"dep": CONDITIONAL_REQUIREMENTS["web_client"].keys()[0]}
             )
         syweb_path = os.path.dirname(syweb.__file__)
         webclient_path = os.path.join(syweb_path, "webclient")
@@ -164,9 +166,6 @@ class SynapseHomeServer(HomeServer):
                 if name == "metrics" and self.get_config().enable_metrics:
                     resources[METRICS_PREFIX] = MetricsResource(self)
 
-                if name == "replication":
-                    resources[REPLICATION_PREFIX] = ReplicationResource(self)
-
         if WEB_CLIENT_PREFIX in resources:
             root_resource = RootRedirect(WEB_CLIENT_PREFIX)
         else:
@@ -220,6 +219,16 @@ class SynapseHomeServer(HomeServer):
                         ),
                         interface=address
                     )
+            elif listener["type"] == "replication":
+                bind_addresses = listener["bind_addresses"]
+                for address in bind_addresses:
+                    factory = ReplicationStreamProtocolFactory(self)
+                    server_listener = reactor.listenTCP(
+                        listener["port"], factory, interface=address
+                    )
+                    reactor.addSystemEventTrigger(
+                        "before", "shutdown", server_listener.stopListening,
+                    )
             else:
                 logger.warn("Unrecognized listener type: %s", listener["type"])
 
@@ -286,7 +295,7 @@ def setup(config_options):
         # generating config files and shouldn't try to continue.
         sys.exit(0)
 
-    config.setup_logging()
+    synapse.config.logger.setup_logging(config, use_worker_options=False)
 
     # check any extra requirements we have now we have a config
     check_requirements(config)
@@ -454,7 +463,12 @@ def run(hs):
     def in_thread():
         # Uncomment to enable tracing of log context changes.
         # sys.settrace(logcontext_tracer)
-        with LoggingContext("run"):
+
+        # make sure that we run the reactor with the sentinel log context,
+        # otherwise other PreserveLoggingContext instances will get confused
+        # and complain when they see the logcontext arbitrarily swapping
+        # between the sentinel and `run` logcontexts.
+        with PreserveLoggingContext():
             change_resource_limit(hs.config.soft_file_limit)
             if hs.config.gc_thresholds:
                 gc.set_threshold(*hs.config.gc_thresholds)

synapse/app/media_repository.py

@@ -24,15 +24,16 @@ from synapse.metrics.resource import MetricsResource, METRICS_PREFIX
 from synapse.replication.slave.storage._base import BaseSlavedStore
 from synapse.replication.slave.storage.appservice import SlavedApplicationServiceStore
 from synapse.replication.slave.storage.registration import SlavedRegistrationStore
+from synapse.replication.slave.storage.transactions import TransactionStore
+from synapse.replication.tcp.client import ReplicationClientHandler
 from synapse.rest.media.v0.content_repository import ContentRepoResource
 from synapse.rest.media.v1.media_repository import MediaRepositoryResource
 from synapse.server import HomeServer
 from synapse.storage.client_ips import ClientIpStore
 from synapse.storage.engines import create_engine
 from synapse.storage.media_repository import MediaRepositoryStore
-from synapse.util.async import sleep
 from synapse.util.httpresourcetree import create_resource_tree
-from synapse.util.logcontext import LoggingContext
+from synapse.util.logcontext import LoggingContext, PreserveLoggingContext
 from synapse.util.manhole import manhole
 from synapse.util.rlimit import change_resource_limit
 from synapse.util.versionstring import get_version_string
@@ -44,7 +45,7 @@ from synapse.crypto import context_factory
 from synapse import events
 
 
-from twisted.internet import reactor, defer
+from twisted.internet import reactor
 from twisted.web.resource import Resource
 
 from daemonize import Daemonize
@@ -59,6 +60,7 @@ logger = logging.getLogger("synapse.app.media_repository")
 class MediaRepositorySlavedStore(
     SlavedApplicationServiceStore,
     SlavedRegistrationStore,
+    TransactionStore,
     BaseSlavedStore,
     MediaRepositoryStore,
     ClientIpStore,
@@ -140,21 +142,10 @@ class MediaRepositoryServer(HomeServer):
             else:
                 logger.warn("Unrecognized listener type: %s", listener["type"])
 
-    @defer.inlineCallbacks
-    def replicate(self):
-        http_client = self.get_simple_http_client()
-        store = self.get_datastore()
-        replication_url = self.config.worker_replication_url
+        self.get_tcp_replication().start_replication(self)
 
-        while True:
-            try:
-                args = store.stream_positions()
-                args["timeout"] = 30000
-                result = yield http_client.get_json(replication_url, args=args)
-                yield store.process_replication(result)
-            except:
-                logger.exception("Error replicating from %r", replication_url)
-                yield sleep(5)
+    def build_tcp_replication(self):
+        return ReplicationClientHandler(self.get_datastore())
 
 
 def start(config_options):
@@ -168,7 +159,7 @@ def start(config_options):
 
     assert config.worker_app == "synapse.app.media_repository"
 
-    setup_logging(config.worker_log_config, config.worker_log_file)
+    setup_logging(config, use_worker_options=True)
 
     events.USE_FROZEN_DICTS = config.use_frozen_dicts
 
@@ -190,7 +181,11 @@ def start(config_options):
     ss.start_listening(config.worker_listeners)
 
     def run():
-        with LoggingContext("run"):
+        # make sure that we run the reactor with the sentinel log context,
+        # otherwise other PreserveLoggingContext instances will get confused
+        # and complain when they see the logcontext arbitrarily swapping
+        # between the sentinel and `run` logcontexts.
+        with PreserveLoggingContext():
             logger.info("Running")
             change_resource_limit(config.soft_file_limit)
             if config.gc_thresholds:
@@ -200,7 +195,6 @@ def start(config_options):
     def start():
         ss.get_state_handler().start_caching()
         ss.get_datastore().start_profiling()
-        ss.replicate()
 
     reactor.callWhenRunning(start)
 

synapse/app/pusher.py

@@ -27,11 +27,12 @@ from synapse.replication.slave.storage.events import SlavedEventStore
 from synapse.replication.slave.storage.pushers import SlavedPusherStore
 from synapse.replication.slave.storage.receipts import SlavedReceiptsStore
 from synapse.replication.slave.storage.account_data import SlavedAccountDataStore
+from synapse.replication.tcp.client import ReplicationClientHandler
 from synapse.storage.engines import create_engine
 from synapse.storage import DataStore
-from synapse.util.async import sleep
 from synapse.util.httpresourcetree import create_resource_tree
-from synapse.util.logcontext import LoggingContext, preserve_fn
+from synapse.util.logcontext import LoggingContext, preserve_fn, \
+    PreserveLoggingContext
 from synapse.util.manhole import manhole
 from synapse.util.rlimit import change_resource_limit
 from synapse.util.versionstring import get_version_string
@@ -88,7 +89,6 @@ class PusherSlaveStore(
 
 
 class PusherServer(HomeServer):
-
     def get_db_conn(self, run_new_connection=True):
         # Any param beginning with cp_ is a parameter for adbapi, and should
         # not be passed to the database engine.
@@ -108,16 +108,7 @@ class PusherServer(HomeServer):
         logger.info("Finished setting up.")
 
     def remove_pusher(self, app_id, push_key, user_id):
-        http_client = self.get_simple_http_client()
-        replication_url = self.config.worker_replication_url
-        url = replication_url + "/remove_pushers"
-        return http_client.post_json_get_json(url, {
-            "remove": [{
-                "app_id": app_id,
-                "push_key": push_key,
-                "user_id": user_id,
-            }]
-        })
+        self.get_tcp_replication().send_remove_pusher(app_id, push_key, user_id)
 
     def _listen_http(self, listener_config):
         port = listener_config["port"]
@@ -165,73 +156,52 @@ class PusherServer(HomeServer):
             else:
                 logger.warn("Unrecognized listener type: %s", listener["type"])
 
+        self.get_tcp_replication().start_replication(self)
+
+    def build_tcp_replication(self):
+        return PusherReplicationHandler(self)
+
+
+class PusherReplicationHandler(ReplicationClientHandler):
+    def __init__(self, hs):
+        super(PusherReplicationHandler, self).__init__(hs.get_datastore())
+
+        self.pusher_pool = hs.get_pusherpool()
+
+    def on_rdata(self, stream_name, token, rows):
+        super(PusherReplicationHandler, self).on_rdata(stream_name, token, rows)
+        preserve_fn(self.poke_pushers)(stream_name, token, rows)
+
     @defer.inlineCallbacks
-    def replicate(self):
-        http_client = self.get_simple_http_client()
-        store = self.get_datastore()
-        replication_url = self.config.worker_replication_url
-        pusher_pool = self.get_pusherpool()
-
-        def stop_pusher(user_id, app_id, pushkey):
-            key = "%s:%s" % (app_id, pushkey)
-            pushers_for_user = pusher_pool.pushers.get(user_id, {})
-            pusher = pushers_for_user.pop(key, None)
-            if pusher is None:
-                return
-            logger.info("Stopping pusher %r / %r", user_id, key)
-            pusher.on_stop()
-
-        def start_pusher(user_id, app_id, pushkey):
-            key = "%s:%s" % (app_id, pushkey)
-            logger.info("Starting pusher %r / %r", user_id, key)
-            return pusher_pool._refresh_pusher(app_id, pushkey, user_id)
-
-        @defer.inlineCallbacks
-        def poke_pushers(results):
-            pushers_rows = set(
-                map(tuple, results.get("pushers", {}).get("rows", []))
+    def poke_pushers(self, stream_name, token, rows):
+        if stream_name == "pushers":
+            for row in rows:
+                if row.deleted:
+                    yield self.stop_pusher(row.user_id, row.app_id, row.pushkey)
+                else:
+                    yield self.start_pusher(row.user_id, row.app_id, row.pushkey)
+        elif stream_name == "events":
+            yield self.pusher_pool.on_new_notifications(
+                token, token,
             )
-            deleted_pushers_rows = set(
-                map(tuple, results.get("deleted_pushers", {}).get("rows", []))
+        elif stream_name == "receipts":
+            yield self.pusher_pool.on_new_receipts(
+                token, token, set(row.room_id for row in rows)
             )
-            for row in sorted(pushers_rows | deleted_pushers_rows):
-                if row in deleted_pushers_rows:
-                    user_id, app_id, pushkey = row[1:4]
-                    stop_pusher(user_id, app_id, pushkey)
-                elif row in pushers_rows:
-                    user_id = row[1]
-                    app_id = row[5]
-                    pushkey = row[8]
-                    yield start_pusher(user_id, app_id, pushkey)
 
-            stream = results.get("events")
-            if stream and stream["rows"]:
-                min_stream_id = stream["rows"][0][0]
-                max_stream_id = stream["position"]
-                preserve_fn(pusher_pool.on_new_notifications)(
-                    min_stream_id, max_stream_id
-                )
+    def stop_pusher(self, user_id, app_id, pushkey):
+        key = "%s:%s" % (app_id, pushkey)
+        pushers_for_user = self.pusher_pool.pushers.get(user_id, {})
+        pusher = pushers_for_user.pop(key, None)
+        if pusher is None:
+            return
+        logger.info("Stopping pusher %r / %r", user_id, key)
+        pusher.on_stop()
 
-            stream = results.get("receipts")
-            if stream and stream["rows"]:
-                rows = stream["rows"]
-                affected_room_ids = set(row[1] for row in rows)
-                min_stream_id = rows[0][0]
-                max_stream_id = stream["position"]
-                preserve_fn(pusher_pool.on_new_receipts)(
-                    min_stream_id, max_stream_id, affected_room_ids
-                )
-
-        while True:
-            try:
-                args = store.stream_positions()
-                args["timeout"] = 30000
-                result = yield http_client.get_json(replication_url, args=args)
-                yield store.process_replication(result)
-                poke_pushers(result)
-            except:
-                logger.exception("Error replicating from %r", replication_url)
-                yield sleep(30)
+    def start_pusher(self, user_id, app_id, pushkey):
+        key = "%s:%s" % (app_id, pushkey)
+        logger.info("Starting pusher %r / %r", user_id, key)
+        return self.pusher_pool._refresh_pusher(app_id, pushkey, user_id)
 
 
 def start(config_options):
@@ -245,7 +215,7 @@ def start(config_options):
 
     assert config.worker_app == "synapse.app.pusher"
 
-    setup_logging(config.worker_log_config, config.worker_log_file)
+    setup_logging(config, use_worker_options=True)
 
     events.USE_FROZEN_DICTS = config.use_frozen_dicts
 
@@ -275,7 +245,11 @@ def start(config_options):
     ps.start_listening(config.worker_listeners)
 
     def run():
-        with LoggingContext("run"):
+        # make sure that we run the reactor with the sentinel log context,
+        # otherwise other PreserveLoggingContext instances will get confused
+        # and complain when they see the logcontext arbitrarily swapping
+        # between the sentinel and `run` logcontexts.
+        with PreserveLoggingContext():
             logger.info("Running")
             change_resource_limit(config.soft_file_limit)
             if config.gc_thresholds:
@@ -283,7 +257,6 @@ def start(config_options):
             reactor.run()
 
     def start():
-        ps.replicate()
         ps.get_pusherpool().start()
         ps.get_datastore().start_profiling()
         ps.get_state_handler().start_caching()

synapse/app/synchrotron.py

@@ -16,12 +16,11 @@
 
 import synapse
 
-from synapse.api.constants import EventTypes, PresenceState
+from synapse.api.constants import EventTypes
 from synapse.config._base import ConfigError
 from synapse.config.homeserver import HomeServerConfig
 from synapse.config.logger import setup_logging
-from synapse.events import FrozenEvent
-from synapse.handlers.presence import PresenceHandler
+from synapse.handlers.presence import PresenceHandler, get_interested_parties
 from synapse.http.site import SynapseSite
 from synapse.http.server import JsonResource
 from synapse.metrics.resource import MetricsResource, METRICS_PREFIX
@@ -41,14 +40,14 @@ from synapse.replication.slave.storage.presence import SlavedPresenceStore
 from synapse.replication.slave.storage.deviceinbox import SlavedDeviceInboxStore
 from synapse.replication.slave.storage.devices import SlavedDeviceStore
 from synapse.replication.slave.storage.room import RoomStore
+from synapse.replication.tcp.client import ReplicationClientHandler
 from synapse.server import HomeServer
 from synapse.storage.client_ips import ClientIpStore
 from synapse.storage.engines import create_engine
-from synapse.storage.presence import PresenceStore, UserPresenceState
+from synapse.storage.presence import UserPresenceState
 from synapse.storage.roommember import RoomMemberStore
-from synapse.util.async import sleep
 from synapse.util.httpresourcetree import create_resource_tree
-from synapse.util.logcontext import LoggingContext, preserve_fn
+from synapse.util.logcontext import LoggingContext, PreserveLoggingContext, preserve_fn
 from synapse.util.manhole import manhole
 from synapse.util.rlimit import change_resource_limit
 from synapse.util.stringutils import random_string
@@ -63,7 +62,6 @@ import sys
 import logging
 import contextlib
 import gc
-import ujson as json
 
 logger = logging.getLogger("synapse.app.synchrotron")
 
@@ -87,15 +85,9 @@ class SynchrotronSlavedStore(
         RoomMemberStore.__dict__["who_forgot_in_room"]
     )
 
-    # XXX: This is a bit broken because we don't persist the accepted list in a
-    # way that can be replicated. This means that we don't have a way to
-    # invalidate the cache correctly.
-    get_presence_list_accepted = PresenceStore.__dict__[
-        "get_presence_list_accepted"
-    ]
-    get_presence_list_observers_accepted = PresenceStore.__dict__[
-        "get_presence_list_observers_accepted"
-    ]
+    did_forget = (
+        RoomMemberStore.__dict__["did_forget"]
+    )
 
 
 UPDATE_SYNCING_USERS_MS = 10 * 1000
@@ -103,11 +95,11 @@ UPDATE_SYNCING_USERS_MS = 10 * 1000
 
 class SynchrotronPresence(object):
     def __init__(self, hs):
+        self.hs = hs
         self.is_mine_id = hs.is_mine_id
         self.http_client = hs.get_simple_http_client()
         self.store = hs.get_datastore()
         self.user_to_num_current_syncs = {}
-        self.syncing_users_url = hs.config.worker_replication_url + "/syncing_users"
         self.clock = hs.get_clock()
         self.notifier = hs.get_notifier()
 
@@ -117,17 +109,52 @@ class SynchrotronPresence(object):
             for state in active_presence
         }
 
+        # user_id -> last_sync_ms. Lists the users that have stopped syncing
+        # but we haven't notified the master of that yet
+        self.users_going_offline = {}
+
+        self._send_stop_syncing_loop = self.clock.looping_call(
+            self.send_stop_syncing, 10 * 1000
+        )
+
         self.process_id = random_string(16)
         logger.info("Presence process_id is %r", self.process_id)
 
-        self._sending_sync = False
-        self._need_to_send_sync = False
-        self.clock.looping_call(
-            self._send_syncing_users_regularly,
-            UPDATE_SYNCING_USERS_MS,
-        )
+    def send_user_sync(self, user_id, is_syncing, last_sync_ms):
+        self.hs.get_tcp_replication().send_user_sync(user_id, is_syncing, last_sync_ms)
 
-        reactor.addSystemEventTrigger("before", "shutdown", self._on_shutdown)
+    def mark_as_coming_online(self, user_id):
+        """A user has started syncing. Send a UserSync to the master, unless they
+        had recently stopped syncing.
+
+        Args:
+            user_id (str)
+        """
+        going_offline = self.users_going_offline.pop(user_id, None)
+        if not going_offline:
+            # Safe to skip because we haven't yet told the master they were offline
+            self.send_user_sync(user_id, True, self.clock.time_msec())
+
+    def mark_as_going_offline(self, user_id):
+        """A user has stopped syncing. We wait before notifying the master as
+        its likely they'll come back soon. This allows us to avoid sending
+        a stopped syncing immediately followed by a started syncing notification
+        to the master
+
+        Args:
+            user_id (str)
+        """
+        self.users_going_offline[user_id] = self.clock.time_msec()
+
+    def send_stop_syncing(self):
+        """Check if there are any users who have stopped syncing a while ago
+        and haven't come back yet. If there are poke the master about them.
+        """
+        now = self.clock.time_msec()
+        for user_id, last_sync_ms in self.users_going_offline.items():
+            if now - last_sync_ms > 10 * 1000:
+                self.users_going_offline.pop(user_id, None)
+                self.send_user_sync(user_id, False, last_sync_ms)
 
     def set_state(self, user, state, ignore_status_msg=False):
         # TODO Hows this supposed to work?
@@ -135,18 +162,16 @@ class SynchrotronPresence(object):
 
     get_states = PresenceHandler.get_states.__func__
     get_state = PresenceHandler.get_state.__func__
-    _get_interested_parties = PresenceHandler._get_interested_parties.__func__
     current_state_for_users = PresenceHandler.current_state_for_users.__func__
 
-    @defer.inlineCallbacks
     def user_syncing(self, user_id, affect_presence):
         if affect_presence:
             curr_sync = self.user_to_num_current_syncs.get(user_id, 0)
             self.user_to_num_current_syncs[user_id] = curr_sync + 1
-            prev_states = yield self.current_state_for_users([user_id])
-            if prev_states[user_id].state == PresenceState.OFFLINE:
-                # TODO: Don't block the sync request on this HTTP hit.
-                yield self._send_syncing_users_now()
+
+            # If we went from no in flight sync to some, notify replication
+            if self.user_to_num_current_syncs[user_id] == 1:
+                self.mark_as_coming_online(user_id)
 
         def _end():
             # We check that the user_id is in user_to_num_current_syncs because
@@ -155,6 +180,10 @@ class SynchrotronPresence(object):
             if affect_presence and user_id in self.user_to_num_current_syncs:
                 self.user_to_num_current_syncs[user_id] -= 1
 
+                # If we went from one in flight sync to non, notify replication
+                if self.user_to_num_current_syncs[user_id] == 0:
+                    self.mark_as_going_offline(user_id)
+
         @contextlib.contextmanager
         def _user_syncing():
             try:
@@ -162,56 +191,12 @@ class SynchrotronPresence(object):
             finally:
                 _end()
 
-        defer.returnValue(_user_syncing())
-
-    @defer.inlineCallbacks
-    def _on_shutdown(self):
-        # When the synchrotron is shutdown tell the master to clear the in
-        # progress syncs for this process
-        self.user_to_num_current_syncs.clear()
-        yield self._send_syncing_users_now()
-
-    def _send_syncing_users_regularly(self):
-        # Only send an update if we aren't in the middle of sending one.
-        if not self._sending_sync:
-            preserve_fn(self._send_syncing_users_now)()
-
-    @defer.inlineCallbacks
-    def _send_syncing_users_now(self):
-        if self._sending_sync:
-            # We don't want to race with sending another update.
-            # Instead we wait for that update to finish and send another
-            # update afterwards.
-            self._need_to_send_sync = True
-            return
-
-        # Flag that we are sending an update.
-        self._sending_sync = True
-
-        yield self.http_client.post_json_get_json(self.syncing_users_url, {
-            "process_id": self.process_id,
-            "syncing_users": [
-                user_id for user_id, count in self.user_to_num_current_syncs.items()
-                if count > 0
-            ],
-        })
-
-        # Unset the flag as we are no longer sending an update.
-        self._sending_sync = False
-        if self._need_to_send_sync:
-            # If something happened while we were sending the update then
-            # we might need to send another update.
-            # TODO: Check if the update that was sent matches the current state
-            # as we only need to send an update if they are different.
-            self._need_to_send_sync = False
-            yield self._send_syncing_users_now()
+        return defer.succeed(_user_syncing())
 
     @defer.inlineCallbacks
     def notify_from_replication(self, states, stream_id):
-        parties = yield self._get_interested_parties(
-            states, calculate_remote_hosts=False
-        )
-        room_ids_to_states, users_to_states, _ = parties
+        parties = yield get_interested_parties(self.store, states)
+        room_ids_to_states, users_to_states = parties
 
         self.notifier.on_new_event(
             "presence_key", stream_id, rooms=room_ids_to_states.keys(),
@@ -219,26 +204,24 @@ class SynchrotronPresence(object):
         )
 
     @defer.inlineCallbacks
-    def process_replication(self, result):
-        stream = result.get("presence", {"rows": []})
-        states = []
-        for row in stream["rows"]:
-            (
-                position, user_id, state, last_active_ts,
-                last_federation_update_ts, last_user_sync_ts, status_msg,
-                currently_active
-            ) = row
-            state = UserPresenceState(
-                user_id, state, last_active_ts,
-                last_federation_update_ts, last_user_sync_ts, status_msg,
-                currently_active
-            )
-            self.user_to_current_state[user_id] = state
-            states.append(state)
+    def process_replication_rows(self, token, rows):
+        states = [UserPresenceState(
+            row.user_id, row.state, row.last_active_ts,
+            row.last_federation_update_ts, row.last_user_sync_ts, row.status_msg,
+            row.currently_active
+        ) for row in rows]
 
-        if states and "position" in stream:
-            stream_id = int(stream["position"])
-            yield self.notify_from_replication(states, stream_id)
+        for state in states:
+            self.user_to_current_state[row.user_id] = state
+
+        stream_id = token
+        yield self.notify_from_replication(states, stream_id)
+
+    def get_currently_syncing_users(self):
+        return [
+            user_id for user_id, count in self.user_to_num_current_syncs.iteritems()
+            if count > 0
+        ]
 
 
 class SynchrotronTyping(object):
@@ -253,16 +236,12 @@ class SynchrotronTyping(object):
         # value which we *must* use for the next replication request.
         return {"typing": self._latest_room_serial}
 
-    def process_replication(self, result):
-        stream = result.get("typing")
-        if stream:
-            self._latest_room_serial = int(stream["position"])
+    def process_replication_rows(self, token, rows):
+        self._latest_room_serial = token
 
-            for row in stream["rows"]:
-                position, room_id, typing_json = row
-                typing = json.loads(typing_json)
-                self._room_serials[room_id] = position
-                self._room_typing[room_id] = typing
+        for row in rows:
+            self._room_serials[row.room_id] = token
+            self._room_typing[row.room_id] = row.user_ids
 
 
 class SynchrotronApplicationService(object):
@@ -347,114 +326,10 @@ class SynchrotronServer(HomeServer):
             else:
                 logger.warn("Unrecognized listener type: %s", listener["type"])
 
-    @defer.inlineCallbacks
-    def replicate(self):
-        http_client = self.get_simple_http_client()
-        store = self.get_datastore()
-        replication_url = self.config.worker_replication_url
-        notifier = self.get_notifier()
-        presence_handler = self.get_presence_handler()
-        typing_handler = self.get_typing_handler()
+        self.get_tcp_replication().start_replication(self)
 
-        def notify_from_stream(
-            result, stream_name, stream_key, room=None, user=None
-        ):
-            stream = result.get(stream_name)
-            if stream:
-                position_index = stream["field_names"].index("position")
-                if room:
-                    room_index = stream["field_names"].index(room)
-                if user:
-                    user_index = stream["field_names"].index(user)
-
-                users = ()
-                rooms = ()
-                for row in stream["rows"]:
-                    position = row[position_index]
-
-                    if user:
-                        users = (row[user_index],)
-
-                    if room:
-                        rooms = (row[room_index],)
-
-                    notifier.on_new_event(
-                        stream_key, position, users=users, rooms=rooms
-                    )
-
-        @defer.inlineCallbacks
-        def notify_device_list_update(result):
-            stream = result.get("device_lists")
-            if not stream:
-                return
-
-            position_index = stream["field_names"].index("position")
-            user_index = stream["field_names"].index("user_id")
-
-            for row in stream["rows"]:
-                position = row[position_index]
-                user_id = row[user_index]
-
-                rooms = yield store.get_rooms_for_user(user_id)
-                room_ids = [r.room_id for r in rooms]
-
-                notifier.on_new_event(
-                    "device_list_key", position, rooms=room_ids,
-                )
-
-        @defer.inlineCallbacks
-        def notify(result):
-            stream = result.get("events")
-            if stream:
-                max_position = stream["position"]
-                for row in stream["rows"]:
-                    position = row[0]
-                    internal = json.loads(row[1])
-                    event_json = json.loads(row[2])
-                    event = FrozenEvent(event_json, internal_metadata_dict=internal)
-                    extra_users = ()
-                    if event.type == EventTypes.Member:
-                        extra_users = (event.state_key,)
-                    notifier.on_new_room_event(
-                        event, position, max_position, extra_users
-                    )
-
-            notify_from_stream(
-                result, "push_rules", "push_rules_key", user="user_id"
-            )
-            notify_from_stream(
-                result, "user_account_data", "account_data_key", user="user_id"
-            )
-            notify_from_stream(
-                result, "room_account_data", "account_data_key", user="user_id"
-            )
-            notify_from_stream(
-                result, "tag_account_data", "account_data_key", user="user_id"
-            )
-            notify_from_stream(
-                result, "receipts", "receipt_key", room="room_id"
-            )
-            notify_from_stream(
-                result, "typing", "typing_key", room="room_id"
-            )
-            notify_from_stream(
-                result, "to_device", "to_device_key", user="user_id"
-            )
-            yield notify_device_list_update(result)
-
-        while True:
-            try:
-                args = store.stream_positions()
-                args.update(typing_handler.stream_positions())
-                args["timeout"] = 30000
-                result = yield http_client.get_json(replication_url, args=args)
-                yield store.process_replication(result)
-                typing_handler.process_replication(result)
-                yield presence_handler.process_replication(result)
-                yield notify(result)
-            except:
-                logger.exception("Error replicating from %r", replication_url)
-                yield sleep(5)
+    def build_tcp_replication(self):
+        return SyncReplicationHandler(self)
 
     def build_presence_handler(self):
         return SynchrotronPresence(self)
@@ -463,6 +338,79 @@ class SynchrotronServer(HomeServer):
         return SynchrotronTyping(self)
 
 
+class SyncReplicationHandler(ReplicationClientHandler):
+    def __init__(self, hs):
+        super(SyncReplicationHandler, self).__init__(hs.get_datastore())
+
+        self.store = hs.get_datastore()
+        self.typing_handler = hs.get_typing_handler()
+        self.presence_handler = hs.get_presence_handler()
+        self.notifier = hs.get_notifier()
+
+        self.presence_handler.sync_callback = self.send_user_sync
+
+    def on_rdata(self, stream_name, token, rows):
+        super(SyncReplicationHandler, self).on_rdata(stream_name, token, rows)
+
+        preserve_fn(self.process_and_notify)(stream_name, token, rows)
+
+    def get_streams_to_replicate(self):
+        args = super(SyncReplicationHandler, self).get_streams_to_replicate()
+        args.update(self.typing_handler.stream_positions())
+        return args
+
+    def get_currently_syncing_users(self):
+        return self.presence_handler.get_currently_syncing_users()
+
+    @defer.inlineCallbacks
+    def process_and_notify(self, stream_name, token, rows):
+        if stream_name == "events":
+            # We shouldn't get multiple rows per token for events stream, so
+            # we don't need to optimise this for multiple rows.
+            for row in rows:
+                event = yield self.store.get_event(row.event_id)
+                extra_users = ()
+                if event.type == EventTypes.Member:
+                    extra_users = (event.state_key,)
+                max_token = self.store.get_room_max_stream_ordering()
+                self.notifier.on_new_room_event(
+                    event, token, max_token, extra_users
+                )
+        elif stream_name == "push_rules":
+            self.notifier.on_new_event(
+                "push_rules_key", token, users=[row.user_id for row in rows],
+            )
+        elif stream_name in ("account_data", "tag_account_data",):
+            self.notifier.on_new_event(
+                "account_data_key", token, users=[row.user_id for row in rows],
+            )
+        elif stream_name == "receipts":
+            self.notifier.on_new_event(
+                "receipt_key", token, rooms=[row.room_id for row in rows],
+            )
+        elif stream_name == "typing":
+            self.typing_handler.process_replication_rows(token, rows)
+            self.notifier.on_new_event(
+                "typing_key", token, rooms=[row.room_id for row in rows],
+            )
+        elif stream_name == "to_device":
+            entities = [row.entity for row in rows if row.entity.startswith("@")]
+            if entities:
+                self.notifier.on_new_event(
+                    "to_device_key", token, users=entities,
+                )
+        elif stream_name == "device_lists":
+            all_room_ids = set()
+            for row in rows:
+                room_ids = yield self.store.get_rooms_for_user(row.user_id)
+                all_room_ids.update(room_ids)
+            self.notifier.on_new_event(
+                "device_list_key", token, rooms=all_room_ids,
+            )
+        elif stream_name == "presence":
+            yield self.presence_handler.process_replication_rows(token, rows)
+
+
 def start(config_options):
     try:
         config = HomeServerConfig.load_config(
@@ -474,7 +422,7 @@ def start(config_options):
 
     assert config.worker_app == "synapse.app.synchrotron"
 
-    setup_logging(config.worker_log_config, config.worker_log_file)
+    setup_logging(config, use_worker_options=True)
 
     synapse.events.USE_FROZEN_DICTS = config.use_frozen_dicts
 
@@ -493,7 +441,11 @@ def start(config_options):
     ss.start_listening(config.worker_listeners)
 
     def run():
-        with LoggingContext("run"):
+        # make sure that we run the reactor with the sentinel log context,
+        # otherwise other PreserveLoggingContext instances will get confused
+        # and complain when they see the logcontext arbitrarily swapping
+        # between the sentinel and `run` logcontexts.
+        with PreserveLoggingContext():
             logger.info("Running")
             change_resource_limit(config.soft_file_limit)
             if config.gc_thresholds:
@@ -502,7 +454,6 @@ def start(config_options):
 
     def start():
         ss.get_datastore().start_profiling()
-        ss.replicate()
         ss.get_state_handler().start_caching()
 
     reactor.callWhenRunning(start)

synapse/app/synctl.py

@@ -23,14 +23,27 @@ import signal
 import subprocess
 import sys
 import yaml
+import errno
+import time
 
 SYNAPSE = [sys.executable, "-B", "-m", "synapse.app.homeserver"]
 
 GREEN = "\x1b[1;32m"
+YELLOW = "\x1b[1;33m"
 RED = "\x1b[1;31m"
 NORMAL = "\x1b[m"
 
 
+def pid_running(pid):
+    try:
+        os.kill(pid, 0)
+        return True
+    except OSError, err:
+        if err.errno == errno.EPERM:
+            return True
+        return False
+
+
 def write(message, colour=NORMAL, stream=sys.stdout):
     if colour == NORMAL:
         stream.write(message + "\n")
@@ -38,6 +51,11 @@ def write(message, colour=NORMAL, stream=sys.stdout):
         stream.write(colour + message + NORMAL + "\n")
 
 
+def abort(message, colour=RED, stream=sys.stderr):
+    write(message, colour, stream)
+    sys.exit(1)
+
+
 def start(configfile):
     write("Starting ...")
     args = SYNAPSE
@@ -45,7 +63,8 @@ def start(configfile):
 
     try:
         subprocess.check_call(args)
-        write("started synapse.app.homeserver(%r)" % (configfile,), colour=GREEN)
+        write("started synapse.app.homeserver(%r)" %
+              (configfile,), colour=GREEN)
     except subprocess.CalledProcessError as e:
         write(
             "error starting (exit code: %d); see above for logs" % e.returncode,
@@ -76,8 +95,16 @@ def start_worker(app, configfile, worker_configfile):
 def stop(pidfile, app):
     if os.path.exists(pidfile):
         pid = int(open(pidfile).read())
-        os.kill(pid, signal.SIGTERM)
-        write("stopped %s" % (app,), colour=GREEN)
+        try:
+            os.kill(pid, signal.SIGTERM)
+            write("stopped %s" % (app,), colour=GREEN)
+        except OSError, err:
+            if err.errno == errno.ESRCH:
+                write("%s not running" % (app,), colour=YELLOW)
+            elif err.errno == errno.EPERM:
+                abort("Cannot stop %s: Operation not permitted" % (app,))
+            else:
+                abort("Cannot stop %s: Unknown error" % (app,))
 
 
 Worker = collections.namedtuple("Worker", [
@@ -98,7 +125,7 @@ def main():
         "configfile",
         nargs="?",
         default="homeserver.yaml",
-        help="the homeserver config file, defaults to homserver.yaml",
+        help="the homeserver config file, defaults to homeserver.yaml",
     )
     parser.add_argument(
         "-w", "--worker",
@@ -175,7 +202,8 @@ def main():
         worker_app = worker_config["worker_app"]
         worker_pidfile = worker_config["worker_pid_file"]
         worker_daemonize = worker_config["worker_daemonize"]
-        assert worker_daemonize  # TODO print something more user friendly
+        assert worker_daemonize, "In config %r: expected '%s' to be True" % (
+            worker_configfile, "worker_daemonize")
         worker_cache_factor = worker_config.get("synctl_cache_factor")
         workers.append(Worker(
             worker_app, worker_configfile, worker_pidfile, worker_cache_factor,
@@ -190,10 +218,25 @@ def main():
         if start_stop_synapse:
             stop(pidfile, "synapse.app.homeserver")
 
-        # TODO: Wait for synapse to actually shutdown before starting it again
+    # Wait for synapse to actually shutdown before starting it again
+    if action == "restart":
+        running_pids = []
+        if start_stop_synapse and os.path.exists(pidfile):
+            running_pids.append(int(open(pidfile).read()))
+        for worker in workers:
+            if os.path.exists(worker.pidfile):
+                running_pids.append(int(open(worker.pidfile).read()))
+        if len(running_pids) > 0:
+            write("Waiting for process to exit before restarting...")
+            for running_pid in running_pids:
+                while pid_running(running_pid):
+                    time.sleep(0.2)
 
     if action == "start" or action == "restart":
         if start_stop_synapse:
+            # Check if synapse is already running
+            if os.path.exists(pidfile) and pid_running(int(open(pidfile).read())):
+                abort("synapse.app.homeserver already running")
             start(configfile)
 
         for worker in workers:

synapse/appservice/__init__.py

@@ -13,6 +13,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 from synapse.api.constants import EventTypes
+from synapse.util.caches.descriptors import cachedInlineCallbacks
 
 from twisted.internet import defer
 
@@ -124,29 +125,23 @@ class ApplicationService(object):
                     raise ValueError(
                         "Expected bool for 'exclusive' in ns '%s'" % ns
                     )
-                if not isinstance(regex_obj.get("regex"), basestring):
+                regex = regex_obj.get("regex")
+                if isinstance(regex, basestring):
+                    regex_obj["regex"] = re.compile(regex)  # Pre-compile regex
+                else:
                     raise ValueError(
                         "Expected string for 'regex' in ns '%s'" % ns
                     )
         return namespaces
 
-    def _matches_regex(self, test_string, namespace_key, return_obj=False):
-        if not isinstance(test_string, basestring):
-            logger.error(
-                "Expected a string to test regex against, but got %s",
-                test_string
-            )
-            return False
-
+    def _matches_regex(self, test_string, namespace_key):
         for regex_obj in self.namespaces[namespace_key]:
-            if re.match(regex_obj["regex"], test_string):
-                if return_obj:
-                    return regex_obj
-                return True
-        return False
+            if regex_obj["regex"].match(test_string):
+                return regex_obj
+        return None
 
     def _is_exclusive(self, ns_key, test_string):
-        regex_obj = self._matches_regex(test_string, ns_key, return_obj=True)
+        regex_obj = self._matches_regex(test_string, ns_key)
         if regex_obj:
             return regex_obj["exclusive"]
         return False
@@ -166,7 +161,14 @@ class ApplicationService(object):
         if not store:
             defer.returnValue(False)
 
-        member_list = yield store.get_users_in_room(event.room_id)
+        does_match = yield self._matches_user_in_member_list(event.room_id, store)
+        defer.returnValue(does_match)
+
+    @cachedInlineCallbacks(num_args=1, cache_context=True)
+    def _matches_user_in_member_list(self, room_id, store, cache_context):
+        member_list = yield store.get_users_in_room(
+            room_id, on_invalidate=cache_context.invalidate
+        )
 
         # check joined member events
         for user_id in member_list:
@@ -219,10 +221,10 @@ class ApplicationService(object):
         )
 
     def is_interested_in_alias(self, alias):
-        return self._matches_regex(alias, ApplicationService.NS_ALIASES)
+        return bool(self._matches_regex(alias, ApplicationService.NS_ALIASES))
 
     def is_interested_in_room(self, room_id):
-        return self._matches_regex(room_id, ApplicationService.NS_ROOMS)
+        return bool(self._matches_regex(room_id, ApplicationService.NS_ROOMS))
 
     def is_exclusive_user(self, user_id):
         return (

synapse/config/emailconfig.py

@@ -71,6 +71,15 @@ class EmailConfig(Config):
             self.email_riot_base_url = email_config.get(
                 "riot_base_url", None
             )
+            self.email_smtp_user = email_config.get(
+                "smtp_user", None
+            )
+            self.email_smtp_pass = email_config.get(
+                "smtp_pass", None
+            )
+            self.require_transport_security = email_config.get(
+                "require_transport_security", False
+            )
             if "app_name" in email_config:
                 self.email_app_name = email_config["app_name"]
             else:
@@ -91,10 +100,17 @@ class EmailConfig(Config):
         # Defining a custom URL for Riot is only needed if email notifications
         # should contain links to a self-hosted installation of Riot; when set
         # the "app_name" setting is ignored.
+        #
+        # If your SMTP server requires authentication, the optional smtp_user &
+        # smtp_pass variables should be used
+        #
         #email:
         #   enable_notifs: false
         #   smtp_host: "localhost"
         #   smtp_port: 25
+        #   smtp_user: "exampleusername"
+        #   smtp_pass: "examplepassword"
+        #   require_transport_security: False
         #   notif_from: "Your Friendly %(app)s Home Server <noreply@example.com>"
         #   app_name: Matrix
         #   template_dir: res/templates

synapse/config/logger.py

@@ -45,7 +45,6 @@ handlers:
     maxBytes: 104857600
     backupCount: 10
     filters: [context]
-    level: INFO
   console:
     class: logging.StreamHandler
     formatter: precise
@@ -56,6 +55,8 @@ loggers:
         level: INFO
 
     synapse.storage.SQL:
+        # beware: increasing this to DEBUG will make synapse log sensitive
+        # information such as access tokens.
         level: INFO
 
 root:
@@ -68,6 +69,7 @@ class LoggingConfig(Config):
 
     def read_config(self, config):
         self.verbosity = config.get("verbose", 0)
+        self.no_redirect_stdio = config.get("no_redirect_stdio", False)
         self.log_config = self.abspath(config.get("log_config"))
         self.log_file = self.abspath(config.get("log_file"))
 
@@ -77,10 +79,10 @@ class LoggingConfig(Config):
             os.path.join(config_dir_path, server_name + ".log.config")
         )
         return """
-        # Logging verbosity level.
+        # Logging verbosity level. Ignored if log_config is specified.
         verbose: 0
 
-        # File to write logging to
+        # File to write logging to. Ignored if log_config is specified.
         log_file: "%(log_file)s"
 
         # A yaml python logging config file
@@ -90,6 +92,8 @@ class LoggingConfig(Config):
     def read_arguments(self, args):
         if args.verbose is not None:
             self.verbosity = args.verbose
+        if args.no_redirect_stdio is not None:
+            self.no_redirect_stdio = args.no_redirect_stdio
         if args.log_config is not None:
             self.log_config = args.log_config
         if args.log_file is not None:
@@ -99,16 +103,22 @@ class LoggingConfig(Config):
         logging_group = parser.add_argument_group("logging")
         logging_group.add_argument(
             '-v', '--verbose', dest="verbose", action='count',
-            help="The verbosity level."
+            help="The verbosity level. Specify multiple times to increase "
+            "verbosity. (Ignored if --log-config is specified.)"
         )
         logging_group.add_argument(
             '-f', '--log-file', dest="log_file",
-            help="File to log to."
+            help="File to log to. (Ignored if --log-config is specified.)"
         )
         logging_group.add_argument(
             '--log-config', dest="log_config", default=None,
             help="Python logging config file"
         )
+        logging_group.add_argument(
+            '-n', '--no-redirect-stdio',
+            action='store_true', default=None,
+            help="Do not redirect stdout/stderr to the log"
+        )
 
     def generate_files(self, config):
         log_config = config.get("log_config")
@@ -118,11 +128,22 @@ class LoggingConfig(Config):
                     DEFAULT_LOG_CONFIG.substitute(log_file=config["log_file"])
                 )
 
-    def setup_logging(self):
-        setup_logging(self.log_config, self.log_file, self.verbosity)
 
+def setup_logging(config, use_worker_options=False):
+    """ Set up python logging
+
+    Args:
+        config (LoggingConfig | synapse.config.workers.WorkerConfig):
+            configuration data
+
+        use_worker_options (bool): True to use 'worker_log_config' and
+            'worker_log_file' options instead of 'log_config' and 'log_file'.
+    """
+    log_config = (config.worker_log_config if use_worker_options
+                  else config.log_config)
+    log_file = (config.worker_log_file if use_worker_options
+                else config.log_file)
 
-def setup_logging(log_config=None, log_file=None, verbosity=None):
     log_format = (
         "%(asctime)s - %(name)s - %(lineno)d - %(levelname)s - %(request)s"
         " - %(message)s"
@@ -131,9 +152,9 @@ def setup_logging(log_config=None, log_file=None, verbosity=None):
 
         level = logging.INFO
         level_for_storage = logging.INFO
-        if verbosity:
+        if config.verbosity:
             level = logging.DEBUG
-            if verbosity > 1:
+            if config.verbosity > 1:
                 level_for_storage = logging.DEBUG
 
         # FIXME: we need a logging.WARN for a -q quiet option
@@ -153,14 +174,6 @@ def setup_logging(log_config=None, log_file=None, verbosity=None):
                 logger.info("Closing log file due to SIGHUP")
                 handler.doRollover()
                 logger.info("Opened new log file due to SIGHUP")
-
-            # TODO(paul): obviously this is a terrible mechanism for
-            #   stealing SIGHUP, because it means no other part of synapse
-            #   can use it instead. If we want to catch SIGHUP anywhere
-            #   else as well, I'd suggest we find a nicer way to broadcast
-            #   it around.
-            if getattr(signal, "SIGHUP"):
-                signal.signal(signal.SIGHUP, sighup)
         else:
             handler = logging.StreamHandler()
         handler.setFormatter(formatter)
@@ -169,8 +182,25 @@ def setup_logging(log_config=None, log_file=None, verbosity=None):
 
         logger.addHandler(handler)
     else:
-        with open(log_config, 'r') as f:
-            logging.config.dictConfig(yaml.load(f))
+        def load_log_config():
+            with open(log_config, 'r') as f:
+                logging.config.dictConfig(yaml.load(f))
+
+        def sighup(signum, stack):
+            # it might be better to use a file watcher or something for this.
+            logging.info("Reloading log config from %s due to SIGHUP",
+                         log_config)
+            load_log_config()
+
+        load_log_config()
+
+    # TODO(paul): obviously this is a terrible mechanism for
+    #   stealing SIGHUP, because it means no other part of synapse
+    #   can use it instead. If we want to catch SIGHUP anywhere
+    #   else as well, I'd suggest we find a nicer way to broadcast
+    #   it around.
+    if getattr(signal, "SIGHUP"):
+        signal.signal(signal.SIGHUP, sighup)
 
     # It's critical to point twisted's internal logging somewhere, otherwise it
     # stacks up and leaks kup to 64K object;
@@ -183,4 +213,7 @@ def setup_logging(log_config=None, log_file=None, verbosity=None):
     #
     # However this may not be too much of a problem if we are just writing to a file.
     observer = STDLibLogObserver()
-    globalLogBeginner.beginLoggingTo([observer])
+    globalLogBeginner.beginLoggingTo(
+        [observer],
+        redirectStandardIO=not config.no_redirect_stdio,
+    )

synapse/config/registration.py

@@ -69,6 +69,7 @@ class RegistrationConfig(Config):
         trusted_third_party_id_servers:
             - matrix.org
             - vector.im
+            - riot.im
         """ % locals()
 
     def add_arguments(self, parser):

synapse/config/server.py

@@ -35,6 +35,8 @@ class ServerConfig(Config):
         # "disable" federation
         self.send_federation = config.get("send_federation", True)
 
+        self.filter_timeline_limit = config.get("filter_timeline_limit", -1)
+
         if self.public_baseurl is not None:
             if self.public_baseurl[-1] != '/':
                 self.public_baseurl += '/'
@@ -144,6 +146,12 @@ class ServerConfig(Config):
         # Whether to serve a web client from the HTTP/HTTPS root resource.
         web_client: True
 
+        # The root directory to server for the above web client.
+        # If left undefined, synapse will serve the matrix-angular-sdk web client.
+        # Make sure matrix-angular-sdk is installed with pip if web_client is True
+        # and web_client_location is undefined
+        # web_client_location: "/path/to/web/root"
+
         # The public-facing base URL for the client API (not including _matrix/...)
         # public_baseurl: https://example.com:8448/
 
@@ -155,6 +163,10 @@ class ServerConfig(Config):
         # The GC threshold parameters to pass to `gc.set_threshold`, if defined
         # gc_thresholds: [700, 10, 10]
 
+        # Set the limit on the returned events in the timeline in the get
+        # and sync operations. The default value is -1, means no upper limit.
+        # filter_timeline_limit: 5000
+
         # List of ports that Synapse should listen on, their purpose and their
         # configuration.
         listeners:

synapse/config/tls.py

@@ -95,7 +95,7 @@ class TlsConfig(Config):
         # make HTTPS requests to this server will check that the TLS
         # certificates returned by this server match one of the fingerprints.
         #
-        # Synapse automatically adds its the fingerprint of its own certificate
+        # Synapse automatically adds the fingerprint of its own certificate
         # to the list. So if federation traffic is handle directly by synapse
         # then no modification to the list is required.
         #

synapse/config/voip.py

@@ -23,6 +23,7 @@ class VoipConfig(Config):
         self.turn_username = config.get("turn_username")
         self.turn_password = config.get("turn_password")
         self.turn_user_lifetime = self.parse_duration(config["turn_user_lifetime"])
+        self.turn_allow_guests = config.get("turn_allow_guests", True)
 
     def default_config(self, **kwargs):
         return """\
@@ -41,4 +42,11 @@ class VoipConfig(Config):
 
         # How long generated TURN credentials last
         turn_user_lifetime: "1h"
+
+        # Whether guests should be allowed to use the TURN server.
+        # This defaults to True, otherwise VoIP will be unreliable for guests.
+        # However, it does introduce a slight security risk as it allows users to
+        # connect to arbitrary endpoints without having first signed up for a
+        # valid account (e.g. by passing a CAPTCHA).
+        turn_allow_guests: True
         """

synapse/config/workers.py

@@ -28,7 +28,9 @@ class WorkerConfig(Config):
         self.worker_pid_file = config.get("worker_pid_file")
         self.worker_log_file = config.get("worker_log_file")
         self.worker_log_config = config.get("worker_log_config")
-        self.worker_replication_url = config.get("worker_replication_url")
+        self.worker_replication_host = config.get("worker_replication_host", None)
+        self.worker_replication_port = config.get("worker_replication_port", None)
+        self.worker_name = config.get("worker_name", self.worker_app)
 
         if self.worker_listeners:
             for listener in self.worker_listeners:

synapse/crypto/keyring.py

@@ -15,7 +15,6 @@
 
 from synapse.crypto.keyclient import fetch_server_key
 from synapse.api.errors import SynapseError, Codes
-from synapse.util.retryutils import get_retry_limiter
 from synapse.util import unwrapFirstError
 from synapse.util.async import ObservableDeferred
 from synapse.util.logcontext import (
@@ -96,10 +95,11 @@ class Keyring(object):
         verify_requests = []
 
         for server_name, json_object in server_and_json:
-            logger.debug("Verifying for %s", server_name)
 
             key_ids = signature_ids(json_object, server_name)
             if not key_ids:
+                logger.warn("Request from %s: no supported signature keys",
+                            server_name)
                 deferred = defer.fail(SynapseError(
                     400,
                     "Not signed with a supported algorithm",
@@ -108,6 +108,9 @@ class Keyring(object):
             else:
                 deferred = defer.Deferred()
 
+            logger.debug("Verifying for %s with key_ids %s",
+                         server_name, key_ids)
+
             verify_request = VerifyKeyRequest(
                 server_name, key_ids, json_object, deferred
             )
@@ -142,6 +145,9 @@ class Keyring(object):
 
             json_object = verify_request.json_object
 
+            logger.debug("Got key %s %s:%s for server %s, verifying" % (
+                key_id, verify_key.alg, verify_key.version, server_name,
+            ))
             try:
                 verify_signed_json(json_object, server_name, verify_key)
             except:
@@ -231,8 +237,14 @@ class Keyring(object):
             d.addBoth(rm, server_name)
 
     def get_server_verify_keys(self, verify_requests):
-        """Takes a dict of KeyGroups and tries to find at least one key for
-        each group.
+        """Tries to find at least one key for each verify request
+
+        For each verify_request, verify_request.deferred is called back with
+        params (server_name, key_id, VerifyKey) if a key is found, or errbacked
+        with a SynapseError if none of the keys are found.
+
+        Args:
+            verify_requests (list[VerifyKeyRequest]): list of verify requests
         """
 
         # These are functions that produce keys given a list of key ids
@@ -245,8 +257,11 @@ class Keyring(object):
         @defer.inlineCallbacks
         def do_iterations():
             with Measure(self.clock, "get_server_verify_keys"):
+                # dict[str, dict[str, VerifyKey]]: results so far.
+                # map server_name -> key_id -> VerifyKey
                 merged_results = {}
 
+                # dict[str, set(str)]: keys to fetch for each server
                 missing_keys = {}
                 for verify_request in verify_requests:
                     missing_keys.setdefault(verify_request.server_name, set()).update(
@@ -308,6 +323,16 @@ class Keyring(object):
 
     @defer.inlineCallbacks
     def get_keys_from_store(self, server_name_and_key_ids):
+        """
+
+        Args:
+            server_name_and_key_ids (list[(str, iterable[str])]):
+                list of (server_name, iterable[key_id]) tuples to fetch keys for
+
+        Returns:
+            Deferred: resolves to dict[str, dict[str, VerifyKey]]: map from
+                server_name -> key_id -> VerifyKey
+        """
         res = yield preserve_context_over_deferred(defer.gatherResults(
             [
                 preserve_fn(self.store.get_server_verify_keys)(
@@ -356,30 +381,24 @@ class Keyring(object):
     def get_keys_from_server(self, server_name_and_key_ids):
         @defer.inlineCallbacks
         def get_key(server_name, key_ids):
-            limiter = yield get_retry_limiter(
-                server_name,
-                self.clock,
-                self.store,
-            )
-            with limiter:
-                keys = None
-                try:
-                    keys = yield self.get_server_verify_key_v2_direct(
-                        server_name, key_ids
-                    )
-                except Exception as e:
-                    logger.info(
-                        "Unable to get key %r for %r directly: %s %s",
-                        key_ids, server_name,
-                        type(e).__name__, str(e.message),
-                    )
+            keys = None
+            try:
+                keys = yield self.get_server_verify_key_v2_direct(
+                    server_name, key_ids
+                )
+            except Exception as e:
+                logger.info(
+                    "Unable to get key %r for %r directly: %s %s",
+                    key_ids, server_name,
+                    type(e).__name__, str(e.message),
+                )
 
-                if not keys:
-                    keys = yield self.get_server_verify_key_v1_direct(
-                        server_name, key_ids
-                    )
+            if not keys:
+                keys = yield self.get_server_verify_key_v1_direct(
+                    server_name, key_ids
+                )
 
-                    keys = {server_name: keys}
+                keys = {server_name: keys}
 
             defer.returnValue(keys)
 

synapse/events/snapshot.py

@@ -15,6 +15,32 @@
 
 
 class EventContext(object):
+    """
+    Attributes:
+        current_state_ids (dict[(str, str), str]):
+            The current state map including the current event.
+            (type, state_key) -> event_id
+
+        prev_state_ids (dict[(str, str), str]):
+            The current state map excluding the current event.
+            (type, state_key) -> event_id
+
+        state_group (int): state group id
+        rejected (bool|str): A rejection reason if the event was rejected, else
+            False
+
+        push_actions (list[(str, list[object])]): list of (user_id, actions)
+            tuples
+
+        prev_group (int): Previously persisted state group. ``None`` for an
+            outlier.
+        delta_ids (dict[(str, str), str]): Delta from ``prev_group``.
+            (type, state_key) -> event_id. ``None`` for an outlier.
+
+        prev_state_events (?): XXX: is this ever set to anything other than
+            the empty list?
+    """
+
     __slots__ = [
         "current_state_ids",
         "prev_state_ids",
@@ -24,6 +50,7 @@ class EventContext(object):
         "prev_group",
         "delta_ids",
         "prev_state_events",
+        "app_service",
     ]
 
     def __init__(self):
@@ -42,3 +69,5 @@ class EventContext(object):
         self.delta_ids = None
 
         self.prev_state_events = None
+
+        self.app_service = None

synapse/events/utils.py

@@ -225,7 +225,22 @@ def format_event_for_client_v2_without_room_id(d):
 
 def serialize_event(e, time_now_ms, as_client_event=True,
                     event_format=format_event_for_client_v1,
-                    token_id=None, only_event_fields=None):
+                    token_id=None, only_event_fields=None, is_invite=False):
+    """Serialize event for clients
+
+    Args:
+        e (EventBase)
+        time_now_ms (int)
+        as_client_event (bool)
+        event_format
+        token_id
+        only_event_fields
+        is_invite (bool): Whether this is an invite that is being sent to the
+            invitee
+
+    Returns:
+        dict
+    """
     # FIXME(erikj): To handle the case of presence events and the like
     if not isinstance(e, EventBase):
         return e
@@ -251,6 +266,12 @@ def serialize_event(e, time_now_ms, as_client_event=True,
             if txn_id is not None:
                 d["unsigned"]["transaction_id"] = txn_id
 
+    # If this is an invite for somebody else, then we don't care about the
+    # invite_room_state as that's meant solely for the invitee. Other clients
+    # will already have the state since they're in the room.
+    if not is_invite:
+        d["unsigned"].pop("invite_room_state", None)
+
     if as_client_event:
         d = event_format(d)
 

synapse/federation/federation_client.py

@@ -29,7 +29,7 @@ from synapse.util.logcontext import preserve_fn, preserve_context_over_deferred
 from synapse.events import FrozenEvent, builder
 import synapse.metrics
 
-from synapse.util.retryutils import get_retry_limiter, NotRetryingDestination
+from synapse.util.retryutils import NotRetryingDestination
 
 import copy
 import itertools
@@ -88,7 +88,7 @@ class FederationClient(FederationBase):
 
     @log_function
     def make_query(self, destination, query_type, args,
-                   retry_on_dns_fail=False):
+                   retry_on_dns_fail=False, ignore_backoff=False):
         """Sends a federation Query to a remote homeserver of the given type
         and arguments.
 
@@ -98,6 +98,8 @@ class FederationClient(FederationBase):
                 handler name used in register_query_handler().
             args (dict): Mapping of strings to strings containing the details
                 of the query request.
+            ignore_backoff (bool): true to ignore the historical backoff data
+                and try the request anyway.
 
         Returns:
             a Deferred which will eventually yield a JSON object from the
@@ -106,7 +108,8 @@ class FederationClient(FederationBase):
         sent_queries_counter.inc(query_type)
 
         return self.transport_layer.make_query(
-            destination, query_type, args, retry_on_dns_fail=retry_on_dns_fail
+            destination, query_type, args, retry_on_dns_fail=retry_on_dns_fail,
+            ignore_backoff=ignore_backoff,
         )
 
     @log_function
@@ -206,8 +209,7 @@ class FederationClient(FederationBase):
 
         Args:
             destinations (list): Which home servers to query
-            pdu_origin (str): The home server that originally sent the pdu.
-            event_id (str)
+            event_id (str): event to fetch
             outlier (bool): Indicates whether the PDU is an `outlier`, i.e. if
                 it's from an arbitary point in the context as opposed to part
                 of the current block of PDUs. Defaults to `False`
@@ -235,31 +237,24 @@ class FederationClient(FederationBase):
                 continue
 
             try:
-                limiter = yield get_retry_limiter(
-                    destination,
-                    self._clock,
-                    self.store,
+                transaction_data = yield self.transport_layer.get_event(
+                    destination, event_id, timeout=timeout,
                 )
 
-                with limiter:
-                    transaction_data = yield self.transport_layer.get_event(
-                        destination, event_id, timeout=timeout,
-                    )
+                logger.debug("transaction_data %r", transaction_data)
 
-                    logger.debug("transaction_data %r", transaction_data)
+                pdu_list = [
+                    self.event_from_pdu_json(p, outlier=outlier)
+                    for p in transaction_data["pdus"]
+                ]
 
-                    pdu_list = [
-                        self.event_from_pdu_json(p, outlier=outlier)
-                        for p in transaction_data["pdus"]
-                    ]
+                if pdu_list and pdu_list[0]:
+                    pdu = pdu_list[0]
 
-                    if pdu_list and pdu_list[0]:
-                        pdu = pdu_list[0]
+                    # Check signatures are correct.
+                    signed_pdu = yield self._check_sigs_and_hashes([pdu])[0]
 
-                        # Check signatures are correct.
-                        signed_pdu = yield self._check_sigs_and_hashes([pdu])[0]
-
-                        break
+                    break
 
                 pdu_attempts[destination] = now
 
@@ -479,8 +474,13 @@ class FederationClient(FederationBase):
             content (object): Any additional data to put into the content field
                 of the event.
         Return:
-            A tuple of (origin (str), event (object)) where origin is the remote
-            homeserver which generated the event.
+            Deferred: resolves to a tuple of (origin (str), event (object))
+            where origin is the remote homeserver which generated the event.
+
+            Fails with a ``CodeMessageException`` if the chosen remote server
+            returns a 300/400 code.
+
+            Fails with a ``RuntimeError`` if no servers were reachable.
         """
         valid_memberships = {Membership.JOIN, Membership.LEAVE}
         if membership not in valid_memberships:
@@ -533,6 +533,27 @@ class FederationClient(FederationBase):
 
     @defer.inlineCallbacks
     def send_join(self, destinations, pdu):
+        """Sends a join event to one of a list of homeservers.
+
+        Doing so will cause the remote server to add the event to the graph,
+        and send the event out to the rest of the federation.
+
+        Args:
+            destinations (str): Candidate homeservers which are probably
+                participating in the room.
+            pdu (BaseEvent): event to be sent
+
+        Return:
+            Deferred: resolves to a dict with members ``origin`` (a string
+            giving the serer the event was sent to, ``state`` (?) and
+            ``auth_chain``.
+
+            Fails with a ``CodeMessageException`` if the chosen remote server
+            returns a 300/400 code.
+
+            Fails with a ``RuntimeError`` if no servers were reachable.
+        """
+
         for destination in destinations:
             if destination == self.server_name:
                 continue
@@ -640,6 +661,26 @@ class FederationClient(FederationBase):
 
     @defer.inlineCallbacks
     def send_leave(self, destinations, pdu):
+        """Sends a leave event to one of a list of homeservers.
+
+        Doing so will cause the remote server to add the event to the graph,
+        and send the event out to the rest of the federation.
+
+        This is mostly useful to reject received invites.
+
+        Args:
+            destinations (str): Candidate homeservers which are probably
+                participating in the room.
+            pdu (BaseEvent): event to be sent
+
+        Return:
+            Deferred: resolves to None.
+
+            Fails with a ``CodeMessageException`` if the chosen remote server
+            returns a non-200 code.
+
+            Fails with a ``RuntimeError`` if no servers were reachable.
+        """
         for destination in destinations:
             if destination == self.server_name:
                 continue

synapse/federation/federation_server.py

@@ -52,7 +52,6 @@ class FederationServer(FederationBase):
 
         self.auth = hs.get_auth()
 
-        self._room_pdu_linearizer = Linearizer("fed_room_pdu")
         self._server_linearizer = Linearizer("fed_server")
 
         # We cache responses to state queries, as they take a while and often
@@ -147,11 +146,15 @@ class FederationServer(FederationBase):
             # check that it's actually being sent from a valid destination to
             # workaround bug #1753 in 0.18.5 and 0.18.6
             if transaction.origin != get_domain_from_id(pdu.event_id):
+                # We continue to accept join events from any server; this is
+                # necessary for the federation join dance to work correctly.
+                # (When we join over federation, the "helper" server is
+                # responsible for sending out the join event, rather than the
+                # origin. See bug #1893).
                 if not (
                     pdu.type == 'm.room.member' and
                     pdu.content and
-                    pdu.content.get("membership", None) == 'join' and
-                    self.hs.is_mine_id(pdu.state_key)
+                    pdu.content.get("membership", None) == 'join'
                 ):
                     logger.info(
                         "Discarding PDU %s from invalid origin %s",
@@ -165,7 +168,7 @@ class FederationServer(FederationBase):
                     )
 
             try:
-                yield self._handle_new_pdu(transaction.origin, pdu)
+                yield self._handle_received_pdu(transaction.origin, pdu)
                 results.append({})
             except FederationError as e:
                 self.send_failure(e, transaction.origin)
@@ -437,6 +440,16 @@ class FederationServer(FederationBase):
                         key_id: json.loads(json_bytes)
                     }
 
+        logger.info(
+            "Claimed one-time-keys: %s",
+            ",".join((
+                "%s for %s:%s" % (key_id, user_id, device_id)
+                for user_id, user_keys in json_result.iteritems()
+                for device_id, device_keys in user_keys.iteritems()
+                for key_id, _ in device_keys.iteritems()
+            )),
+        )
+
         defer.returnValue({"one_time_keys": json_result})
 
     @defer.inlineCallbacks
@@ -497,27 +510,16 @@ class FederationServer(FederationBase):
         )
 
     @defer.inlineCallbacks
-    @log_function
-    def _handle_new_pdu(self, origin, pdu, get_missing=True):
+    def _handle_received_pdu(self, origin, pdu):
+        """ Process a PDU received in a federation /send/ transaction.
 
-        # We reprocess pdus when we have seen them only as outliers
-        existing = yield self._get_persisted_pdu(
-            origin, pdu.event_id, do_auth=False
-        )
-
-        # FIXME: Currently we fetch an event again when we already have it
-        # if it has been marked as an outlier.
-
-        already_seen = (
-            existing and (
-                not existing.internal_metadata.is_outlier()
-                or pdu.internal_metadata.is_outlier()
-            )
-        )
-        if already_seen:
-            logger.debug("Already seen pdu %s", pdu.event_id)
-            return
+        Args:
+            origin (str): server which sent the pdu
+            pdu (FrozenEvent): received pdu
 
+        Returns (Deferred): completes with None
+        Raises: FederationError if the signatures / hash do not match
+    """
         # Check signature.
         try:
             pdu = yield self._check_sigs_and_hash(pdu)
@@ -529,143 +531,7 @@ class FederationServer(FederationBase):
                 affected=pdu.event_id,
             )
 
-        state = None
-
-        auth_chain = []
-
-        have_seen = yield self.store.have_events(
-            [ev for ev, _ in pdu.prev_events]
-        )
-
-        fetch_state = False
-
-        # Get missing pdus if necessary.
-        if not pdu.internal_metadata.is_outlier():
-            # We only backfill backwards to the min depth.
-            min_depth = yield self.handler.get_min_depth_for_context(
-                pdu.room_id
-            )
-
-            logger.debug(
-                "_handle_new_pdu min_depth for %s: %d",
-                pdu.room_id, min_depth
-            )
-
-            prevs = {e_id for e_id, _ in pdu.prev_events}
-            seen = set(have_seen.keys())
-
-            if min_depth and pdu.depth < min_depth:
-                # This is so that we don't notify the user about this
-                # message, to work around the fact that some events will
-                # reference really really old events we really don't want to
-                # send to the clients.
-                pdu.internal_metadata.outlier = True
-            elif min_depth and pdu.depth > min_depth:
-                if get_missing and prevs - seen:
-                    # If we're missing stuff, ensure we only fetch stuff one
-                    # at a time.
-                    logger.info(
-                        "Acquiring lock for room %r to fetch %d missing events: %r...",
-                        pdu.room_id, len(prevs - seen), list(prevs - seen)[:5],
-                    )
-                    with (yield self._room_pdu_linearizer.queue(pdu.room_id)):
-                        logger.info(
-                            "Acquired lock for room %r to fetch %d missing events",
-                            pdu.room_id, len(prevs - seen),
-                        )
-
-                        # We recalculate seen, since it may have changed.
-                        have_seen = yield self.store.have_events(prevs)
-                        seen = set(have_seen.keys())
-
-                        if prevs - seen:
-                            latest = yield self.store.get_latest_event_ids_in_room(
-                                pdu.room_id
-                            )
-
-                            # We add the prev events that we have seen to the latest
-                            # list to ensure the remote server doesn't give them to us
-                            latest = set(latest)
-                            latest |= seen
-
-                            logger.info(
-                                "Missing %d events for room %r: %r...",
-                                len(prevs - seen), pdu.room_id, list(prevs - seen)[:5]
-                            )
-
-                            # XXX: we set timeout to 10s to help workaround
-                            # https://github.com/matrix-org/synapse/issues/1733.
-                            # The reason is to avoid holding the linearizer lock
-                            # whilst processing inbound /send transactions, causing
-                            # FDs to stack up and block other inbound transactions
-                            # which empirically can currently take up to 30 minutes.
-                            #
-                            # N.B. this explicitly disables retry attempts.
-                            #
-                            # N.B. this also increases our chances of falling back to
-                            # fetching fresh state for the room if the missing event
-                            # can't be found, which slightly reduces our security.
-                            # it may also increase our DAG extremity count for the room,
-                            # causing additional state resolution?  See #1760.
-                            # However, fetching state doesn't hold the linearizer lock
-                            # apparently.
-                            #
-                            # see https://github.com/matrix-org/synapse/pull/1744
-
-                            missing_events = yield self.get_missing_events(
-                                origin,
-                                pdu.room_id,
-                                earliest_events_ids=list(latest),
-                                latest_events=[pdu],
-                                limit=10,
-                                min_depth=min_depth,
-                                timeout=10000,
-                            )
-
-                            # We want to sort these by depth so we process them and
-                            # tell clients about them in order.
-                            missing_events.sort(key=lambda x: x.depth)
-
-                            for e in missing_events:
-                                yield self._handle_new_pdu(
-                                    origin,
-                                    e,
-                                    get_missing=False
-                                )
-
-                            have_seen = yield self.store.have_events(
-                                [ev for ev, _ in pdu.prev_events]
-                            )
-
-            prevs = {e_id for e_id, _ in pdu.prev_events}
-            seen = set(have_seen.keys())
-            if prevs - seen:
-                logger.info(
-                    "Still missing %d events for room %r: %r...",
-                    len(prevs - seen), pdu.room_id, list(prevs - seen)[:5]
-                )
-                fetch_state = True
-
-        if fetch_state:
-            # We need to get the state at this event, since we haven't
-            # processed all the prev events.
-            logger.debug(
-                "_handle_new_pdu getting state for %s",
-                pdu.room_id
-            )
-            try:
-                state, auth_chain = yield self.get_state_for_room(
-                    origin, pdu.room_id, pdu.event_id,
-                )
-            except:
-                logger.exception("Failed to get state for event: %s", pdu.event_id)
-
-        yield self.handler.on_receive_pdu(
-            origin,
-            pdu,
-            state=state,
-            auth_chain=auth_chain,
-        )
+        yield self.handler.on_receive_pdu(origin, pdu, get_missing=True)
 
     def __str__(self):
         return "<ReplicationLayer(%s)>" % self.server_name

synapse/federation/send_queue.py

@@ -31,41 +31,41 @@ Events are replicated via a separate events stream.
 
 from .units import Edu
 
+from synapse.storage.presence import UserPresenceState
 from synapse.util.metrics import Measure
 import synapse.metrics
 
 from blist import sorteddict
-import ujson
+from collections import namedtuple
+
+import logging
+
+logger = logging.getLogger(__name__)
 
 
 metrics = synapse.metrics.get_metrics_for(__name__)
 
 
-PRESENCE_TYPE = "p"
-KEYED_EDU_TYPE = "k"
-EDU_TYPE = "e"
-FAILURE_TYPE = "f"
-DEVICE_MESSAGE_TYPE = "d"
-
-
 class FederationRemoteSendQueue(object):
     """A drop in replacement for TransactionQueue"""
 
     def __init__(self, hs):
         self.server_name = hs.hostname
         self.clock = hs.get_clock()
+        self.notifier = hs.get_notifier()
+        self.is_mine_id = hs.is_mine_id
 
-        self.presence_map = {}
-        self.presence_changed = sorteddict()
+        self.presence_map = {}  # Pending presence map user_id -> UserPresenceState
+        self.presence_changed = sorteddict()  # Stream position -> user_id
 
-        self.keyed_edu = {}
-        self.keyed_edu_changed = sorteddict()
+        self.keyed_edu = {}  # (destination, key) -> EDU
+        self.keyed_edu_changed = sorteddict()  # stream position -> (destination, key)
 
-        self.edus = sorteddict()
+        self.edus = sorteddict()  # stream position -> Edu
 
-        self.failures = sorteddict()
+        self.failures = sorteddict()  # stream position -> (destination, Failure)
 
-        self.device_messages = sorteddict()
+        self.device_messages = sorteddict()  # stream position -> destination
 
         self.pos = 1
         self.pos_time = sorteddict()
@@ -121,7 +121,9 @@ class FederationRemoteSendQueue(object):
                 del self.presence_changed[key]
 
             user_ids = set(
-                user_id for uids in self.presence_changed.values() for _, user_id in uids
+                user_id
+                for uids in self.presence_changed.itervalues()
+                for user_id in uids
             )
 
             to_del = [
@@ -186,37 +188,50 @@ class FederationRemoteSendQueue(object):
         else:
             self.edus[pos] = edu
 
-    def send_presence(self, destination, states):
-        """As per TransactionQueue"""
+        self.notifier.on_new_replication_data()
+
+    def send_presence(self, states):
+        """As per TransactionQueue
+
+        Args:
+            states (list(UserPresenceState))
+        """
         pos = self._next_pos()
 
-        self.presence_map.update({
-            state.user_id: state
-            for state in states
-        })
+        # We only want to send presence for our own users, so lets always just
+        # filter here just in case.
+        local_states = filter(lambda s: self.is_mine_id(s.user_id), states)
 
-        self.presence_changed[pos] = [
-            (destination, state.user_id) for state in states
-        ]
+        self.presence_map.update({state.user_id: state for state in local_states})
+        self.presence_changed[pos] = [state.user_id for state in local_states]
+
+        self.notifier.on_new_replication_data()
 
     def send_failure(self, failure, destination):
         """As per TransactionQueue"""
         pos = self._next_pos()
 
         self.failures[pos] = (destination, str(failure))
+        self.notifier.on_new_replication_data()
 
     def send_device_messages(self, destination):
         """As per TransactionQueue"""
         pos = self._next_pos()
         self.device_messages[pos] = destination
+        self.notifier.on_new_replication_data()
 
     def get_current_token(self):
         return self.pos - 1
 
-    def get_replication_rows(self, token, limit, federation_ack=None):
-        """
+    def federation_ack(self, token):
+        self._clear_queue_before_pos(token)
+
+    def get_replication_rows(self, from_token, to_token, limit, federation_ack=None):
+        """Get rows to be sent over federation between the two tokens
+
         Args:
-            token (int)
+            from_token (int)
+            to_token(int)
             limit (int)
             federation_ack (int): Optional. The position where the worker is
                 explicitly acknowledged it has handled. Allows us to drop
@@ -225,9 +240,11 @@ class FederationRemoteSendQueue(object):
         # TODO: Handle limit.
 
         # To handle restarts where we wrap around
-        if token > self.pos:
-            token = -1
+        if from_token > self.pos:
+            from_token = -1
 
+        # list of tuple(int, BaseFederationRow), where the first is the position
+        # of the federation stream.
         rows = []
 
         # There should be only one reader, so lets delete everything its
@@ -237,62 +254,295 @@ class FederationRemoteSendQueue(object):
 
         # Fetch changed presence
         keys = self.presence_changed.keys()
-        i = keys.bisect_right(token)
-        dest_user_ids = set(
-            (pos, dest_user_id)
-            for pos in keys[i:]
-            for dest_user_id in self.presence_changed[pos]
-        )
+        i = keys.bisect_right(from_token)
+        j = keys.bisect_right(to_token) + 1
+        dest_user_ids = [
+            (pos, user_id)
+            for pos in keys[i:j]
+            for user_id in self.presence_changed[pos]
+        ]
 
-        for (key, (dest, user_id)) in dest_user_ids:
-            rows.append((key, PRESENCE_TYPE, ujson.dumps({
-                "destination": dest,
-                "state": self.presence_map[user_id].as_dict(),
-            })))
+        for (key, user_id) in dest_user_ids:
+            rows.append((key, PresenceRow(
+                state=self.presence_map[user_id],
+            )))
 
         # Fetch changes keyed edus
         keys = self.keyed_edu_changed.keys()
-        i = keys.bisect_right(token)
-        keyed_edus = set((k, self.keyed_edu_changed[k]) for k in keys[i:])
+        i = keys.bisect_right(from_token)
+        j = keys.bisect_right(to_token) + 1
+        # We purposefully clobber based on the key here, python dict comprehensions
+        # always use the last value, so this will correctly point to the last
+        # stream position.
+        keyed_edus = {self.keyed_edu_changed[k]: k for k in keys[i:j]}
 
-        for (pos, (destination, edu_key)) in keyed_edus:
-            rows.append(
-                (pos, KEYED_EDU_TYPE, ujson.dumps({
-                    "key": edu_key,
-                    "edu": self.keyed_edu[(destination, edu_key)].get_internal_dict(),
-                }))
-            )
+        for ((destination, edu_key), pos) in keyed_edus.iteritems():
+            rows.append((pos, KeyedEduRow(
+                key=edu_key,
+                edu=self.keyed_edu[(destination, edu_key)],
+            )))
 
         # Fetch changed edus
         keys = self.edus.keys()
-        i = keys.bisect_right(token)
-        edus = set((k, self.edus[k]) for k in keys[i:])
+        i = keys.bisect_right(from_token)
+        j = keys.bisect_right(to_token) + 1
+        edus = ((k, self.edus[k]) for k in keys[i:j])
 
         for (pos, edu) in edus:
-            rows.append((pos, EDU_TYPE, ujson.dumps(edu.get_internal_dict())))
+            rows.append((pos, EduRow(edu)))
 
         # Fetch changed failures
         keys = self.failures.keys()
-        i = keys.bisect_right(token)
-        failures = set((k, self.failures[k]) for k in keys[i:])
+        i = keys.bisect_right(from_token)
+        j = keys.bisect_right(to_token) + 1
+        failures = ((k, self.failures[k]) for k in keys[i:j])
 
         for (pos, (destination, failure)) in failures:
-            rows.append((pos, FAILURE_TYPE, ujson.dumps({
-                "destination": destination,
-                "failure": failure,
-            })))
+            rows.append((pos, FailureRow(
+                destination=destination,
+                failure=failure,
+            )))
 
         # Fetch changed device messages
         keys = self.device_messages.keys()
-        i = keys.bisect_right(token)
-        device_messages = set((k, self.device_messages[k]) for k in keys[i:])
+        i = keys.bisect_right(from_token)
+        j = keys.bisect_right(to_token) + 1
+        device_messages = {self.device_messages[k]: k for k in keys[i:j]}
 
-        for (pos, destination) in device_messages:
-            rows.append((pos, DEVICE_MESSAGE_TYPE, ujson.dumps({
-                "destination": destination,
-            })))
+        for (destination, pos) in device_messages.iteritems():
+            rows.append((pos, DeviceRow(
+                destination=destination,
+            )))
 
         # Sort rows based on pos
         rows.sort()
 
-        return rows
+        return [(pos, row.TypeId, row.to_data()) for pos, row in rows]
+
+
+class BaseFederationRow(object):
+    """Base class for rows to be sent in the federation stream.
+
+    Specifies how to identify, serialize and deserialize the different types.
+    """
+
+    TypeId = None  # Unique string that ids the type. Must be overriden in sub classes.
+
+    @staticmethod
+    def from_data(data):
+        """Parse the data from the federation stream into a row.
+
+        Args:
+            data: The value of ``data`` from FederationStreamRow.data, type
+                depends on the type of stream
+        """
+        raise NotImplementedError()
+
+    def to_data(self):
+        """Serialize this row to be sent over the federation stream.
+
+        Returns:
+            The value to be sent in FederationStreamRow.data. The type depends
+            on the type of stream.
+        """
+        raise NotImplementedError()
+
+    def add_to_buffer(self, buff):
+        """Add this row to the appropriate field in the buffer ready for this
+        to be sent over federation.
+
+        We use a buffer so that we can batch up events that have come in at
+        the same time and send them all at once.
+
+        Args:
+            buff (BufferedToSend)
+        """
+        raise NotImplementedError()
+
+
+class PresenceRow(BaseFederationRow, namedtuple("PresenceRow", (
+    "state",  # UserPresenceState
+))):
+    TypeId = "p"
+
+    @staticmethod
+    def from_data(data):
+        return PresenceRow(
+            state=UserPresenceState.from_dict(data)
+        )
+
+    def to_data(self):
+        return self.state.as_dict()
+
+    def add_to_buffer(self, buff):
+        buff.presence.append(self.state)
+
+
+class KeyedEduRow(BaseFederationRow, namedtuple("KeyedEduRow", (
+    "key",  # tuple(str) - the edu key passed to send_edu
+    "edu",  # Edu
+))):
+    """Streams EDUs that have an associated key that is ued to clobber. For example,
+    typing EDUs clobber based on room_id.
+    """
+
+    TypeId = "k"
+
+    @staticmethod
+    def from_data(data):
+        return KeyedEduRow(
+            key=tuple(data["key"]),
+            edu=Edu(**data["edu"]),
+        )
+
+    def to_data(self):
+        return {
+            "key": self.key,
+            "edu": self.edu.get_internal_dict(),
+        }
+
+    def add_to_buffer(self, buff):
+        buff.keyed_edus.setdefault(
+            self.edu.destination, {}
+        )[self.key] = self.edu
+
+
+class EduRow(BaseFederationRow, namedtuple("EduRow", (
+    "edu",  # Edu
+))):
+    """Streams EDUs that don't have keys. See KeyedEduRow
+    """
+    TypeId = "e"
+
+    @staticmethod
+    def from_data(data):
+        return EduRow(Edu(**data))
+
+    def to_data(self):
+        return self.edu.get_internal_dict()
+
+    def add_to_buffer(self, buff):
+        buff.edus.setdefault(self.edu.destination, []).append(self.edu)
+
+
+class FailureRow(BaseFederationRow, namedtuple("FailureRow", (
+    "destination",  # str
+    "failure",
+))):
+    """Streams failures to a remote server. Failures are issued when there was
+    something wrong with a transaction the remote sent us, e.g. it included
+    an event that was invalid.
+    """
+
+    TypeId = "f"
+
+    @staticmethod
+    def from_data(data):
+        return FailureRow(
+            destination=data["destination"],
+            failure=data["failure"],
+        )
+
+    def to_data(self):
+        return {
+            "destination": self.destination,
+            "failure": self.failure,
+        }
+
+    def add_to_buffer(self, buff):
+        buff.failures.setdefault(self.destination, []).append(self.failure)
+
+
+class DeviceRow(BaseFederationRow, namedtuple("DeviceRow", (
+    "destination",  # str
+))):
+    """Streams the fact that either a) there is pending to device messages for
+    users on the remote, or b) a local users device has changed and needs to
+    be sent to the remote.
+    """
+    TypeId = "d"
+
+    @staticmethod
+    def from_data(data):
+        return DeviceRow(destination=data["destination"])
+
+    def to_data(self):
+        return {"destination": self.destination}
+
+    def add_to_buffer(self, buff):
+        buff.device_destinations.add(self.destination)
+
+
+TypeToRow = {
+    Row.TypeId: Row
+    for Row in (
+        PresenceRow,
+        KeyedEduRow,
+        EduRow,
+        FailureRow,
+        DeviceRow,
+    )
+}
+
+
+ParsedFederationStreamData = namedtuple("ParsedFederationStreamData", (
+    "presence",  # list(UserPresenceState)
+    "keyed_edus",  # dict of destination -> { key -> Edu }
+    "edus",  # dict of destination -> [Edu]
+    "failures",  # dict of destination -> [failures]
+    "device_destinations",  # set of destinations
+))
+
+
+def process_rows_for_federation(transaction_queue, rows):
+    """Parse a list of rows from the federation stream and put them in the
+    transaction queue ready for sending to the relevant homeservers.
+
+    Args:
+        transaction_queue (TransactionQueue)
+        rows (list(synapse.replication.tcp.streams.FederationStreamRow))
+    """
+
+    # The federation stream contains a bunch of different types of
+    # rows that need to be handled differently. We parse the rows, put
+    # them into the appropriate collection and then send them off.
+
+    buff = ParsedFederationStreamData(
+        presence=[],
+        keyed_edus={},
+        edus={},
+        failures={},
+        device_destinations=set(),
+    )
+
+    # Parse the rows in the stream and add to the buffer
+    for row in rows:
+        if row.type not in TypeToRow:
+            logger.error("Unrecognized federation row type %r", row.type)
+            continue
+
+        RowType = TypeToRow[row.type]
+        parsed_row = RowType.from_data(row.data)
+        parsed_row.add_to_buffer(buff)
+
+    if buff.presence:
+        transaction_queue.send_presence(buff.presence)
+
+    for destination, edu_map in buff.keyed_edus.iteritems():
+        for key, edu in edu_map.items():
+            transaction_queue.send_edu(
+                edu.destination, edu.edu_type, edu.content, key=key,
+            )
+
+    for destination, edu_list in buff.edus.iteritems():
+        for edu in edu_list:
+            transaction_queue.send_edu(
+                edu.destination, edu.edu_type, edu.content, key=None,
+            )
+
+    for destination, failure_list in buff.failures.iteritems():
+        for failure in failure_list:
+            transaction_queue.send_failure(destination, failure)
+
+    for destination in buff.device_destinations:
+        transaction_queue.send_device_messages(destination)

synapse/federation/transaction_queue.py

@@ -12,7 +12,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-
+import datetime
 
 from twisted.internet import defer
 
@@ -21,13 +21,10 @@ from .units import Transaction, Edu
 
 from synapse.api.errors import HttpResponseException
 from synapse.util.async import run_on_reactor
-from synapse.util.logcontext import preserve_context_over_fn
-from synapse.util.retryutils import (
-    get_retry_limiter, NotRetryingDestination,
-)
+from synapse.util.logcontext import preserve_context_over_fn, preserve_fn
+from synapse.util.retryutils import NotRetryingDestination, get_retry_limiter
 from synapse.util.metrics import measure_func
-from synapse.types import get_domain_from_id
-from synapse.handlers.presence import format_user_presence_state
+from synapse.handlers.presence import format_user_presence_state, get_interested_remotes
 import synapse.metrics
 
 import logging
@@ -43,6 +40,8 @@ sent_pdus_destination_dist = client_metrics.register_distribution(
 )
 sent_edus_counter = client_metrics.register_counter("sent_edus")
 
+sent_transactions_counter = client_metrics.register_counter("sent_transactions")
+
 
 class TransactionQueue(object):
     """This class makes sure we only have one transaction in flight at
@@ -79,8 +78,18 @@ class TransactionQueue(object):
         # destination -> list of tuple(edu, deferred)
         self.pending_edus_by_dest = edus = {}
 
-        # Presence needs to be separate as we send single aggragate EDUs
+        # Map of user_id -> UserPresenceState for all the pending presence
+        # to be sent out by user_id. Entries here get processed and put in
+        # pending_presence_by_dest
+        self.pending_presence = {}
+
+        # Map of destination -> user_id -> UserPresenceState of pending presence
+        # to be sent to each destinations
         self.pending_presence_by_dest = presence = {}
+
+        # Pending EDUs by their "key". Keyed EDUs are EDUs that get clobbered
+        # based on their key (e.g. typing events by room_id)
+        # Map of destination -> (edu_type, key) -> Edu
         self.pending_edus_keyed_by_dest = edus_keyed = {}
 
         metrics.register_callback(
@@ -99,7 +108,12 @@ class TransactionQueue(object):
         # destination -> list of tuple(failure, deferred)
         self.pending_failures_by_dest = {}
 
+        # destination -> stream_id of last successfully sent to-device message.
+        # NB: may be a long or an int.
         self.last_device_stream_id_by_dest = {}
+
+        # destination -> stream_id of last successfully sent device list
+        # update.
         self.last_device_list_stream_id_by_dest = {}
 
         # HACK to get unique tx id
@@ -110,6 +124,8 @@ class TransactionQueue(object):
         self._is_processing = False
         self._last_poked_id = -1
 
+        self._processing_pending_presence = False
+
     def can_send_to(self, destination):
         """Can we send messages to the given server?
 
@@ -166,15 +182,12 @@ class TransactionQueue(object):
                     # Otherwise if the last member on a server in a room is
                     # banned then it won't receive the event because it won't
                     # be in the room after the ban.
-                    users_in_room = yield self.state.get_current_user_in_room(
+                    destinations = yield self.state.get_current_hosts_in_room(
                         event.room_id, latest_event_ids=[
                             prev_id for prev_id, _ in event.prev_events
                         ],
                     )
 
-                    destinations = set(
-                        get_domain_from_id(user_id) for user_id in users_in_room
-                    )
                     if send_on_behalf_of is not None:
                         # If we are sending the event on behalf of another server
                         # then it already has the event and there is no reason to
@@ -221,17 +234,71 @@ class TransactionQueue(object):
                 self._attempt_new_transaction, destination
             )
 
-    def send_presence(self, destination, states):
-        if not self.can_send_to(destination):
-            return
+    @preserve_fn  # the caller should not yield on this
+    @defer.inlineCallbacks
+    def send_presence(self, states):
+        """Send the new presence states to the appropriate destinations.
 
-        self.pending_presence_by_dest.setdefault(destination, {}).update({
+        This actually queues up the presence states ready for sending and
+        triggers a background task to process them and send out the transactions.
+
+        Args:
+            states (list(UserPresenceState))
+        """
+
+        # First we queue up the new presence by user ID, so multiple presence
+        # updates in quick successtion are correctly handled
+        # We only want to send presence for our own users, so lets always just
+        # filter here just in case.
+        self.pending_presence.update({
             state.user_id: state for state in states
+            if self.is_mine_id(state.user_id)
         })
 
-        preserve_context_over_fn(
-            self._attempt_new_transaction, destination
-        )
+        # We then handle the new pending presence in batches, first figuring
+        # out the destinations we need to send each state to and then poking it
+        # to attempt a new transaction. We linearize this so that we don't
+        # accidentally mess up the ordering and send multiple presence updates
+        # in the wrong order
+        if self._processing_pending_presence:
+            return
+
+        self._processing_pending_presence = True
+        try:
+            while True:
+                states_map = self.pending_presence
+                self.pending_presence = {}
+
+                if not states_map:
+                    break
+
+                yield self._process_presence_inner(states_map.values())
+        finally:
+            self._processing_pending_presence = False
+
+    @measure_func("txnqueue._process_presence")
+    @defer.inlineCallbacks
+    def _process_presence_inner(self, states):
+        """Given a list of states populate self.pending_presence_by_dest and
+        poke to send a new transaction to each destination
+
+        Args:
+            states (list(UserPresenceState))
+        """
+        hosts_and_states = yield get_interested_remotes(self.store, states, self.state)
+
+        for destinations, states in hosts_and_states:
+            for destination in destinations:
+                if not self.can_send_to(destination):
+                    continue
+
+                self.pending_presence_by_dest.setdefault(
+                    destination, {}
+                ).update({
+                    state.user_id: state for state in states
+                })
+
+                preserve_fn(self._attempt_new_transaction)(destination)
 
     def send_edu(self, destination, edu_type, content, key=None):
         edu = Edu(
@@ -300,12 +367,33 @@ class TransactionQueue(object):
             )
             return
 
+        pending_pdus = []
         try:
             self.pending_transactions[destination] = 1
 
+            # This will throw if we wouldn't retry. We do this here so we fail
+            # quickly, but we will later check this again in the http client,
+            # hence why we throw the result away.
+            yield get_retry_limiter(destination, self.clock, self.store)
+
+            # XXX: what's this for?
             yield run_on_reactor()
 
+            pending_pdus = []
             while True:
+                device_message_edus, device_stream_id, dev_list_id = (
+                    yield self._get_new_device_messages(destination)
+                )
+
+                # BEGIN CRITICAL SECTION
+                #
+                # In order to avoid a race condition, we need to make sure that
+                # the following code (from popping the queues up to the point
+                # where we decide if we actually have any pending messages) is
+                # atomic - otherwise new PDUs or EDUs might arrive in the
+                # meantime, but not get sent because we hold the
+                # pending_transactions flag.
+
                 pending_pdus = self.pending_pdus_by_dest.pop(destination, [])
                 pending_edus = self.pending_edus_by_dest.pop(destination, [])
                 pending_presence = self.pending_presence_by_dest.pop(destination, {})
@@ -315,17 +403,6 @@ class TransactionQueue(object):
                     self.pending_edus_keyed_by_dest.pop(destination, {}).values()
                 )
 
-                limiter = yield get_retry_limiter(
-                    destination,
-                    self.clock,
-                    self.store,
-                    backoff_on_404=True,  # If we get a 404 the other side has gone
-                )
-
-                device_message_edus, device_stream_id, dev_list_id = (
-                    yield self._get_new_device_messages(destination)
-                )
-
                 pending_edus.extend(device_message_edus)
                 if pending_presence:
                     pending_edus.append(
@@ -355,11 +432,13 @@ class TransactionQueue(object):
                     )
                     return
 
+                # END CRITICAL SECTION
+
                 success = yield self._send_new_transaction(
                     destination, pending_pdus, pending_edus, pending_failures,
-                    limiter=limiter,
                 )
                 if success:
+                    sent_transactions_counter.inc()
                     # Remove the acknowledged device messages from the database
                     # Only bother if we actually sent some device messages
                     if device_message_edus:
@@ -375,12 +454,24 @@ class TransactionQueue(object):
                     self.last_device_list_stream_id_by_dest[destination] = dev_list_id
                 else:
                     break
-        except NotRetryingDestination:
+        except NotRetryingDestination as e:
             logger.debug(
-                "TX [%s] not ready for retry yet - "
+                "TX [%s] not ready for retry yet (next retry at %s) - "
                 "dropping transaction for now",
                 destination,
+                datetime.datetime.fromtimestamp(
+                    (e.retry_last_ts + e.retry_interval) / 1000.0
+                ),
             )
+        except Exception as e:
+            logger.warn(
+                "TX [%s] Failed to send transaction: %s",
+                destination,
+                e,
+            )
+            for p, _ in pending_pdus:
+                logger.info("Failed to send event %s to %s", p.event_id,
+                            destination)
         finally:
             # We want to be *very* sure we delete this after we stop processing
             self.pending_transactions.pop(destination, None)
@@ -420,7 +511,7 @@ class TransactionQueue(object):
     @measure_func("_send_new_transaction")
     @defer.inlineCallbacks
     def _send_new_transaction(self, destination, pending_pdus, pending_edus,
-                              pending_failures, limiter):
+                              pending_failures):
 
         # Sort based on the order field
         pending_pdus.sort(key=lambda t: t[1])
@@ -430,132 +521,104 @@ class TransactionQueue(object):
 
         success = True
 
+        logger.debug("TX [%s] _attempt_new_transaction", destination)
+
+        txn_id = str(self._next_txn_id)
+
+        logger.debug(
+            "TX [%s] {%s} Attempting new transaction"
+            " (pdus: %d, edus: %d, failures: %d)",
+            destination, txn_id,
+            len(pdus),
+            len(edus),
+            len(failures)
+        )
+
+        logger.debug("TX [%s] Persisting transaction...", destination)
+
+        transaction = Transaction.create_new(
+            origin_server_ts=int(self.clock.time_msec()),
+            transaction_id=txn_id,
+            origin=self.server_name,
+            destination=destination,
+            pdus=pdus,
+            edus=edus,
+            pdu_failures=failures,
+        )
+
+        self._next_txn_id += 1
+
+        yield self.transaction_actions.prepare_to_send(transaction)
+
+        logger.debug("TX [%s] Persisted transaction", destination)
+        logger.info(
+            "TX [%s] {%s} Sending transaction [%s],"
+            " (PDUs: %d, EDUs: %d, failures: %d)",
+            destination, txn_id,
+            transaction.transaction_id,
+            len(pdus),
+            len(edus),
+            len(failures),
+        )
+
+        # Actually send the transaction
+
+        # FIXME (erikj): This is a bit of a hack to make the Pdu age
+        # keys work
+        def json_data_cb():
+            data = transaction.get_dict()
+            now = int(self.clock.time_msec())
+            if "pdus" in data:
+                for p in data["pdus"]:
+                    if "age_ts" in p:
+                        unsigned = p.setdefault("unsigned", {})
+                        unsigned["age"] = now - int(p["age_ts"])
+                        del p["age_ts"]
+            return data
+
         try:
-            logger.debug("TX [%s] _attempt_new_transaction", destination)
-
-            txn_id = str(self._next_txn_id)
-
-            logger.debug(
-                "TX [%s] {%s} Attempting new transaction"
-                " (pdus: %d, edus: %d, failures: %d)",
-                destination, txn_id,
-                len(pdus),
-                len(edus),
-                len(failures)
+            response = yield self.transport_layer.send_transaction(
+                transaction, json_data_cb
             )
+            code = 200
 
-            logger.debug("TX [%s] Persisting transaction...", destination)
-
-            transaction = Transaction.create_new(
-                origin_server_ts=int(self.clock.time_msec()),
-                transaction_id=txn_id,
-                origin=self.server_name,
-                destination=destination,
-                pdus=pdus,
-                edus=edus,
-                pdu_failures=failures,
-            )
-
-            self._next_txn_id += 1
-
-            yield self.transaction_actions.prepare_to_send(transaction)
-
-            logger.debug("TX [%s] Persisted transaction", destination)
-            logger.info(
-                "TX [%s] {%s} Sending transaction [%s],"
-                " (PDUs: %d, EDUs: %d, failures: %d)",
-                destination, txn_id,
-                transaction.transaction_id,
-                len(pdus),
-                len(edus),
-                len(failures),
-            )
-
-            with limiter:
-                # Actually send the transaction
-
-                # FIXME (erikj): This is a bit of a hack to make the Pdu age
-                # keys work
-                def json_data_cb():
-                    data = transaction.get_dict()
-                    now = int(self.clock.time_msec())
-                    if "pdus" in data:
-                        for p in data["pdus"]:
-                            if "age_ts" in p:
-                                unsigned = p.setdefault("unsigned", {})
-                                unsigned["age"] = now - int(p["age_ts"])
-                                del p["age_ts"]
-                    return data
-
-                try:
-                    response = yield self.transport_layer.send_transaction(
-                        transaction, json_data_cb
-                    )
-                    code = 200
-
-                    if response:
-                        for e_id, r in response.get("pdus", {}).items():
-                            if "error" in r:
-                                logger.warn(
-                                    "Transaction returned error for %s: %s",
-                                    e_id, r,
-                                )
-                except HttpResponseException as e:
-                    code = e.code
-                    response = e.response
-
-                    if e.code in (401, 404, 429) or 500 <= e.code:
-                        logger.info(
-                            "TX [%s] {%s} got %d response",
-                            destination, txn_id, code
+            if response:
+                for e_id, r in response.get("pdus", {}).items():
+                    if "error" in r:
+                        logger.warn(
+                            "Transaction returned error for %s: %s",
+                            e_id, r,
                         )
-                        raise e
+        except HttpResponseException as e:
+            code = e.code
+            response = e.response
 
+            if e.code in (401, 404, 429) or 500 <= e.code:
                 logger.info(
                     "TX [%s] {%s} got %d response",
                     destination, txn_id, code
                 )
+                raise e
 
-                logger.debug("TX [%s] Sent transaction", destination)
-                logger.debug("TX [%s] Marking as delivered...", destination)
+        logger.info(
+            "TX [%s] {%s} got %d response",
+            destination, txn_id, code
+        )
 
-            yield self.transaction_actions.delivered(
-                transaction, code, response
-            )
+        logger.debug("TX [%s] Sent transaction", destination)
+        logger.debug("TX [%s] Marking as delivered...", destination)
 
-            logger.debug("TX [%s] Marked as delivered", destination)
+        yield self.transaction_actions.delivered(
+            transaction, code, response
+        )
 
-            if code != 200:
-                for p in pdus:
-                    logger.info(
-                        "Failed to send event %s to %s", p.event_id, destination
-                    )
-                success = False
-        except RuntimeError as e:
-            # We capture this here as there as nothing actually listens
-            # for this finishing functions deferred.
-            logger.warn(
-                "TX [%s] Problem in _attempt_transaction: %s",
-                destination,
-                e,
-            )
-
-            success = False
+        logger.debug("TX [%s] Marked as delivered", destination)
 
+        if code != 200:
             for p in pdus:
-                logger.info("Failed to send event %s to %s", p.event_id, destination)
-        except Exception as e:
-            # We capture this here as there as nothing actually listens
-            # for this finishing functions deferred.
-            logger.warn(
-                "TX [%s] Problem in _attempt_transaction: %s",
-                destination,
-                e,
-            )
-
+                logger.info(
+                    "Failed to send event %s to %s", p.event_id, destination
+                )
             success = False
 
-            for p in pdus:
-                logger.info("Failed to send event %s to %s", p.event_id, destination)
-
         defer.returnValue(success)

synapse/federation/transport/client.py

@@ -163,6 +163,7 @@ class TransportLayerClient(object):
             data=json_data,
             json_data_callback=json_data_callback,
             long_retries=True,
+            backoff_on_404=True,  # If we get a 404 the other side has gone
         )
 
         logger.debug(
@@ -174,7 +175,8 @@ class TransportLayerClient(object):
 
     @defer.inlineCallbacks
     @log_function
-    def make_query(self, destination, query_type, args, retry_on_dns_fail):
+    def make_query(self, destination, query_type, args, retry_on_dns_fail,
+                   ignore_backoff=False):
         path = PREFIX + "/query/%s" % query_type
 
         content = yield self.client.get_json(
@@ -183,6 +185,7 @@ class TransportLayerClient(object):
             args=args,
             retry_on_dns_fail=retry_on_dns_fail,
             timeout=10000,
+            ignore_backoff=ignore_backoff,
         )
 
         defer.returnValue(content)
@@ -190,6 +193,26 @@ class TransportLayerClient(object):
     @defer.inlineCallbacks
     @log_function
     def make_membership_event(self, destination, room_id, user_id, membership):
+        """Asks a remote server to build and sign us a membership event
+
+        Note that this does not append any events to any graphs.
+
+        Args:
+            destination (str): address of remote homeserver
+            room_id (str): room to join/leave
+            user_id (str): user to be joined/left
+            membership (str): one of join/leave
+
+        Returns:
+            Deferred: Succeeds when we get a 2xx HTTP response. The result
+            will be the decoded JSON body (ie, the new event).
+
+            Fails with ``HTTPRequestException`` if we get an HTTP response
+            code >= 300.
+
+            Fails with ``NotRetryingDestination`` if we are not yet ready
+            to retry this server.
+        """
         valid_memberships = {Membership.JOIN, Membership.LEAVE}
         if membership not in valid_memberships:
             raise RuntimeError(
@@ -198,11 +221,23 @@ class TransportLayerClient(object):
             )
         path = PREFIX + "/make_%s/%s/%s" % (membership, room_id, user_id)
 
+        ignore_backoff = False
+        retry_on_dns_fail = False
+
+        if membership == Membership.LEAVE:
+            # we particularly want to do our best to send leave events. The
+            # problem is that if it fails, we won't retry it later, so if the
+            # remote server was just having a momentary blip, the room will be
+            # out of sync.
+            ignore_backoff = True
+            retry_on_dns_fail = True
+
         content = yield self.client.get_json(
             destination=destination,
             path=path,
-            retry_on_dns_fail=False,
+            retry_on_dns_fail=retry_on_dns_fail,
             timeout=20000,
+            ignore_backoff=ignore_backoff,
         )
 
         defer.returnValue(content)
@@ -229,6 +264,12 @@ class TransportLayerClient(object):
             destination=destination,
             path=path,
             data=content,
+
+            # we want to do our best to send this through. The problem is
+            # that if it fails, we won't retry it later, so if the remote
+            # server was just having a momentary blip, the room will be out of
+            # sync.
+            ignore_backoff=True,
         )
 
         defer.returnValue(response)
@@ -242,6 +283,7 @@ class TransportLayerClient(object):
             destination=destination,
             path=path,
             data=content,
+            ignore_backoff=True,
         )
 
         defer.returnValue(response)
@@ -269,6 +311,7 @@ class TransportLayerClient(object):
             destination=remote_server,
             path=path,
             args=args,
+            ignore_backoff=True,
         )
 
         defer.returnValue(response)

synapse/federation/transport/server.py

@@ -24,6 +24,7 @@ from synapse.http.servlet import (
 )
 from synapse.util.ratelimitutils import FederationRateLimiter
 from synapse.util.versionstring import get_version_string
+from synapse.util.logcontext import preserve_fn
 from synapse.types import ThirdPartyInstanceID
 
 import functools
@@ -79,6 +80,7 @@ class Authenticator(object):
     def __init__(self, hs):
         self.keyring = hs.get_keyring()
         self.server_name = hs.hostname
+        self.store = hs.get_datastore()
 
     # A method just so we can pass 'self' as the authenticator to the Servlets
     @defer.inlineCallbacks
@@ -138,6 +140,13 @@ class Authenticator(object):
         logger.info("Request from %s", origin)
         request.authenticated_entity = origin
 
+        # If we get a valid signed request from the other side, its probably
+        # alive
+        retry_timings = yield self.store.get_destination_retry_timings(origin)
+        if retry_timings and retry_timings["retry_last_ts"]:
+            logger.info("Marking origin %r as up", origin)
+            preserve_fn(self.store.set_destination_retry_timings)(origin, 0, 0)
+
         defer.returnValue(origin)
 
 

synapse/handlers/_base.py

@@ -53,7 +53,20 @@ class BaseHandler(object):
 
         self.event_builder_factory = hs.get_event_builder_factory()
 
-    def ratelimit(self, requester):
+    @defer.inlineCallbacks
+    def ratelimit(self, requester, update=True):
+        """Ratelimits requests.
+
+        Args:
+            requester (Requester)
+            update (bool): Whether to record that a request is being processed.
+                Set to False when doing multiple checks for one request (e.g.
+                to check up front if we would reject the request), and set to
+                True for the last call for a given request.
+
+        Raises:
+            LimitExceededError if the request should be ratelimited
+        """
         time_now = self.clock.time()
         user_id = requester.user.to_string()
 
@@ -67,10 +80,25 @@ class BaseHandler(object):
         if requester.app_service and not requester.app_service.is_rate_limited():
             return
 
+        # Check if there is a per user override in the DB.
+        override = yield self.store.get_ratelimit_for_user(user_id)
+        if override:
+            # If overriden with a null Hz then ratelimiting has been entirely
+            # disabled for the user
+            if not override.messages_per_second:
+                return
+
+            messages_per_second = override.messages_per_second
+            burst_count = override.burst_count
+        else:
+            messages_per_second = self.hs.config.rc_messages_per_second
+            burst_count = self.hs.config.rc_message_burst_count
+
         allowed, time_allowed = self.ratelimiter.send_message(
             user_id, time_now,
-            msg_rate_hz=self.hs.config.rc_messages_per_second,
-            burst_count=self.hs.config.rc_message_burst_count,
+            msg_rate_hz=messages_per_second,
+            burst_count=burst_count,
+            update=update,
         )
         if not allowed:
             raise LimitExceededError(

synapse/handlers/admin.py

@@ -19,7 +19,6 @@ from ._base import BaseHandler
 
 import logging
 
-
 logger = logging.getLogger(__name__)
 
 
@@ -54,3 +53,46 @@ class AdminHandler(BaseHandler):
         }
 
         defer.returnValue(ret)
+
+    @defer.inlineCallbacks
+    def get_users(self):
+        """Function to reterive a list of users in users table.
+
+        Args:
+        Returns:
+            defer.Deferred: resolves to list[dict[str, Any]]
+        """
+        ret = yield self.store.get_users()
+
+        defer.returnValue(ret)
+
+    @defer.inlineCallbacks
+    def get_users_paginate(self, order, start, limit):
+        """Function to reterive a paginated list of users from
+        users list. This will return a json object, which contains
+        list of users and the total number of users in users table.
+
+        Args:
+            order (str): column name to order the select by this column
+            start (int): start number to begin the query from
+            limit (int): number of rows to reterive
+        Returns:
+            defer.Deferred: resolves to json object {list[dict[str, Any]], count}
+        """
+        ret = yield self.store.get_users_paginate(order, start, limit)
+
+        defer.returnValue(ret)
+
+    @defer.inlineCallbacks
+    def search_users(self, term):
+        """Function to search users list for one or more users with
+        the matched term.
+
+        Args:
+            term (str): search term
+        Returns:
+            defer.Deferred: resolves to list[dict[str, Any]]
+        """
+        ret = yield self.store.search_users(term)
+
+        defer.returnValue(ret)

synapse/handlers/auth.py

@@ -1,5 +1,6 @@
 # -*- coding: utf-8 -*-
 # Copyright 2014 - 2016 OpenMarket Ltd
+# Copyright 2017 Vector Creations Ltd
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -47,6 +48,7 @@ class AuthHandler(BaseHandler):
             LoginType.PASSWORD: self._check_password_auth,
             LoginType.RECAPTCHA: self._check_recaptcha,
             LoginType.EMAIL_IDENTITY: self._check_email_identity,
+            LoginType.MSISDN: self._check_msisdn,
             LoginType.DUMMY: self._check_dummy_auth,
         }
         self.bcrypt_rounds = hs.config.bcrypt_rounds
@@ -307,31 +309,47 @@ class AuthHandler(BaseHandler):
                 defer.returnValue(True)
         raise LoginError(401, "", errcode=Codes.UNAUTHORIZED)
 
-    @defer.inlineCallbacks
     def _check_email_identity(self, authdict, _):
+        return self._check_threepid('email', authdict)
+
+    def _check_msisdn(self, authdict, _):
+        return self._check_threepid('msisdn', authdict)
+
+    @defer.inlineCallbacks
+    def _check_dummy_auth(self, authdict, _):
+        yield run_on_reactor()
+        defer.returnValue(True)
+
+    @defer.inlineCallbacks
+    def _check_threepid(self, medium, authdict):
         yield run_on_reactor()
 
         if 'threepid_creds' not in authdict:
             raise LoginError(400, "Missing threepid_creds", Codes.MISSING_PARAM)
 
         threepid_creds = authdict['threepid_creds']
+
         identity_handler = self.hs.get_handlers().identity_handler
 
-        logger.info("Getting validated threepid. threepidcreds: %r" % (threepid_creds,))
+        logger.info("Getting validated threepid. threepidcreds: %r", (threepid_creds,))
         threepid = yield identity_handler.threepid_from_creds(threepid_creds)
 
         if not threepid:
             raise LoginError(401, "", errcode=Codes.UNAUTHORIZED)
 
+        if threepid['medium'] != medium:
+            raise LoginError(
+                401,
+                "Expecting threepid of type '%s', got '%s'" % (
+                    medium, threepid['medium'],
+                ),
+                errcode=Codes.UNAUTHORIZED
+            )
+
         threepid['threepid_creds'] = authdict['threepid_creds']
 
         defer.returnValue(threepid)
 
-    @defer.inlineCallbacks
-    def _check_dummy_auth(self, authdict, _):
-        yield run_on_reactor()
-        defer.returnValue(True)
-
     def _get_params_recaptcha(self):
         return {"public_key": self.hs.config.recaptcha_public_key}
 

synapse/handlers/device.py

@@ -12,11 +12,12 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-
 from synapse.api import errors
 from synapse.api.constants import EventTypes
 from synapse.util import stringutils
 from synapse.util.async import Linearizer
+from synapse.util.caches.expiringcache import ExpiringCache
+from synapse.util.retryutils import NotRetryingDestination
 from synapse.util.metrics import measure_func
 from synapse.types import get_domain_from_id, RoomStreamToken
 from twisted.internet import defer
@@ -35,10 +36,11 @@ class DeviceHandler(BaseHandler):
         self.state = hs.get_state_handler()
         self.federation_sender = hs.get_federation_sender()
         self.federation = hs.get_replication_layer()
-        self._remote_edue_linearizer = Linearizer(name="remote_device_list")
+
+        self._edu_updater = DeviceListEduUpdater(hs, self)
 
         self.federation.register_edu_handler(
-            "m.device_list_update", self._incoming_device_list_update,
+            "m.device_list_update", self._edu_updater.incoming_device_list_update,
         )
         self.federation.register_query_handler(
             "user_devices", self.on_federation_query_user_devices,
@@ -168,6 +170,40 @@ class DeviceHandler(BaseHandler):
 
         yield self.notify_device_update(user_id, [device_id])
 
+    @defer.inlineCallbacks
+    def delete_devices(self, user_id, device_ids):
+        """ Delete several devices
+
+        Args:
+            user_id (str):
+            device_ids (str): The list of device IDs to delete
+
+        Returns:
+            defer.Deferred:
+        """
+
+        try:
+            yield self.store.delete_devices(user_id, device_ids)
+        except errors.StoreError, e:
+            if e.code == 404:
+                # no match
+                pass
+            else:
+                raise
+
+        # Delete access tokens and e2e keys for each device. Not optimised as it is not
+        # considered as part of a critical path.
+        for device_id in device_ids:
+            yield self.store.user_delete_access_tokens(
+                user_id, device_id=device_id,
+                delete_refresh_tokens=True,
+            )
+            yield self.store.delete_e2e_keys_by_device(
+                user_id=user_id, device_id=device_id
+            )
+
+        yield self.notify_device_update(user_id, device_ids)
+
     @defer.inlineCallbacks
     def update_device(self, user_id, device_id, content):
         """ Update the given device
@@ -213,8 +249,7 @@ class DeviceHandler(BaseHandler):
             user_id, device_ids, list(hosts)
         )
 
-        rooms = yield self.store.get_rooms_for_user(user_id)
-        room_ids = [r.room_id for r in rooms]
+        room_ids = yield self.store.get_rooms_for_user(user_id)
 
         yield self.notifier.on_new_event(
             "device_list_key", position, rooms=room_ids,
@@ -235,8 +270,7 @@ class DeviceHandler(BaseHandler):
             user_id (str)
             from_token (StreamToken)
         """
-        rooms = yield self.store.get_rooms_for_user(user_id)
-        room_ids = set(r.room_id for r in rooms)
+        room_ids = yield self.store.get_rooms_for_user(user_id)
 
         # First we check if any devices have changed
         changed = yield self.store.get_user_whose_devices_changed(
@@ -246,30 +280,51 @@ class DeviceHandler(BaseHandler):
         # Then work out if any users have since joined
         rooms_changed = self.store.get_rooms_that_changed(room_ids, from_token.room_key)
 
+        stream_ordering = RoomStreamToken.parse_stream_token(
+            from_token.room_key).stream
+
         possibly_changed = set(changed)
         for room_id in rooms_changed:
-            # Fetch  the current state at the time.
-            stream_ordering = RoomStreamToken.parse_stream_token(from_token.room_key)
-
+            # Fetch the current state at the time.
             try:
                 event_ids = yield self.store.get_forward_extremeties_for_room(
                     room_id, stream_ordering=stream_ordering
                 )
-                prev_state_ids = yield self.store.get_state_ids_for_events(event_ids)
-            except:
-                prev_state_ids = {}
+            except errors.StoreError:
+                # we have purged the stream_ordering index since the stream
+                # ordering: treat it the same as a new room
+                event_ids = []
 
-            current_state_ids = yield self.state.get_current_state_ids(room_id)
+            current_state_ids = yield self.store.get_current_state_ids(room_id)
+
+            # special-case for an empty prev state: include all members
+            # in the changed list
+            if not event_ids:
+                for key, event_id in current_state_ids.iteritems():
+                    etype, state_key = key
+                    if etype != EventTypes.Member:
+                        continue
+                    possibly_changed.add(state_key)
+                continue
+
+            # mapping from event_id -> state_dict
+            prev_state_ids = yield self.store.get_state_ids_for_events(event_ids)
 
             # If there has been any change in membership, include them in the
             # possibly changed list. We'll check if they are joined below,
             # and we're not toooo worried about spuriously adding users.
             for key, event_id in current_state_ids.iteritems():
                 etype, state_key = key
-                if etype == EventTypes.Member:
-                    prev_event_id = prev_state_ids.get(key, None)
+                if etype != EventTypes.Member:
+                    continue
+
+                # check if this member has changed since any of the extremities
+                # at the stream_ordering, and add them to the list if so.
+                for state_dict in prev_state_ids.values():
+                    prev_event_id = state_dict.get(key, None)
                     if not prev_event_id or prev_event_id != event_id:
                         possibly_changed.add(state_key)
+                        break
 
         users_who_share_room = yield self.store.get_users_who_share_room_with_user(
             user_id
@@ -279,58 +334,6 @@ class DeviceHandler(BaseHandler):
         # and those that actually still share a room with the user
         defer.returnValue(users_who_share_room & possibly_changed)
 
-    @measure_func("_incoming_device_list_update")
-    @defer.inlineCallbacks
-    def _incoming_device_list_update(self, origin, edu_content):
-        user_id = edu_content["user_id"]
-        device_id = edu_content["device_id"]
-        stream_id = edu_content["stream_id"]
-        prev_ids = edu_content.get("prev_id", [])
-
-        if get_domain_from_id(user_id) != origin:
-            # TODO: Raise?
-            logger.warning("Got device list update edu for %r from %r", user_id, origin)
-            return
-
-        rooms = yield self.store.get_rooms_for_user(user_id)
-        if not rooms:
-            # We don't share any rooms with this user. Ignore update, as we
-            # probably won't get any further updates.
-            return
-
-        with (yield self._remote_edue_linearizer.queue(user_id)):
-            # If the prev id matches whats in our cache table, then we don't need
-            # to resync the users device list, otherwise we do.
-            resync = True
-            if len(prev_ids) == 1:
-                extremity = yield self.store.get_device_list_last_stream_id_for_remote(
-                    user_id
-                )
-                logger.info("Extrem: %r, prev_ids: %r", extremity, prev_ids)
-                if str(extremity) == str(prev_ids[0]):
-                    resync = False
-
-            if resync:
-                # Fetch all devices for the user.
-                result = yield self.federation.query_user_devices(origin, user_id)
-                stream_id = result["stream_id"]
-                devices = result["devices"]
-                yield self.store.update_remote_device_list_cache(
-                    user_id, devices, stream_id,
-                )
-                device_ids = [device["device_id"] for device in devices]
-                yield self.notify_device_update(user_id, device_ids)
-            else:
-                # Simply update the single device, since we know that is the only
-                # change (becuase of the single prev_id matching the current cache)
-                content = dict(edu_content)
-                for key in ("user_id", "device_id", "stream_id", "prev_ids"):
-                    content.pop(key, None)
-                yield self.store.update_remote_device_list_cache_entry(
-                    user_id, device_id, content, stream_id,
-                )
-                yield self.notify_device_update(user_id, [device_id])
-
     @defer.inlineCallbacks
     def on_federation_query_user_devices(self, user_id):
         stream_id, devices = yield self.store.get_devices_with_keys_by_user(user_id)
@@ -343,8 +346,8 @@ class DeviceHandler(BaseHandler):
     @defer.inlineCallbacks
     def user_left_room(self, user, room_id):
         user_id = user.to_string()
-        rooms = yield self.store.get_rooms_for_user(user_id)
-        if not rooms:
+        room_ids = yield self.store.get_rooms_for_user(user_id)
+        if not room_ids:
             # We no longer share rooms with this user, so we'll no longer
             # receive device updates. Mark this in DB.
             yield self.store.mark_remote_user_device_list_as_unsubscribed(user_id)
@@ -356,3 +359,155 @@ def _update_device_from_client_ips(device, client_ips):
         "last_seen_ts": ip.get("last_seen"),
         "last_seen_ip": ip.get("ip"),
     })
+
+
+class DeviceListEduUpdater(object):
+    "Handles incoming device list updates from federation and updates the DB"
+
+    def __init__(self, hs, device_handler):
+        self.store = hs.get_datastore()
+        self.federation = hs.get_replication_layer()
+        self.clock = hs.get_clock()
+        self.device_handler = device_handler
+
+        self._remote_edu_linearizer = Linearizer(name="remote_device_list")
+
+        # user_id -> list of updates waiting to be handled.
+        self._pending_updates = {}
+
+        # Recently seen stream ids. We don't bother keeping these in the DB,
+        # but they're useful to have them about to reduce the number of spurious
+        # resyncs.
+        self._seen_updates = ExpiringCache(
+            cache_name="device_update_edu",
+            clock=self.clock,
+            max_len=10000,
+            expiry_ms=30 * 60 * 1000,
+            iterable=True,
+        )
+
+    @defer.inlineCallbacks
+    def incoming_device_list_update(self, origin, edu_content):
+        """Called on incoming device list update from federation. Responsible
+        for parsing the EDU and adding to pending updates list.
+        """
+
+        user_id = edu_content.pop("user_id")
+        device_id = edu_content.pop("device_id")
+        stream_id = str(edu_content.pop("stream_id"))  # They may come as ints
+        prev_ids = edu_content.pop("prev_id", [])
+        prev_ids = [str(p) for p in prev_ids]   # They may come as ints
+
+        if get_domain_from_id(user_id) != origin:
+            # TODO: Raise?
+            logger.warning("Got device list update edu for %r from %r", user_id, origin)
+            return
+
+        room_ids = yield self.store.get_rooms_for_user(user_id)
+        if not room_ids:
+            # We don't share any rooms with this user. Ignore update, as we
+            # probably won't get any further updates.
+            return
+
+        self._pending_updates.setdefault(user_id, []).append(
+            (device_id, stream_id, prev_ids, edu_content)
+        )
+
+        yield self._handle_device_updates(user_id)
+
+    @measure_func("_incoming_device_list_update")
+    @defer.inlineCallbacks
+    def _handle_device_updates(self, user_id):
+        "Actually handle pending updates."
+
+        with (yield self._remote_edu_linearizer.queue(user_id)):
+            pending_updates = self._pending_updates.pop(user_id, [])
+            if not pending_updates:
+                # This can happen since we batch updates
+                return
+
+            # Given a list of updates we check if we need to resync. This
+            # happens if we've missed updates.
+            resync = yield self._need_to_do_resync(user_id, pending_updates)
+
+            if resync:
+                # Fetch all devices for the user.
+                origin = get_domain_from_id(user_id)
+                try:
+                    result = yield self.federation.query_user_devices(origin, user_id)
+                except NotRetryingDestination:
+                    # TODO: Remember that we are now out of sync and try again
+                    # later
+                    logger.warn(
+                        "Failed to handle device list update for %s,"
+                        " we're not retrying the remote",
+                        user_id,
+                    )
+                    # We abort on exceptions rather than accepting the update
+                    # as otherwise synapse will 'forget' that its device list
+                    # is out of date. If we bail then we will retry the resync
+                    # next time we get a device list update for this user_id.
+                    # This makes it more likely that the device lists will
+                    # eventually become consistent.
+                    return
+                except Exception:
+                    # TODO: Remember that we are now out of sync and try again
+                    # later
+                    logger.exception(
+                        "Failed to handle device list update for %s", user_id
+                    )
+                    return
+
+                stream_id = result["stream_id"]
+                devices = result["devices"]
+                yield self.store.update_remote_device_list_cache(
+                    user_id, devices, stream_id,
+                )
+                device_ids = [device["device_id"] for device in devices]
+                yield self.device_handler.notify_device_update(user_id, device_ids)
+            else:
+                # Simply update the single device, since we know that is the only
+                # change (becuase of the single prev_id matching the current cache)
+                for device_id, stream_id, prev_ids, content in pending_updates:
+                    yield self.store.update_remote_device_list_cache_entry(
+                        user_id, device_id, content, stream_id,
+                    )
+
+                yield self.device_handler.notify_device_update(
+                    user_id, [device_id for device_id, _, _, _ in pending_updates]
+                )
+
+            self._seen_updates.setdefault(user_id, set()).update(
+                stream_id for _, stream_id, _, _ in pending_updates
+            )
+
+    @defer.inlineCallbacks
+    def _need_to_do_resync(self, user_id, updates):
+        """Given a list of updates for a user figure out if we need to do a full
+        resync, or whether we have enough data that we can just apply the delta.
+        """
+        seen_updates = self._seen_updates.get(user_id, set())
+
+        extremity = yield self.store.get_device_list_last_stream_id_for_remote(
+            user_id
+        )
+
+        stream_id_in_updates = set()  # stream_ids in updates list
+        for _, stream_id, prev_ids, _ in updates:
+            if not prev_ids:
+                # We always do a resync if there are no previous IDs
+                defer.returnValue(True)
+
+            for prev_id in prev_ids:
+                if prev_id == extremity:
+                    continue
+                elif prev_id in seen_updates:
+                    continue
+                elif prev_id in stream_id_in_updates:
+                    continue
+                else:
+                    defer.returnValue(True)
+
+            stream_id_in_updates.add(stream_id)
+
+        defer.returnValue(False)

synapse/handlers/directory.py

@@ -175,6 +175,7 @@ class DirectoryHandler(BaseHandler):
                         "room_alias": room_alias.to_string(),
                     },
                     retry_on_dns_fail=False,
+                    ignore_backoff=True,
                 )
             except CodeMessageException as e:
                 logging.warn("Error retrieving alias")

synapse/handlers/e2e_keys.py

@@ -21,8 +21,8 @@ from twisted.internet import defer
 
 from synapse.api.errors import SynapseError, CodeMessageException
 from synapse.types import get_domain_from_id
-from synapse.util.logcontext import preserve_fn, preserve_context_over_deferred
-from synapse.util.retryutils import get_retry_limiter, NotRetryingDestination
+from synapse.util.logcontext import preserve_fn, make_deferred_yieldable
+from synapse.util.retryutils import NotRetryingDestination
 
 logger = logging.getLogger(__name__)
 
@@ -121,15 +121,11 @@ class E2eKeysHandler(object):
         def do_remote_query(destination):
             destination_query = remote_queries_not_in_cache[destination]
             try:
-                limiter = yield get_retry_limiter(
-                    destination, self.clock, self.store
+                remote_result = yield self.federation.query_client_keys(
+                    destination,
+                    {"device_keys": destination_query},
+                    timeout=timeout
                 )
-                with limiter:
-                    remote_result = yield self.federation.query_client_keys(
-                        destination,
-                        {"device_keys": destination_query},
-                        timeout=timeout
-                    )
 
                 for user_id, keys in remote_result["device_keys"].items():
                     if user_id in destination_query:
@@ -149,7 +145,7 @@ class E2eKeysHandler(object):
                     "status": 503, "message": e.message
                 }
 
-        yield preserve_context_over_deferred(defer.gatherResults([
+        yield make_deferred_yieldable(defer.gatherResults([
             preserve_fn(do_remote_query)(destination)
             for destination in remote_queries_not_in_cache
         ]))
@@ -239,18 +235,14 @@ class E2eKeysHandler(object):
         def claim_client_keys(destination):
             device_keys = remote_queries[destination]
             try:
-                limiter = yield get_retry_limiter(
-                    destination, self.clock, self.store
+                remote_result = yield self.federation.claim_client_keys(
+                    destination,
+                    {"one_time_keys": device_keys},
+                    timeout=timeout
                 )
-                with limiter:
-                    remote_result = yield self.federation.claim_client_keys(
-                        destination,
-                        {"one_time_keys": device_keys},
-                        timeout=timeout
-                    )
-                    for user_id, keys in remote_result["one_time_keys"].items():
-                        if user_id in device_keys:
-                            json_result[user_id] = keys
+                for user_id, keys in remote_result["one_time_keys"].items():
+                    if user_id in device_keys:
+                        json_result[user_id] = keys
             except CodeMessageException as e:
                 failures[destination] = {
                     "status": e.code, "message": e.message
@@ -265,11 +257,21 @@ class E2eKeysHandler(object):
                     "status": 503, "message": e.message
                 }
 
-        yield preserve_context_over_deferred(defer.gatherResults([
+        yield make_deferred_yieldable(defer.gatherResults([
             preserve_fn(claim_client_keys)(destination)
             for destination in remote_queries
         ]))
 
+        logger.info(
+            "Claimed one-time-keys: %s",
+            ",".join((
+                "%s for %s:%s" % (key_id, user_id, device_id)
+                for user_id, user_keys in json_result.iteritems()
+                for device_id, device_keys in user_keys.iteritems()
+                for key_id, _ in device_keys.iteritems()
+            )),
+        )
+
         defer.returnValue({
             "one_time_keys": json_result,
             "failures": failures
@@ -296,19 +298,8 @@ class E2eKeysHandler(object):
 
         one_time_keys = keys.get("one_time_keys", None)
         if one_time_keys:
-            logger.info(
-                "Adding %d one_time_keys for device %r for user %r at %d",
-                len(one_time_keys), device_id, user_id, time_now
-            )
-            key_list = []
-            for key_id, key_json in one_time_keys.items():
-                algorithm, key_id = key_id.split(":")
-                key_list.append((
-                    algorithm, key_id, encode_canonical_json(key_json)
-                ))
-
-            yield self.store.add_e2e_one_time_keys(
-                user_id, device_id, time_now, key_list
+            yield self._upload_one_time_keys_for_user(
+                user_id, device_id, time_now, one_time_keys,
             )
 
         # the device should have been registered already, but it may have been
@@ -316,8 +307,63 @@ class E2eKeysHandler(object):
         # old access_token without an associated device_id. Either way, we
         # need to double-check the device is registered to avoid ending up with
         # keys without a corresponding device.
-        self.device_handler.check_device_registered(user_id, device_id)
+        yield self.device_handler.check_device_registered(user_id, device_id)
 
         result = yield self.store.count_e2e_one_time_keys(user_id, device_id)
 
         defer.returnValue({"one_time_key_counts": result})
+
+    @defer.inlineCallbacks
+    def _upload_one_time_keys_for_user(self, user_id, device_id, time_now,
+                                       one_time_keys):
+        logger.info(
+            "Adding one_time_keys %r for device %r for user %r at %d",
+            one_time_keys.keys(), device_id, user_id, time_now,
+        )
+
+        # make a list of (alg, id, key) tuples
+        key_list = []
+        for key_id, key_obj in one_time_keys.items():
+            algorithm, key_id = key_id.split(":")
+            key_list.append((
+                algorithm, key_id, key_obj
+            ))
+
+        # First we check if we have already persisted any of the keys.
+        existing_key_map = yield self.store.get_e2e_one_time_keys(
+            user_id, device_id, [k_id for _, k_id, _ in key_list]
+        )
+
+        new_keys = []  # Keys that we need to insert. (alg, id, json) tuples.
+        for algorithm, key_id, key in key_list:
+            ex_json = existing_key_map.get((algorithm, key_id), None)
+            if ex_json:
+                if not _one_time_keys_match(ex_json, key):
+                    raise SynapseError(
+                        400,
+                        ("One time key %s:%s already exists. "
+                         "Old key: %s; new key: %r") %
+                        (algorithm, key_id, ex_json, key)
+                    )
+            else:
+                new_keys.append((algorithm, key_id, encode_canonical_json(key)))
+
+        yield self.store.add_e2e_one_time_keys(
+            user_id, device_id, time_now, new_keys
+        )
+
+
+def _one_time_keys_match(old_key_json, new_key):
+    old_key = json.loads(old_key_json)
+
+    # if either is a string rather than an object, they must match exactly
+    if not isinstance(old_key, dict) or not isinstance(new_key, dict):
+        return old_key == new_key
+
+    # otherwise, we strip off the 'signatures' if any, because it's legitimate
+    # for different upload attempts to have different signatures.
+    old_key.pop("signatures", None)
+    new_key_copy = dict(new_key)
+    new_key_copy.pop("signatures", None)
+
+    return old_key == new_key_copy

synapse/handlers/federation.py

@@ -14,6 +14,7 @@
 # limitations under the License.
 
 """Contains handlers for federation events."""
+import synapse.util.logcontext
 from signedjson.key import decode_verify_key_bytes
 from signedjson.sign import verify_signed_json
 from unpaddedbase64 import decode_base64
@@ -27,11 +28,11 @@ from synapse.api.constants import EventTypes, Membership, RejectedReason
 from synapse.events.validator import EventValidator
 from synapse.util import unwrapFirstError
 from synapse.util.logcontext import (
-    PreserveLoggingContext, preserve_fn, preserve_context_over_deferred
+    preserve_fn, preserve_context_over_deferred
 )
 from synapse.util.metrics import measure_func
 from synapse.util.logutils import log_function
-from synapse.util.async import run_on_reactor
+from synapse.util.async import run_on_reactor, Linearizer
 from synapse.util.frozenutils import unfreeze
 from synapse.crypto.event_signing import (
     compute_event_signature, add_hashes_and_signatures,
@@ -79,29 +80,227 @@ class FederationHandler(BaseHandler):
 
         # When joining a room we need to queue any events for that room up
         self.room_queues = {}
+        self._room_pdu_linearizer = Linearizer("fed_room_pdu")
 
-    @log_function
     @defer.inlineCallbacks
-    def on_receive_pdu(self, origin, pdu, state=None, auth_chain=None):
-        """ Called by the ReplicationLayer when we have a new pdu. We need to
-        do auth checks and put it through the StateHandler.
+    @log_function
+    def on_receive_pdu(self, origin, pdu, get_missing=True):
+        """ Process a PDU received via a federation /send/ transaction, or
+        via backfill of missing prev_events
 
-        auth_chain and state are None if we already have the necessary state
-        and prev_events in the db
+        Args:
+            origin (str): server which initiated the /send/ transaction. Will
+                be used to fetch missing events or state.
+            pdu (FrozenEvent): received PDU
+            get_missing (bool): True if we should fetch missing prev_events
+
+        Returns (Deferred): completes with None
         """
-        event = pdu
 
-        logger.debug("Got event: %s", event.event_id)
+        # We reprocess pdus when we have seen them only as outliers
+        existing = yield self.get_persisted_pdu(
+            origin, pdu.event_id, do_auth=False
+        )
+
+        # FIXME: Currently we fetch an event again when we already have it
+        # if it has been marked as an outlier.
+
+        already_seen = (
+            existing and (
+                not existing.internal_metadata.is_outlier()
+                or pdu.internal_metadata.is_outlier()
+            )
+        )
+        if already_seen:
+            logger.debug("Already seen pdu %s", pdu.event_id)
+            return
 
         # If we are currently in the process of joining this room, then we
         # queue up events for later processing.
-        if event.room_id in self.room_queues:
-            self.room_queues[event.room_id].append((pdu, origin))
+        if pdu.room_id in self.room_queues:
+            logger.info("Ignoring PDU %s for room %s from %s for now; join "
+                        "in progress", pdu.event_id, pdu.room_id, origin)
+            self.room_queues[pdu.room_id].append((pdu, origin))
             return
 
-        logger.debug("Processing event: %s", event.event_id)
+        state = None
 
-        logger.debug("Event: %s", event)
+        auth_chain = []
+
+        have_seen = yield self.store.have_events(
+            [ev for ev, _ in pdu.prev_events]
+        )
+
+        fetch_state = False
+
+        # Get missing pdus if necessary.
+        if not pdu.internal_metadata.is_outlier():
+            # We only backfill backwards to the min depth.
+            min_depth = yield self.get_min_depth_for_context(
+                pdu.room_id
+            )
+
+            logger.debug(
+                "_handle_new_pdu min_depth for %s: %d",
+                pdu.room_id, min_depth
+            )
+
+            prevs = {e_id for e_id, _ in pdu.prev_events}
+            seen = set(have_seen.keys())
+
+            if min_depth and pdu.depth < min_depth:
+                # This is so that we don't notify the user about this
+                # message, to work around the fact that some events will
+                # reference really really old events we really don't want to
+                # send to the clients.
+                pdu.internal_metadata.outlier = True
+            elif min_depth and pdu.depth > min_depth:
+                if get_missing and prevs - seen:
+                    # If we're missing stuff, ensure we only fetch stuff one
+                    # at a time.
+                    logger.info(
+                        "Acquiring lock for room %r to fetch %d missing events: %r...",
+                        pdu.room_id, len(prevs - seen), list(prevs - seen)[:5],
+                    )
+                    with (yield self._room_pdu_linearizer.queue(pdu.room_id)):
+                        logger.info(
+                            "Acquired lock for room %r to fetch %d missing events",
+                            pdu.room_id, len(prevs - seen),
+                        )
+
+                        yield self._get_missing_events_for_pdu(
+                            origin, pdu, prevs, min_depth
+                        )
+
+                        # Update the set of things we've seen after trying to
+                        # fetch the missing stuff
+                        have_seen = yield self.store.have_events(prevs)
+                        seen = set(have_seen.iterkeys())
+
+                        if not prevs - seen:
+                            logger.info(
+                                "Found all missing prev events for %s", pdu.event_id
+                            )
+                elif prevs - seen:
+                    logger.info(
+                        "Not fetching %d missing events for room %r,event %s: %r...",
+                        len(prevs - seen), pdu.room_id, pdu.event_id,
+                        list(prevs - seen)[:5],
+                    )
+
+            if prevs - seen:
+                logger.info(
+                    "Still missing %d events for room %r: %r...",
+                    len(prevs - seen), pdu.room_id, list(prevs - seen)[:5]
+                )
+                fetch_state = True
+
+        if fetch_state:
+            # We need to get the state at this event, since we haven't
+            # processed all the prev events.
+            logger.debug(
+                "_handle_new_pdu getting state for %s",
+                pdu.room_id
+            )
+            try:
+                state, auth_chain = yield self.replication_layer.get_state_for_room(
+                    origin, pdu.room_id, pdu.event_id,
+                )
+            except:
+                logger.exception("Failed to get state for event: %s", pdu.event_id)
+
+        yield self._process_received_pdu(
+            origin,
+            pdu,
+            state=state,
+            auth_chain=auth_chain,
+        )
+
+    @defer.inlineCallbacks
+    def _get_missing_events_for_pdu(self, origin, pdu, prevs, min_depth):
+        """
+        Args:
+            origin (str): Origin of the pdu. Will be called to get the missing events
+            pdu: received pdu
+            prevs (set(str)): List of event ids which we are missing
+            min_depth (int): Minimum depth of events to return.
+        """
+        # We recalculate seen, since it may have changed.
+        have_seen = yield self.store.have_events(prevs)
+        seen = set(have_seen.keys())
+
+        if not prevs - seen:
+            return
+
+        latest = yield self.store.get_latest_event_ids_in_room(
+            pdu.room_id
+        )
+
+        # We add the prev events that we have seen to the latest
+        # list to ensure the remote server doesn't give them to us
+        latest = set(latest)
+        latest |= seen
+
+        logger.info(
+            "Missing %d events for room %r pdu %s: %r...",
+            len(prevs - seen), pdu.room_id, pdu.event_id, list(prevs - seen)[:5]
+        )
+
+        # XXX: we set timeout to 10s to help workaround
+        # https://github.com/matrix-org/synapse/issues/1733.
+        # The reason is to avoid holding the linearizer lock
+        # whilst processing inbound /send transactions, causing
+        # FDs to stack up and block other inbound transactions
+        # which empirically can currently take up to 30 minutes.
+        #
+        # N.B. this explicitly disables retry attempts.
+        #
+        # N.B. this also increases our chances of falling back to
+        # fetching fresh state for the room if the missing event
+        # can't be found, which slightly reduces our security.
+        # it may also increase our DAG extremity count for the room,
+        # causing additional state resolution?  See #1760.
+        # However, fetching state doesn't hold the linearizer lock
+        # apparently.
+        #
+        # see https://github.com/matrix-org/synapse/pull/1744
+
+        missing_events = yield self.replication_layer.get_missing_events(
+            origin,
+            pdu.room_id,
+            earliest_events_ids=list(latest),
+            latest_events=[pdu],
+            limit=10,
+            min_depth=min_depth,
+            timeout=10000,
+        )
+
+        logger.info(
+            "Got %d events: %r...",
+            len(missing_events), [e.event_id for e in missing_events[:5]]
+        )
+
+        # We want to sort these by depth so we process them and
+        # tell clients about them in order.
+        missing_events.sort(key=lambda x: x.depth)
+
+        for e in missing_events:
+            logger.info("Handling found event %s", e.event_id)
+            yield self.on_receive_pdu(
+                origin,
+                e,
+                get_missing=False
+            )
+
+    @log_function
+    @defer.inlineCallbacks
+    def _process_received_pdu(self, origin, pdu, state, auth_chain):
+        """ Called when we have a new pdu. We need to do auth checks and put it
+        through the StateHandler.
+        """
+        event = pdu
+
+        logger.debug("Processing event: %s", event)
 
         # FIXME (erikj): Awful hack to make the case where we are not currently
         # in the room work
@@ -181,13 +380,6 @@ class FederationHandler(BaseHandler):
                     affected=event.event_id,
                 )
 
-        # if we're receiving valid events from an origin,
-        # it's probably a good idea to mark it as not in retry-state
-        # for sending (although this is a bit of a leap)
-        retry_timings = yield self.store.get_destination_retry_timings(origin)
-        if retry_timings and retry_timings["retry_last_ts"]:
-            self.store.set_destination_retry_timings(origin, 0, 0)
-
         room = yield self.store.get_room(event.room_id)
 
         if not room:
@@ -206,11 +398,10 @@ class FederationHandler(BaseHandler):
             target_user = UserID.from_string(target_user_id)
             extra_users.append(target_user)
 
-        with PreserveLoggingContext():
-            self.notifier.on_new_room_event(
-                event, event_stream_id, max_stream_id,
-                extra_users=extra_users
-            )
+        self.notifier.on_new_room_event(
+            event, event_stream_id, max_stream_id,
+            extra_users=extra_users
+        )
 
         if event.type == EventTypes.Member:
             if event.membership == Membership.JOIN:
@@ -670,8 +861,6 @@ class FederationHandler(BaseHandler):
         """
         logger.debug("Joining %s to %s", joinee, room_id)
 
-        yield self.store.clean_room_for_join(room_id)
-
         origin, event = yield self._make_and_verify_event(
             target_hosts,
             room_id,
@@ -680,7 +869,15 @@ class FederationHandler(BaseHandler):
             content,
         )
 
+        # This shouldn't happen, because the RoomMemberHandler has a
+        # linearizer lock which only allows one operation per user per room
+        # at a time - so this is just paranoia.
+        assert (room_id not in self.room_queues)
+
         self.room_queues[room_id] = []
+
+        yield self.store.clean_room_for_join(room_id)
+
         handled_events = set()
 
         try:
@@ -722,28 +919,46 @@ class FederationHandler(BaseHandler):
                 origin, auth_chain, state, event
             )
 
-            with PreserveLoggingContext():
-                self.notifier.on_new_room_event(
-                    event, event_stream_id, max_stream_id,
-                    extra_users=[joinee]
-                )
+            self.notifier.on_new_room_event(
+                event, event_stream_id, max_stream_id,
+                extra_users=[joinee]
+            )
 
             logger.debug("Finished joining %s to %s", joinee, room_id)
         finally:
             room_queue = self.room_queues[room_id]
             del self.room_queues[room_id]
 
-            for p, origin in room_queue:
-                if p.event_id in handled_events:
-                    continue
+            # we don't need to wait for the queued events to be processed -
+            # it's just a best-effort thing at this point. We do want to do
+            # them roughly in order, though, otherwise we'll end up making
+            # lots of requests for missing prev_events which we do actually
+            # have. Hence we fire off the deferred, but don't wait for it.
 
-                try:
-                    self.on_receive_pdu(origin, p)
-                except:
-                    logger.exception("Couldn't handle pdu")
+            synapse.util.logcontext.preserve_fn(self._handle_queued_pdus)(
+                room_queue
+            )
 
         defer.returnValue(True)
 
+    @defer.inlineCallbacks
+    def _handle_queued_pdus(self, room_queue):
+        """Process PDUs which got queued up while we were busy send_joining.
+
+        Args:
+            room_queue (list[FrozenEvent, str]): list of PDUs to be processed
+                and the servers that sent them
+        """
+        for p, origin in room_queue:
+            try:
+                logger.info("Processing queued PDU %s which was received "
+                            "while we were joining %s", p.event_id, p.room_id)
+                yield self.on_receive_pdu(origin, p)
+            except Exception as e:
+                logger.warn(
+                    "Error handling queued PDU %s from %s: %s",
+                    p.event_id, origin, e)
+
     @defer.inlineCallbacks
     @log_function
     def on_make_join_request(self, room_id, user_id):
@@ -791,9 +1006,19 @@ class FederationHandler(BaseHandler):
         )
 
         event.internal_metadata.outlier = False
-        # Send this event on behalf of the origin server since they may not
-        # have an up to data view of the state of the room at this event so
-        # will not know which servers to send the event to.
+        # Send this event on behalf of the origin server.
+        #
+        # The reasons we have the destination server rather than the origin
+        # server send it are slightly mysterious: the origin server should have
+        # all the neccessary state once it gets the response to the send_join,
+        # so it could send the event itself if it wanted to. It may be that
+        # doing it this way reduces failure modes, or avoids certain attacks
+        # where a new server selectively tells a subset of the federation that
+        # it has joined.
+        #
+        # The fact is that, as of the current writing, Synapse doesn't send out
+        # the join event over federation after joining, and changing it now
+        # would introduce the danger of backwards-compatibility problems.
         event.internal_metadata.send_on_behalf_of = origin
 
         context, event_stream_id, max_stream_id = yield self._handle_new_event(
@@ -812,10 +1037,9 @@ class FederationHandler(BaseHandler):
             target_user = UserID.from_string(target_user_id)
             extra_users.append(target_user)
 
-        with PreserveLoggingContext():
-            self.notifier.on_new_room_event(
-                event, event_stream_id, max_stream_id, extra_users=extra_users
-            )
+        self.notifier.on_new_room_event(
+            event, event_stream_id, max_stream_id, extra_users=extra_users
+        )
 
         if event.type == EventTypes.Member:
             if event.content["membership"] == Membership.JOIN:
@@ -861,48 +1085,35 @@ class FederationHandler(BaseHandler):
         )
 
         target_user = UserID.from_string(event.state_key)
-        with PreserveLoggingContext():
-            self.notifier.on_new_room_event(
-                event, event_stream_id, max_stream_id,
-                extra_users=[target_user],
-            )
+        self.notifier.on_new_room_event(
+            event, event_stream_id, max_stream_id,
+            extra_users=[target_user],
+        )
 
         defer.returnValue(event)
 
     @defer.inlineCallbacks
     def do_remotely_reject_invite(self, target_hosts, room_id, user_id):
-        try:
-            origin, event = yield self._make_and_verify_event(
-                target_hosts,
-                room_id,
-                user_id,
-                "leave"
-            )
-            signed_event = self._sign_event(event)
-        except SynapseError:
-            raise
-        except CodeMessageException as e:
-            logger.warn("Failed to reject invite: %s", e)
-            raise SynapseError(500, "Failed to reject invite")
+        origin, event = yield self._make_and_verify_event(
+            target_hosts,
+            room_id,
+            user_id,
+            "leave"
+        )
+        event = self._sign_event(event)
 
-        # Try the host we successfully got a response to /make_join/
-        # request first.
+        # Try the host that we succesfully called /make_leave/ on first for
+        # the /send_leave/ request.
         try:
             target_hosts.remove(origin)
             target_hosts.insert(0, origin)
         except ValueError:
             pass
 
-        try:
-            yield self.replication_layer.send_leave(
-                target_hosts,
-                signed_event
-            )
-        except SynapseError:
-            raise
-        except CodeMessageException as e:
-            logger.warn("Failed to reject invite: %s", e)
-            raise SynapseError(500, "Failed to reject invite")
+        yield self.replication_layer.send_leave(
+            target_hosts,
+            event
+        )
 
         context = yield self.state_handler.compute_event_context(event)
 
@@ -1023,10 +1234,9 @@ class FederationHandler(BaseHandler):
             target_user = UserID.from_string(target_user_id)
             extra_users.append(target_user)
 
-        with PreserveLoggingContext():
-            self.notifier.on_new_room_event(
-                event, event_stream_id, max_stream_id, extra_users=extra_users
-            )
+        self.notifier.on_new_room_event(
+            event, event_stream_id, max_stream_id, extra_users=extra_users
+        )
 
         defer.returnValue(None)
 
@@ -1096,7 +1306,7 @@ class FederationHandler(BaseHandler):
                     if prev_id != event.event_id:
                         results[(event.type, event.state_key)] = prev_id
                 else:
-                    del results[(event.type, event.state_key)]
+                    results.pop((event.type, event.state_key), None)
 
             defer.returnValue(results.values())
         else:
@@ -1325,7 +1535,17 @@ class FederationHandler(BaseHandler):
 
     @defer.inlineCallbacks
     def _prep_event(self, origin, event, state=None, auth_events=None):
+        """
 
+        Args:
+            origin:
+            event:
+            state:
+            auth_events:
+
+        Returns:
+            Deferred, which resolves to synapse.events.snapshot.EventContext
+        """
         context = yield self.state_handler.compute_event_context(
             event, old_state=state,
         )

synapse/handlers/identity.py

@@ -1,5 +1,6 @@
 # -*- coding: utf-8 -*-
 # Copyright 2015, 2016 OpenMarket Ltd
+# Copyright 2017 Vector Creations Ltd
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -17,7 +18,7 @@
 from twisted.internet import defer
 
 from synapse.api.errors import (
-    CodeMessageException
+    MatrixCodeMessageException, CodeMessageException
 )
 from ._base import BaseHandler
 from synapse.util.async import run_on_reactor
@@ -89,6 +90,9 @@ class IdentityHandler(BaseHandler):
                 ),
                 {'sid': creds['sid'], 'client_secret': client_secret}
             )
+        except MatrixCodeMessageException as e:
+            logger.info("getValidated3pid failed with Matrix error: %r", e)
+            raise SynapseError(e.code, e.msg, e.errcode)
         except CodeMessageException as e:
             data = json.loads(e.msg)
 
@@ -150,7 +154,7 @@ class IdentityHandler(BaseHandler):
         params.update(kwargs)
 
         try:
-            data = yield self.http_client.post_urlencoded_get_json(
+            data = yield self.http_client.post_json_get_json(
                 "https://%s%s" % (
                     id_server,
                     "/_matrix/identity/api/v1/validate/email/requestToken"
@@ -158,6 +162,46 @@ class IdentityHandler(BaseHandler):
                 params
             )
             defer.returnValue(data)
+        except MatrixCodeMessageException as e:
+            logger.info("Proxied requestToken failed with Matrix error: %r", e)
+            raise SynapseError(e.code, e.msg, e.errcode)
+        except CodeMessageException as e:
+            logger.info("Proxied requestToken failed: %r", e)
+            raise e
+
+    @defer.inlineCallbacks
+    def requestMsisdnToken(
+            self, id_server, country, phone_number,
+            client_secret, send_attempt, **kwargs
+    ):
+        yield run_on_reactor()
+
+        if not self._should_trust_id_server(id_server):
+            raise SynapseError(
+                400, "Untrusted ID server '%s'" % id_server,
+                Codes.SERVER_NOT_TRUSTED
+            )
+
+        params = {
+            'country': country,
+            'phone_number': phone_number,
+            'client_secret': client_secret,
+            'send_attempt': send_attempt,
+        }
+        params.update(kwargs)
+
+        try:
+            data = yield self.http_client.post_json_get_json(
+                "https://%s%s" % (
+                    id_server,
+                    "/_matrix/identity/api/v1/validate/msisdn/requestToken"
+                ),
+                params
+            )
+            defer.returnValue(data)
+        except MatrixCodeMessageException as e:
+            logger.info("Proxied requestToken failed with Matrix error: %r", e)
+            raise SynapseError(e.code, e.msg, e.errcode)
         except CodeMessageException as e:
             logger.info("Proxied requestToken failed: %r", e)
             raise e

synapse/handlers/initial_sync.py

@@ -19,6 +19,7 @@ from synapse.api.constants import EventTypes, Membership
 from synapse.api.errors import AuthError, Codes
 from synapse.events.utils import serialize_event
 from synapse.events.validator import EventValidator
+from synapse.handlers.presence import format_user_presence_state
 from synapse.streams.config import PaginationConfig
 from synapse.types import (
     UserID, StreamToken,
@@ -225,9 +226,17 @@ class InitialSyncHandler(BaseHandler):
                 "content": content,
             })
 
+        now = self.clock.time_msec()
+
         ret = {
             "rooms": rooms_ret,
-            "presence": presence,
+            "presence": [
+                {
+                    "type": "m.presence",
+                    "content": format_user_presence_state(event, now),
+                }
+                for event in presence
+            ],
             "account_data": account_data_events,
             "receipts": receipt,
             "end": now_token.to_string(),

synapse/handlers/message.py

@@ -16,7 +16,7 @@
 from twisted.internet import defer
 
 from synapse.api.constants import EventTypes, Membership
-from synapse.api.errors import AuthError, Codes, SynapseError, LimitExceededError
+from synapse.api.errors import AuthError, Codes, SynapseError
 from synapse.crypto.event_signing import add_hashes_and_signatures
 from synapse.events.utils import serialize_event
 from synapse.events.validator import EventValidator
@@ -175,7 +175,8 @@ class MessageHandler(BaseHandler):
         defer.returnValue(chunk)
 
     @defer.inlineCallbacks
-    def create_event(self, event_dict, token_id=None, txn_id=None, prev_event_ids=None):
+    def create_event(self, requester, event_dict, token_id=None, txn_id=None,
+                     prev_event_ids=None):
         """
         Given a dict from a client, create a new event.
 
@@ -185,6 +186,7 @@ class MessageHandler(BaseHandler):
         Adds display names to Join membership events.
 
         Args:
+            requester
             event_dict (dict): An entire event
             token_id (str)
             txn_id (str)
@@ -226,6 +228,7 @@ class MessageHandler(BaseHandler):
 
             event, context = yield self._create_new_client_event(
                 builder=builder,
+                requester=requester,
                 prev_event_ids=prev_event_ids,
             )
 
@@ -251,17 +254,7 @@ class MessageHandler(BaseHandler):
         # We check here if we are currently being rate limited, so that we
         # don't do unnecessary work. We check again just before we actually
         # send the event.
-        time_now = self.clock.time()
-        allowed, time_allowed = self.ratelimiter.send_message(
-            event.sender, time_now,
-            msg_rate_hz=self.hs.config.rc_messages_per_second,
-            burst_count=self.hs.config.rc_message_burst_count,
-            update=False,
-        )
-        if not allowed:
-            raise LimitExceededError(
-                retry_after_ms=int(1000 * (time_allowed - time_now)),
-            )
+        yield self.ratelimit(requester, update=False)
 
         user = UserID.from_string(event.sender)
 
@@ -319,6 +312,7 @@ class MessageHandler(BaseHandler):
         See self.create_event and self.send_nonmember_event.
         """
         event, context = yield self.create_event(
+            requester,
             event_dict,
             token_id=requester.access_token_id,
             txn_id=txn_id
@@ -416,7 +410,7 @@ class MessageHandler(BaseHandler):
 
     @measure_func("_create_new_client_event")
     @defer.inlineCallbacks
-    def _create_new_client_event(self, builder, prev_event_ids=None):
+    def _create_new_client_event(self, builder, requester=None, prev_event_ids=None):
         if prev_event_ids:
             prev_events = yield self.store.add_event_hashes(prev_event_ids)
             prev_max_depth = yield self.store.get_max_depth_of_events(prev_event_ids)
@@ -456,6 +450,8 @@ class MessageHandler(BaseHandler):
         state_handler = self.state_handler
 
         context = yield state_handler.compute_event_context(builder)
+        if requester:
+            context.app_service = requester.app_service
 
         if builder.is_state():
             builder.prev_state = yield self.store.add_event_hashes(
@@ -493,7 +489,7 @@ class MessageHandler(BaseHandler):
         # We now need to go and hit out to wherever we need to hit out to.
 
         if ratelimit:
-            self.ratelimit(requester)
+            yield self.ratelimit(requester)
 
         try:
             yield self.auth.check_from_context(event, context)
@@ -531,9 +527,9 @@ class MessageHandler(BaseHandler):
 
                 state_to_include_ids = [
                     e_id
-                    for k, e_id in context.current_state_ids.items()
+                    for k, e_id in context.current_state_ids.iteritems()
                     if k[0] in self.hs.config.room_invite_state_types
-                    or k[0] == EventTypes.Member and k[1] == event.sender
+                    or k == (EventTypes.Member, event.sender)
                 ]
 
                 state_to_include = yield self.store.get_events(state_to_include_ids)
@@ -545,7 +541,7 @@ class MessageHandler(BaseHandler):
                         "content": e.content,
                         "sender": e.sender,
                     }
-                    for e in state_to_include.values()
+                    for e in state_to_include.itervalues()
                 ]
 
                 invitee = UserID.from_string(event.state_key)
@@ -612,12 +608,9 @@ class MessageHandler(BaseHandler):
         @defer.inlineCallbacks
         def _notify():
             yield run_on_reactor()
-            yield self.notifier.on_new_room_event(
+            self.notifier.on_new_room_event(
                 event, event_stream_id, max_stream_id,
                 extra_users=extra_users
             )
 
         preserve_fn(_notify)()
-
-        # If invite, remove room_state from unsigned before sending.
-        event.unsigned.pop("invite_room_state", None)

synapse/handlers/presence.py

@@ -29,6 +29,8 @@ from synapse.api.errors import SynapseError
 from synapse.api.constants import PresenceState
 from synapse.storage.presence import UserPresenceState
 
+from synapse.util.caches.descriptors import cachedInlineCallbacks
+from synapse.util.async import Linearizer
 from synapse.util.logcontext import preserve_fn
 from synapse.util.logutils import log_function
 from synapse.util.metrics import Measure
@@ -186,6 +188,7 @@ class PresenceHandler(object):
         # process_id to millisecond timestamp last updated.
         self.external_process_to_current_syncs = {}
         self.external_process_last_updated_ms = {}
+        self.external_sync_linearizer = Linearizer(name="external_sync_linearizer")
 
         # Start a LoopingCall in 30s that fires every 5s.
         # The initial delay is to allow disconnected clients a chance to
@@ -315,11 +318,7 @@ class PresenceHandler(object):
             if to_federation_ping:
                 federation_presence_out_counter.inc_by(len(to_federation_ping))
 
-                _, _, hosts_to_states = yield self._get_interested_parties(
-                    to_federation_ping.values()
-                )
-
-                self._push_to_remotes(hosts_to_states)
+                self._push_to_remotes(to_federation_ping.values())
 
     def _handle_timeouts(self):
         """Checks the presence of users that have timed out and updates as
@@ -507,6 +506,73 @@ class PresenceHandler(object):
         self.external_process_last_updated_ms[process_id] = self.clock.time_msec()
         self.external_process_to_current_syncs[process_id] = syncing_user_ids
 
+    @defer.inlineCallbacks
+    def update_external_syncs_row(self, process_id, user_id, is_syncing, sync_time_msec):
+        """Update the syncing users for an external process as a delta.
+
+        Args:
+            process_id (str): An identifier for the process the users are
+                syncing against. This allows synapse to process updates
+                as user start and stop syncing against a given process.
+            user_id (str): The user who has started or stopped syncing
+            is_syncing (bool): Whether or not the user is now syncing
+            sync_time_msec(int): Time in ms when the user was last syncing
+        """
+        with (yield self.external_sync_linearizer.queue(process_id)):
+            prev_state = yield self.current_state_for_user(user_id)
+
+            process_presence = self.external_process_to_current_syncs.setdefault(
+                process_id, set()
+            )
+
+            updates = []
+            if is_syncing and user_id not in process_presence:
+                if prev_state.state == PresenceState.OFFLINE:
+                    updates.append(prev_state.copy_and_replace(
+                        state=PresenceState.ONLINE,
+                        last_active_ts=sync_time_msec,
+                        last_user_sync_ts=sync_time_msec,
+                    ))
+                else:
+                    updates.append(prev_state.copy_and_replace(
+                        last_user_sync_ts=sync_time_msec,
+                    ))
+                process_presence.add(user_id)
+            elif user_id in process_presence:
+                updates.append(prev_state.copy_and_replace(
+                    last_user_sync_ts=sync_time_msec,
+                ))
+
+            if not is_syncing:
+                process_presence.discard(user_id)
+
+            if updates:
+                yield self._update_states(updates)
+
+            self.external_process_last_updated_ms[process_id] = self.clock.time_msec()
+
+    @defer.inlineCallbacks
+    def update_external_syncs_clear(self, process_id):
+        """Marks all users that had been marked as syncing by a given process
+        as offline.
+
+        Used when the process has stopped/disappeared.
+        """
+        with (yield self.external_sync_linearizer.queue(process_id)):
+            process_presence = self.external_process_to_current_syncs.pop(
+                process_id, set()
+            )
+            prev_states = yield self.current_state_for_users(process_presence)
+            time_now_ms = self.clock.time_msec()
+
+            yield self._update_states([
+                prev_state.copy_and_replace(
+                    last_user_sync_ts=time_now_ms,
+                )
+                for prev_state in prev_states.itervalues()
+            ])
+            self.external_process_last_updated_ms.pop(process_id, None)
+
     @defer.inlineCallbacks
     def current_state_for_user(self, user_id):
         """Get the current presence state for a user.
@@ -526,14 +592,14 @@ class PresenceHandler(object):
             for user_id in user_ids
         }
 
-        missing = [user_id for user_id, state in states.items() if not state]
+        missing = [user_id for user_id, state in states.iteritems() if not state]
         if missing:
             # There are things not in our in memory cache. Lets pull them out of
             # the database.
             res = yield self.store.get_presence_for_users(missing)
-            states.update({state.user_id: state for state in res})
+            states.update(res)
 
-            missing = [user_id for user_id, state in states.items() if not state]
+            missing = [user_id for user_id, state in states.iteritems() if not state]
             if missing:
                 new = {
                     user_id: UserPresenceState.default(user_id)
@@ -544,55 +610,6 @@ class PresenceHandler(object):
 
         defer.returnValue(states)
 
-    @defer.inlineCallbacks
-    def _get_interested_parties(self, states, calculate_remote_hosts=True):
-        """Given a list of states return which entities (rooms, users, servers)
-        are interested in the given states.
-
-        Returns:
-            3-tuple: `(room_ids_to_states, users_to_states, hosts_to_states)`,
-            with each item being a dict of `entity_name` -> `[UserPresenceState]`
-        """
-        room_ids_to_states = {}
-        users_to_states = {}
-        for state in states:
-            events = yield self.store.get_rooms_for_user(state.user_id)
-            for e in events:
-                room_ids_to_states.setdefault(e.room_id, []).append(state)
-
-            plist = yield self.store.get_presence_list_observers_accepted(state.user_id)
-            for u in plist:
-                users_to_states.setdefault(u, []).append(state)
-
-            # Always notify self
-            users_to_states.setdefault(state.user_id, []).append(state)
-
-        hosts_to_states = {}
-        if calculate_remote_hosts:
-            for room_id, states in room_ids_to_states.items():
-                local_states = filter(lambda s: self.is_mine_id(s.user_id), states)
-                if not local_states:
-                    continue
-
-                users = yield self.store.get_users_in_room(room_id)
-                hosts = set(get_domain_from_id(u) for u in users)
-
-                for host in hosts:
-                    hosts_to_states.setdefault(host, []).extend(local_states)
-
-        for user_id, states in users_to_states.items():
-            local_states = filter(lambda s: self.is_mine_id(s.user_id), states)
-            if not local_states:
-                continue
-
-            host = get_domain_from_id(user_id)
-            hosts_to_states.setdefault(host, []).extend(local_states)
-
-        # TODO: de-dup hosts_to_states, as a single host might have multiple
-        # of same presence
-
-        defer.returnValue((room_ids_to_states, users_to_states, hosts_to_states))
-
     @defer.inlineCallbacks
     def _persist_and_notify(self, states):
         """Persist states in the database, poke the notifier and send to
@@ -600,34 +617,33 @@ class PresenceHandler(object):
         """
         stream_id, max_token = yield self.store.update_presence(states)
 
-        parties = yield self._get_interested_parties(states)
-        room_ids_to_states, users_to_states, hosts_to_states = parties
+        parties = yield get_interested_parties(self.store, states)
+        room_ids_to_states, users_to_states = parties
 
         self.notifier.on_new_event(
             "presence_key", stream_id, rooms=room_ids_to_states.keys(),
-            users=[UserID.from_string(u) for u in users_to_states.keys()]
+            users=[UserID.from_string(u) for u in users_to_states]
         )
 
-        self._push_to_remotes(hosts_to_states)
+        self._push_to_remotes(states)
 
     @defer.inlineCallbacks
     def notify_for_states(self, state, stream_id):
-        parties = yield self._get_interested_parties([state])
-        room_ids_to_states, users_to_states, hosts_to_states = parties
+        parties = yield get_interested_parties(self.store, [state])
+        room_ids_to_states, users_to_states = parties
 
         self.notifier.on_new_event(
             "presence_key", stream_id, rooms=room_ids_to_states.keys(),
-            users=[UserID.from_string(u) for u in users_to_states.keys()]
+            users=[UserID.from_string(u) for u in users_to_states]
         )
 
-    def _push_to_remotes(self, hosts_to_states):
+    def _push_to_remotes(self, states):
         """Sends state updates to remote servers.
 
         Args:
-            hosts_to_states (dict): Mapping `server_name` -> `[UserPresenceState]`
+            states (list(UserPresenceState))
         """
-        for host, states in hosts_to_states.items():
-            self.federation.send_presence(host, states)
+        self.federation.send_presence(states)
 
     @defer.inlineCallbacks
     def incoming_presence(self, origin, content):
@@ -719,9 +735,7 @@ class PresenceHandler(object):
                 for state in updates
             ])
         else:
-            defer.returnValue([
-                format_user_presence_state(state, now) for state in updates
-            ])
+            defer.returnValue(updates)
 
     @defer.inlineCallbacks
     def set_state(self, target_user, state, ignore_status_msg=False):
@@ -766,18 +780,17 @@ class PresenceHandler(object):
         # don't need to send to local clients here, as that is done as part
         # of the event stream/sync.
         # TODO: Only send to servers not already in the room.
-        user_ids = yield self.store.get_users_in_room(room_id)
         if self.is_mine(user):
             state = yield self.current_state_for_user(user.to_string())
 
-            hosts = set(get_domain_from_id(u) for u in user_ids)
-            self._push_to_remotes({host: (state,) for host in hosts})
+            self._push_to_remotes([state])
         else:
+            user_ids = yield self.store.get_users_in_room(room_id)
             user_ids = filter(self.is_mine_id, user_ids)
 
             states = yield self.current_state_for_users(user_ids)
 
-            self._push_to_remotes({user.domain: states.values()})
+            self._push_to_remotes(states.values())
 
     @defer.inlineCallbacks
     def get_presence_list(self, observer_user, accepted=None):
@@ -795,6 +808,9 @@ class PresenceHandler(object):
             as_event=False,
         )
 
+        now = self.clock.time_msec()
+        results[:] = [format_user_presence_state(r, now) for r in results]
+
         is_accepted = {
             row["observed_user_id"]: row["accepted"] for row in presence_list
         }
@@ -847,6 +863,7 @@ class PresenceHandler(object):
             )
 
             state_dict = yield self.get_state(observed_user, as_event=False)
+            state_dict = format_user_presence_state(state_dict, self.clock.time_msec())
 
             self.federation.send_edu(
                 destination=observer_user.domain,
@@ -910,11 +927,12 @@ class PresenceHandler(object):
     def is_visible(self, observed_user, observer_user):
         """Returns whether a user can see another user's presence.
         """
-        observer_rooms = yield self.store.get_rooms_for_user(observer_user.to_string())
-        observed_rooms = yield self.store.get_rooms_for_user(observed_user.to_string())
-
-        observer_room_ids = set(r.room_id for r in observer_rooms)
-        observed_room_ids = set(r.room_id for r in observed_rooms)
+        observer_room_ids = yield self.store.get_rooms_for_user(
+            observer_user.to_string()
+        )
+        observed_room_ids = yield self.store.get_rooms_for_user(
+            observed_user.to_string()
+        )
 
         if observer_room_ids & observed_room_ids:
             defer.returnValue(True)
@@ -979,14 +997,18 @@ def should_notify(old_state, new_state):
     return False
 
 
-def format_user_presence_state(state, now):
+def format_user_presence_state(state, now, include_user_id=True):
     """Convert UserPresenceState to a format that can be sent down to clients
     and to other servers.
+
+    The "user_id" is optional so that this function can be used to format presence
+    updates for client /sync responses and for federation /send requests.
     """
     content = {
         "presence": state.state,
-        "user_id": state.user_id,
     }
+    if include_user_id:
+        content["user_id"] = state.user_id
     if state.last_active_ts:
         content["last_active_ago"] = now - state.last_active_ts
     if state.status_msg and state.state != PresenceState.OFFLINE:
@@ -1025,7 +1047,6 @@ class PresenceEventSource(object):
         # sending down the rare duplicate is not a concern.
 
         with Measure(self.clock, "presence.get_new_events"):
-            user_id = user.to_string()
             if from_key is not None:
                 from_key = int(from_key)
 
@@ -1034,18 +1055,7 @@ class PresenceEventSource(object):
 
             max_token = self.store.get_current_presence_token()
 
-            plist = yield self.store.get_presence_list_accepted(user.localpart)
-            users_interested_in = set(row["observed_user_id"] for row in plist)
-            users_interested_in.add(user_id)  # So that we receive our own presence
-
-            users_who_share_room = yield self.store.get_users_who_share_room_with_user(
-                user_id
-            )
-            users_interested_in.update(users_who_share_room)
-
-            if explicit_room_id:
-                user_ids = yield self.store.get_users_in_room(explicit_room_id)
-                users_interested_in.update(user_ids)
+            users_interested_in = yield self._get_interested_in(user, explicit_room_id)
 
             user_ids_changed = set()
             changed = None
@@ -1073,16 +1083,13 @@ class PresenceEventSource(object):
 
             updates = yield presence.current_state_for_users(user_ids_changed)
 
-        now = self.clock.time_msec()
-
-        defer.returnValue(([
-            {
-                "type": "m.presence",
-                "content": format_user_presence_state(s, now),
-            }
-            for s in updates.values()
-            if include_offline or s.state != PresenceState.OFFLINE
-        ], max_token))
+        if include_offline:
+            defer.returnValue((updates.values(), max_token))
+        else:
+            defer.returnValue(([
+                s for s in updates.itervalues()
+                if s.state != PresenceState.OFFLINE
+            ], max_token))
 
     def get_current_key(self):
         return self.store.get_current_presence_token()
@@ -1090,6 +1097,31 @@ class PresenceEventSource(object):
     def get_pagination_rows(self, user, pagination_config, key):
         return self.get_new_events(user, from_key=None, include_offline=False)
 
+    @cachedInlineCallbacks(num_args=2, cache_context=True)
+    def _get_interested_in(self, user, explicit_room_id, cache_context):
+        """Returns the set of users that the given user should see presence
+        updates for
+        """
+        user_id = user.to_string()
+        plist = yield self.store.get_presence_list_accepted(
+            user.localpart, on_invalidate=cache_context.invalidate,
+        )
+        users_interested_in = set(row["observed_user_id"] for row in plist)
+        users_interested_in.add(user_id)  # So that we receive our own presence
+
+        users_who_share_room = yield self.store.get_users_who_share_room_with_user(
+            user_id, on_invalidate=cache_context.invalidate,
+        )
+        users_interested_in.update(users_who_share_room)
+
+        if explicit_room_id:
+            user_ids = yield self.store.get_users_in_room(
+                explicit_room_id, on_invalidate=cache_context.invalidate,
+            )
+            users_interested_in.update(user_ids)
+
+        defer.returnValue(users_interested_in)
+
 
 def handle_timeouts(user_states, is_mine_fn, syncing_user_ids, now):
     """Checks the presence of users that have timed out and updates as
@@ -1157,7 +1189,10 @@ def handle_timeout(state, is_mine, syncing_user_ids, now):
         # If there are have been no sync for a while (and none ongoing),
         # set presence to offline
         if user_id not in syncing_user_ids:
-            if now - state.last_user_sync_ts > SYNC_ONLINE_TIMEOUT:
+            # If the user has done something recently but hasn't synced,
+            # don't set them as offline.
+            sync_or_active = max(state.last_user_sync_ts, state.last_active_ts)
+            if now - sync_or_active > SYNC_ONLINE_TIMEOUT:
                 state = state.copy_and_replace(
                     state=PresenceState.OFFLINE,
                     status_msg=None,
@@ -1255,3 +1290,66 @@ def handle_update(prev_state, new_state, is_mine, wheel_timer, now):
         persist_and_notify = True
 
     return new_state, persist_and_notify, federation_ping
+
+
+@defer.inlineCallbacks
+def get_interested_parties(store, states):
+    """Given a list of states return which entities (rooms, users)
+    are interested in the given states.
+
+    Args:
+        states (list(UserPresenceState))
+
+    Returns:
+        2-tuple: `(room_ids_to_states, users_to_states)`,
+        with each item being a dict of `entity_name` -> `[UserPresenceState]`
+    """
+    room_ids_to_states = {}
+    users_to_states = {}
+    for state in states:
+        room_ids = yield store.get_rooms_for_user(state.user_id)
+        for room_id in room_ids:
+            room_ids_to_states.setdefault(room_id, []).append(state)
+
+        plist = yield store.get_presence_list_observers_accepted(state.user_id)
+        for u in plist:
+            users_to_states.setdefault(u, []).append(state)
+
+        # Always notify self
+        users_to_states.setdefault(state.user_id, []).append(state)
+
+    defer.returnValue((room_ids_to_states, users_to_states))
+
+
+@defer.inlineCallbacks
+def get_interested_remotes(store, states, state_handler):
+    """Given a list of presence states figure out which remote servers
+    should be sent which.
+
+    All the presence states should be for local users only.
+
+    Args:
+        store (DataStore)
+        states (list(UserPresenceState))
+
+    Returns:
+        Deferred list of ([destinations], [UserPresenceState]), where for
+        each row the list of UserPresenceState should be sent to each
+        destination
+    """
+    hosts_and_states = []
+
+    # First we look up the rooms each user is in (as well as any explicit
+    # subscriptions), then for each distinct room we look up the remote
+    # hosts in those rooms.
+    room_ids_to_states, users_to_states = yield get_interested_parties(store, states)
+
+    for room_id, states in room_ids_to_states.iteritems():
+        hosts = yield state_handler.get_current_hosts_in_room(room_id)
+        hosts_and_states.append((hosts, states))
+
+    for user_id, states in users_to_states.iteritems():
+        host = get_domain_from_id(user_id)
+        hosts_and_states.append(([host], states))
+
+    defer.returnValue(hosts_and_states)

synapse/handlers/profile.py

@@ -52,7 +52,8 @@ class ProfileHandler(BaseHandler):
                     args={
                         "user_id": target_user.to_string(),
                         "field": "displayname",
-                    }
+                    },
+                    ignore_backoff=True,
                 )
             except CodeMessageException as e:
                 if e.code != 404:
@@ -99,7 +100,8 @@ class ProfileHandler(BaseHandler):
                     args={
                         "user_id": target_user.to_string(),
                         "field": "avatar_url",
-                    }
+                    },
+                    ignore_backoff=True,
                 )
             except CodeMessageException as e:
                 if e.code != 404:
@@ -154,13 +156,13 @@ class ProfileHandler(BaseHandler):
         if not self.hs.is_mine(user):
             return
 
-        self.ratelimit(requester)
+        yield self.ratelimit(requester)
 
-        joins = yield self.store.get_rooms_for_user(
+        room_ids = yield self.store.get_rooms_for_user(
             user.to_string(),
         )
 
-        for j in joins:
+        for room_id in room_ids:
             handler = self.hs.get_handlers().room_member_handler
             try:
                 # Assume the user isn't a guest because we don't let guests set
@@ -171,12 +173,12 @@ class ProfileHandler(BaseHandler):
                 yield handler.update_membership(
                     requester,
                     user,
-                    j.room_id,
+                    room_id,
                     "join",  # We treat a profile update like a join.
                     ratelimit=False,  # Try to hide that these events aren't atomic.
                 )
             except Exception as e:
                 logger.warn(
                     "Failed to update join event for room %s - %s",
-                    j.room_id, str(e.message)
+                    room_id, str(e.message)
                 )

synapse/handlers/read_marker.py

@@ -0,0 +1,64 @@
+# -*- coding: utf-8 -*-
+# Copyright 2017 Vector Creations Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from ._base import BaseHandler
+
+from twisted.internet import defer
+
+from synapse.util.async import Linearizer
+
+import logging
+logger = logging.getLogger(__name__)
+
+
+class ReadMarkerHandler(BaseHandler):
+    def __init__(self, hs):
+        super(ReadMarkerHandler, self).__init__(hs)
+        self.server_name = hs.config.server_name
+        self.store = hs.get_datastore()
+        self.read_marker_linearizer = Linearizer(name="read_marker")
+        self.notifier = hs.get_notifier()
+
+    @defer.inlineCallbacks
+    def received_client_read_marker(self, room_id, user_id, event_id):
+        """Updates the read marker for a given user in a given room if the event ID given
+        is ahead in the stream relative to the current read marker.
+
+        This uses a notifier to indicate that account data should be sent down /sync if
+        the read marker has changed.
+        """
+
+        with (yield self.read_marker_linearizer.queue((room_id, user_id))):
+            account_data = yield self.store.get_account_data_for_room(user_id, room_id)
+
+            existing_read_marker = account_data.get("m.fully_read", None)
+
+            should_update = True
+
+            if existing_read_marker:
+                # Only update if the new marker is ahead in the stream
+                should_update = yield self.store.is_event_after(
+                    event_id,
+                    existing_read_marker['event_id']
+                )
+
+            if should_update:
+                content = {
+                    "event_id": event_id
+                }
+                max_id = yield self.store.add_account_data_to_room(
+                    user_id, room_id, "m.fully_read", content
+                )
+                self.notifier.on_new_event("account_data_key", max_id, users=[user_id])

synapse/handlers/receipts.py

@@ -210,10 +210,9 @@ class ReceiptEventSource(object):
         else:
             from_key = None
 
-        rooms = yield self.store.get_rooms_for_user(user.to_string())
-        rooms = [room.room_id for room in rooms]
+        room_ids = yield self.store.get_rooms_for_user(user.to_string())
         events = yield self.store.get_linearized_receipts_for_rooms(
-            rooms,
+            room_ids,
             from_key=from_key,
             to_key=to_key,
         )

synapse/handlers/register.py

@@ -54,6 +54,13 @@ class RegistrationHandler(BaseHandler):
                 Codes.INVALID_USERNAME
             )
 
+        if not localpart:
+            raise SynapseError(
+                400,
+                "User ID cannot be empty",
+                Codes.INVALID_USERNAME
+            )
+
         if localpart[0] == '_':
             raise SynapseError(
                 400,

synapse/handlers/room.py

@@ -75,7 +75,7 @@ class RoomCreationHandler(BaseHandler):
         """
         user_id = requester.user.to_string()
 
-        self.ratelimit(requester)
+        yield self.ratelimit(requester)
 
         if "room_alias_name" in config:
             for wchar in string.whitespace:

synapse/handlers/room_list.py

@@ -21,6 +21,7 @@ from synapse.api.constants import (
     EventTypes, JoinRules,
 )
 from synapse.util.async import concurrently_execute
+from synapse.util.caches.descriptors import cachedInlineCallbacks
 from synapse.util.caches.response_cache import ResponseCache
 from synapse.types import ThirdPartyInstanceID
 
@@ -62,6 +63,10 @@ class RoomListHandler(BaseHandler):
                 appservice and network id to use an appservice specific one.
                 Setting to None returns all public rooms across all lists.
         """
+        logger.info(
+            "Getting public room list: limit=%r, since=%r, search=%r, network=%r",
+            limit, since_token, bool(search_filter), network_tuple,
+        )
         if search_filter:
             # We explicitly don't bother caching searches or requests for
             # appservice specific lists.
@@ -91,7 +96,6 @@ class RoomListHandler(BaseHandler):
 
         rooms_to_order_value = {}
         rooms_to_num_joined = {}
-        rooms_to_latest_event_ids = {}
 
         newly_visible = []
         newly_unpublished = []
@@ -116,19 +120,26 @@ class RoomListHandler(BaseHandler):
 
         @defer.inlineCallbacks
         def get_order_for_room(room_id):
-            latest_event_ids = rooms_to_latest_event_ids.get(room_id, None)
-            if not latest_event_ids:
+            # Most of the rooms won't have changed between the since token and
+            # now (especially if the since token is "now"). So, we can ask what
+            # the current users are in a room (that will hit a cache) and then
+            # check if the room has changed since the since token. (We have to
+            # do it in that order to avoid races).
+            # If things have changed then fall back to getting the current state
+            # at the since token.
+            joined_users = yield self.store.get_users_in_room(room_id)
+            if self.store.has_room_changed_since(room_id, stream_token):
                 latest_event_ids = yield self.store.get_forward_extremeties_for_room(
                     room_id, stream_token
                 )
-                rooms_to_latest_event_ids[room_id] = latest_event_ids
 
-            if not latest_event_ids:
-                return
+                if not latest_event_ids:
+                    return
+
+                joined_users = yield self.state_handler.get_current_user_in_room(
+                    room_id, latest_event_ids,
+                )
 
-            joined_users = yield self.state_handler.get_current_user_in_room(
-                room_id, latest_event_ids,
-            )
             num_joined_users = len(joined_users)
             rooms_to_num_joined[room_id] = num_joined_users
 
@@ -165,19 +176,19 @@ class RoomListHandler(BaseHandler):
                 rooms_to_scan = rooms_to_scan[:since_token.current_limit]
                 rooms_to_scan.reverse()
 
-        # Actually generate the entries. _generate_room_entry will append to
+        # Actually generate the entries. _append_room_entry_to_chunk will append to
         # chunk but will stop if len(chunk) > limit
         chunk = []
         if limit and not search_filter:
             step = limit + 1
             for i in xrange(0, len(rooms_to_scan), step):
                 # We iterate here because the vast majority of cases we'll stop
-                # at first iteration, but occaisonally _generate_room_entry
+                # at first iteration, but occaisonally _append_room_entry_to_chunk
                 # won't append to the chunk and so we need to loop again.
                 # We don't want to scan over the entire range either as that
                 # would potentially waste a lot of work.
                 yield concurrently_execute(
-                    lambda r: self._generate_room_entry(
+                    lambda r: self._append_room_entry_to_chunk(
                         r, rooms_to_num_joined[r],
                         chunk, limit, search_filter
                     ),
@@ -187,7 +198,7 @@ class RoomListHandler(BaseHandler):
                     break
         else:
             yield concurrently_execute(
-                lambda r: self._generate_room_entry(
+                lambda r: self._append_room_entry_to_chunk(
                     r, rooms_to_num_joined[r],
                     chunk, limit, search_filter
                 ),
@@ -256,21 +267,35 @@ class RoomListHandler(BaseHandler):
         defer.returnValue(results)
 
     @defer.inlineCallbacks
-    def _generate_room_entry(self, room_id, num_joined_users, chunk, limit,
-                             search_filter):
+    def _append_room_entry_to_chunk(self, room_id, num_joined_users, chunk, limit,
+                                    search_filter):
+        """Generate the entry for a room in the public room list and append it
+        to the `chunk` if it matches the search filter
+        """
         if limit and len(chunk) > limit + 1:
             # We've already got enough, so lets just drop it.
             return
 
+        result = yield self._generate_room_entry(room_id, num_joined_users)
+
+        if result and _matches_room_entry(result, search_filter):
+            chunk.append(result)
+
+    @cachedInlineCallbacks(num_args=1, cache_context=True)
+    def _generate_room_entry(self, room_id, num_joined_users, cache_context):
+        """Returns the entry for a room
+        """
         result = {
             "room_id": room_id,
             "num_joined_members": num_joined_users,
         }
 
-        current_state_ids = yield self.state_handler.get_current_state_ids(room_id)
+        current_state_ids = yield self.store.get_current_state_ids(
+            room_id, on_invalidate=cache_context.invalidate,
+        )
 
         event_map = yield self.store.get_events([
-            event_id for key, event_id in current_state_ids.items()
+            event_id for key, event_id in current_state_ids.iteritems()
             if key[0] in (
                 EventTypes.JoinRules,
                 EventTypes.Name,
@@ -294,7 +319,9 @@ class RoomListHandler(BaseHandler):
             if join_rule and join_rule != JoinRules.PUBLIC:
                 defer.returnValue(None)
 
-        aliases = yield self.store.get_aliases_for_room(room_id)
+        aliases = yield self.store.get_aliases_for_room(
+            room_id, on_invalidate=cache_context.invalidate
+        )
         if aliases:
             result["aliases"] = aliases
 
@@ -334,8 +361,7 @@ class RoomListHandler(BaseHandler):
             if avatar_url:
                 result["avatar_url"] = avatar_url
 
-        if _matches_room_entry(result, search_filter):
-            chunk.append(result)
+        defer.returnValue(result)
 
     @defer.inlineCallbacks
     def get_remote_public_room_list(self, server_name, limit=None, since_token=None,

synapse/handlers/room_member.py

@@ -70,6 +70,7 @@ class RoomMemberHandler(BaseHandler):
             content["kind"] = "guest"
 
         event, context = yield msg_handler.create_event(
+            requester,
             {
                 "type": EventTypes.Member,
                 "content": content,
@@ -139,13 +140,6 @@ class RoomMemberHandler(BaseHandler):
         )
         yield user_joined_room(self.distributor, user, room_id)
 
-    def reject_remote_invite(self, user_id, room_id, remote_room_hosts):
-        return self.hs.get_handlers().federation_handler.do_remotely_reject_invite(
-            remote_room_hosts,
-            room_id,
-            user_id
-        )
-
     @defer.inlineCallbacks
     def update_membership(
             self,
@@ -286,13 +280,21 @@ class RoomMemberHandler(BaseHandler):
                 else:
                     # send the rejection to the inviter's HS.
                     remote_room_hosts = remote_room_hosts + [inviter.domain]
-
+                    fed_handler = self.hs.get_handlers().federation_handler
                     try:
-                        ret = yield self.reject_remote_invite(
-                            target.to_string(), room_id, remote_room_hosts
+                        ret = yield fed_handler.do_remotely_reject_invite(
+                            remote_room_hosts,
+                            room_id,
+                            target.to_string(),
                         )
                         defer.returnValue(ret)
-                    except SynapseError as e:
+                    except Exception as e:
+                        # if we were unable to reject the exception, just mark
+                        # it as rejected on our end and plough ahead.
+                        #
+                        # The 'except' clause is very broad, but we need to
+                        # capture everything from DNS failures upwards
+                        #
                         logger.warn("Failed to reject invite: %s", e)
 
                         yield self.store.locally_reject_invite(
@@ -719,7 +721,9 @@ class RoomMemberHandler(BaseHandler):
         )
         membership = member.membership if member else None
 
-        if membership is not None and membership != Membership.LEAVE:
+        if membership is not None and membership not in [
+            Membership.LEAVE, Membership.BAN
+        ]:
             raise SynapseError(400, "User %s in room %s" % (
                 user_id, room_id
             ))
@@ -735,10 +739,11 @@ class RoomMemberHandler(BaseHandler):
         if len(current_state_ids) == 1 and create_event_id:
             defer.returnValue(self.hs.is_mine_id(create_event_id))
 
-        for (etype, state_key), event_id in current_state_ids.items():
+        for etype, state_key in current_state_ids:
             if etype != EventTypes.Member or not self.hs.is_mine_id(state_key):
                 continue
 
+            event_id = current_state_ids[(etype, state_key)]
             event = yield self.store.get_event(event_id, allow_none=True)
             if not event:
                 continue

synapse/handlers/sync.py

@@ -20,6 +20,7 @@ from synapse.util.metrics import Measure, measure_func
 from synapse.util.caches.response_cache import ResponseCache
 from synapse.push.clientformat import format_push_rules_for_user
 from synapse.visibility import filter_events_for_client
+from synapse.types import RoomStreamToken
 
 from twisted.internet import defer
 
@@ -225,8 +226,7 @@ class SyncHandler(object):
         with Measure(self.clock, "ephemeral_by_room"):
             typing_key = since_token.typing_key if since_token else "0"
 
-            rooms = yield self.store.get_rooms_for_user(sync_config.user.to_string())
-            room_ids = [room.room_id for room in rooms]
+            room_ids = yield self.store.get_rooms_for_user(sync_config.user.to_string())
 
             typing_source = self.event_sources.sources["typing"]
             typing, typing_key = yield typing_source.get_new_events(
@@ -568,16 +568,15 @@ class SyncHandler(object):
         since_token = sync_result_builder.since_token
 
         if since_token and since_token.device_list_key:
-            rooms = yield self.store.get_rooms_for_user(user_id)
-            room_ids = set(r.room_id for r in rooms)
+            room_ids = yield self.store.get_rooms_for_user(user_id)
 
             user_ids_changed = set()
             changed = yield self.store.get_user_whose_devices_changed(
                 since_token.device_list_key
             )
             for other_user_id in changed:
-                other_rooms = yield self.store.get_rooms_for_user(other_user_id)
-                if room_ids.intersection(e.room_id for e in other_rooms):
+                other_room_ids = yield self.store.get_rooms_for_user(other_user_id)
+                if room_ids.intersection(other_room_ids):
                     user_ids_changed.add(other_user_id)
 
             defer.returnValue(user_ids_changed)
@@ -609,14 +608,14 @@ class SyncHandler(object):
             deleted = yield self.store.delete_messages_for_device(
                 user_id, device_id, since_stream_id
             )
-            logger.info("Deleted %d to-device messages up to %d",
-                        deleted, since_stream_id)
+            logger.debug("Deleted %d to-device messages up to %d",
+                         deleted, since_stream_id)
 
             messages, stream_id = yield self.store.get_new_messages_for_device(
                 user_id, device_id, since_stream_id, now_token.to_device_key
             )
 
-            logger.info(
+            logger.debug(
                 "Returning %d to-device messages between %d and %d (current token: %d)",
                 len(messages), since_stream_id, stream_id, now_token.to_device_key
             )
@@ -721,14 +720,14 @@ class SyncHandler(object):
             extra_users_ids.update(users)
         extra_users_ids.discard(user.to_string())
 
-        states = yield self.presence_handler.get_states(
-            extra_users_ids,
-            as_event=True,
-        )
-        presence.extend(states)
+        if extra_users_ids:
+            states = yield self.presence_handler.get_states(
+                extra_users_ids,
+            )
+            presence.extend(states)
 
-        # Deduplicate the presence entries so that there's at most one per user
-        presence = {p["content"]["user_id"]: p for p in presence}.values()
+            # Deduplicate the presence entries so that there's at most one per user
+            presence = {p.user_id: p for p in presence}.values()
 
         presence = sync_config.filter_collection.filter_presence(
             presence
@@ -765,6 +764,21 @@ class SyncHandler(object):
             )
             sync_result_builder.now_token = now_token
 
+        # We check up front if anything has changed, if it hasn't then there is
+        # no point in going futher.
+        since_token = sync_result_builder.since_token
+        if not sync_result_builder.full_state:
+            if since_token and not ephemeral_by_room and not account_data_by_room:
+                have_changed = yield self._have_rooms_changed(sync_result_builder)
+                if not have_changed:
+                    tags_by_room = yield self.store.get_updated_tags(
+                        user_id,
+                        since_token.account_data_key,
+                    )
+                    if not tags_by_room:
+                        logger.debug("no-oping sync")
+                        defer.returnValue(([], []))
+
         ignored_account_data = yield self.store.get_global_account_data_by_type_for_user(
             "m.ignored_user_list", user_id=user_id,
         )
@@ -774,13 +788,12 @@ class SyncHandler(object):
         else:
             ignored_users = frozenset()
 
-        if sync_result_builder.since_token:
+        if since_token:
             res = yield self._get_rooms_changed(sync_result_builder, ignored_users)
             room_entries, invited, newly_joined_rooms = res
 
             tags_by_room = yield self.store.get_updated_tags(
-                user_id,
-                sync_result_builder.since_token.account_data_key,
+                user_id, since_token.account_data_key,
             )
         else:
             res = yield self._get_all_rooms(sync_result_builder, ignored_users)
@@ -805,7 +818,7 @@ class SyncHandler(object):
 
         # Now we want to get any newly joined users
         newly_joined_users = set()
-        if sync_result_builder.since_token:
+        if since_token:
             for joined_sync in sync_result_builder.joined:
                 it = itertools.chain(
                     joined_sync.timeline.events, joined_sync.state.values()
@@ -817,6 +830,38 @@ class SyncHandler(object):
 
         defer.returnValue((newly_joined_rooms, newly_joined_users))
 
+    @defer.inlineCallbacks
+    def _have_rooms_changed(self, sync_result_builder):
+        """Returns whether there may be any new events that should be sent down
+        the sync. Returns True if there are.
+        """
+        user_id = sync_result_builder.sync_config.user.to_string()
+        since_token = sync_result_builder.since_token
+        now_token = sync_result_builder.now_token
+
+        assert since_token
+
+        # Get a list of membership change events that have happened.
+        rooms_changed = yield self.store.get_membership_changes_for_user(
+            user_id, since_token.room_key, now_token.room_key
+        )
+
+        if rooms_changed:
+            defer.returnValue(True)
+
+        app_service = self.store.get_app_service_by_user_id(user_id)
+        if app_service:
+            rooms = yield self.store.get_app_service_rooms(app_service)
+            joined_room_ids = set(r.room_id for r in rooms)
+        else:
+            joined_room_ids = yield self.store.get_rooms_for_user(user_id)
+
+        stream_id = RoomStreamToken.parse_stream_token(since_token.room_key).stream
+        for room_id in joined_room_ids:
+            if self.store.has_room_changed_since(room_id, stream_id):
+                defer.returnValue(True)
+        defer.returnValue(False)
+
     @defer.inlineCallbacks
     def _get_rooms_changed(self, sync_result_builder, ignored_users):
         """Gets the the changes that have happened since the last sync.
@@ -841,8 +886,7 @@ class SyncHandler(object):
             rooms = yield self.store.get_app_service_rooms(app_service)
             joined_room_ids = set(r.room_id for r in rooms)
         else:
-            rooms = yield self.store.get_rooms_for_user(user_id)
-            joined_room_ids = set(r.room_id for r in rooms)
+            joined_room_ids = yield self.store.get_rooms_for_user(user_id)
 
         # Get a list of membership change events that have happened.
         rooms_changed = yield self.store.get_membership_changes_for_user(

synapse/handlers/typing.py

@@ -24,7 +24,6 @@ from synapse.types import UserID, get_domain_from_id
 import logging
 
 from collections import namedtuple
-import ujson as json
 
 logger = logging.getLogger(__name__)
 
@@ -288,11 +287,13 @@ class TypingHandler(object):
         for room_id, serial in self._room_serials.items():
             if last_id < serial and serial <= current_id:
                 typing = self._room_typing[room_id]
-                typing_bytes = json.dumps(list(typing), ensure_ascii=False)
-                rows.append((serial, room_id, typing_bytes))
+                rows.append((serial, room_id, list(typing)))
         rows.sort()
         return rows
 
+    def get_current_token(self):
+        return self._latest_room_serial
+
 
 class TypingNotificationEventSource(object):
     def __init__(self, hs):

synapse/http/client.py

@@ -16,9 +16,10 @@ from OpenSSL import SSL
 from OpenSSL.SSL import VERIFY_NONE
 
 from synapse.api.errors import (
-    CodeMessageException, SynapseError, Codes,
+    CodeMessageException, MatrixCodeMessageException, SynapseError, Codes,
 )
 from synapse.util.logcontext import preserve_context_over_fn
+from synapse.util import logcontext
 import synapse.metrics
 from synapse.http.endpoint import SpiderEndpoint
 
@@ -72,39 +73,45 @@ class SimpleHttpClient(object):
             contextFactory=hs.get_http_client_context_factory()
         )
         self.user_agent = hs.version_string
+        self.clock = hs.get_clock()
         if hs.config.user_agent_suffix:
             self.user_agent = "%s %s" % (self.user_agent, hs.config.user_agent_suffix,)
 
+    @defer.inlineCallbacks
     def request(self, method, uri, *args, **kwargs):
         # A small wrapper around self.agent.request() so we can easily attach
         # counters to it
         outgoing_requests_counter.inc(method)
-        d = preserve_context_over_fn(
-            self.agent.request,
-            method, uri, *args, **kwargs
-        )
+
+        def send_request():
+            request_deferred = self.agent.request(
+                method, uri, *args, **kwargs
+            )
+
+            return self.clock.time_bound_deferred(
+                request_deferred,
+                time_out=60,
+            )
 
         logger.info("Sending request %s %s", method, uri)
 
-        def _cb(response):
+        try:
+            with logcontext.PreserveLoggingContext():
+                response = yield send_request()
+
             incoming_responses_counter.inc(method, response.code)
             logger.info(
                 "Received response to  %s %s: %s",
                 method, uri, response.code
             )
-            return response
-
-        def _eb(failure):
+            defer.returnValue(response)
+        except Exception as e:
             incoming_responses_counter.inc(method, "ERR")
             logger.info(
                 "Error sending request to  %s %s: %s %s",
-                method, uri, failure.type, failure.getErrorMessage()
+                method, uri, type(e).__name__, e.message
             )
-            return failure
-
-        d.addCallbacks(_cb, _eb)
-
-        return d
+            raise e
 
     @defer.inlineCallbacks
     def post_urlencoded_get_json(self, uri, args={}):
@@ -145,6 +152,11 @@ class SimpleHttpClient(object):
 
         body = yield preserve_context_over_fn(readBody, response)
 
+        if 200 <= response.code < 300:
+            defer.returnValue(json.loads(body))
+        else:
+            raise self._exceptionFromFailedRequest(response, body)
+
         defer.returnValue(json.loads(body))
 
     @defer.inlineCallbacks
@@ -164,8 +176,11 @@ class SimpleHttpClient(object):
             On a non-2xx HTTP response. The response body will be used as the
             error message.
         """
-        body = yield self.get_raw(uri, args)
-        defer.returnValue(json.loads(body))
+        try:
+            body = yield self.get_raw(uri, args)
+            defer.returnValue(json.loads(body))
+        except CodeMessageException as e:
+            raise self._exceptionFromFailedRequest(e.code, e.msg)
 
     @defer.inlineCallbacks
     def put_json(self, uri, json_body, args={}):
@@ -246,6 +261,15 @@ class SimpleHttpClient(object):
         else:
             raise CodeMessageException(response.code, body)
 
+    def _exceptionFromFailedRequest(self, response, body):
+        try:
+            jsonBody = json.loads(body)
+            errcode = jsonBody['errcode']
+            error = jsonBody['error']
+            return MatrixCodeMessageException(response.code, error, errcode)
+        except (ValueError, KeyError):
+            return CodeMessageException(response.code, body)
+
     # XXX: FIXME: This is horribly copy-pasted from matrixfederationclient.
     # The two should be factored out.
 

synapse/http/matrixfederationclient.py

@@ -12,8 +12,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-
-
+import synapse.util.retryutils
 from twisted.internet import defer, reactor, protocol
 from twisted.internet.error import DNSLookupError
 from twisted.web.client import readBody, HTTPConnectionPool, Agent
@@ -22,7 +21,7 @@ from twisted.web._newclient import ResponseDone
 
 from synapse.http.endpoint import matrix_federation_endpoint
 from synapse.util.async import sleep
-from synapse.util.logcontext import preserve_context_over_fn
+from synapse.util import logcontext
 import synapse.metrics
 
 from canonicaljson import encode_canonical_json
@@ -94,6 +93,7 @@ class MatrixFederationHttpClient(object):
             reactor, MatrixFederationEndpointFactory(hs), pool=pool
         )
         self.clock = hs.get_clock()
+        self._store = hs.get_datastore()
         self.version_string = hs.version_string
         self._next_id = 1
 
@@ -103,123 +103,154 @@ class MatrixFederationHttpClient(object):
         )
 
     @defer.inlineCallbacks
-    def _create_request(self, destination, method, path_bytes,
-                        body_callback, headers_dict={}, param_bytes=b"",
-                        query_bytes=b"", retry_on_dns_fail=True,
-                        timeout=None, long_retries=False):
-        """ Creates and sends a request to the given url
+    def _request(self, destination, method, path,
+                 body_callback, headers_dict={}, param_bytes=b"",
+                 query_bytes=b"", retry_on_dns_fail=True,
+                 timeout=None, long_retries=False,
+                 ignore_backoff=False,
+                 backoff_on_404=False):
+        """ Creates and sends a request to the given server
+        Args:
+            destination (str): The remote server to send the HTTP request to.
+            method (str): HTTP method
+            path (str): The HTTP path
+            ignore_backoff (bool): true to ignore the historical backoff data
+                and try the request anyway.
+            backoff_on_404 (bool): Back off if we get a 404
+
+        Returns:
+            Deferred: resolves with the http response object on success.
+
+            Fails with ``HTTPRequestException``: if we get an HTTP response
+                code >= 300.
+            Fails with ``NotRetryingDestination`` if we are not yet ready
+                to retry this server.
+            (May also fail with plenty of other Exceptions for things like DNS
+                failures, connection failures, SSL failures.)
         """
-        headers_dict[b"User-Agent"] = [self.version_string]
-        headers_dict[b"Host"] = [destination]
-
-        url_bytes = self._create_url(
-            destination, path_bytes, param_bytes, query_bytes
+        limiter = yield synapse.util.retryutils.get_retry_limiter(
+            destination,
+            self.clock,
+            self._store,
+            backoff_on_404=backoff_on_404,
+            ignore_backoff=ignore_backoff,
         )
 
-        txn_id = "%s-O-%s" % (method, self._next_id)
-        self._next_id = (self._next_id + 1) % (sys.maxint - 1)
+        destination = destination.encode("ascii")
+        path_bytes = path.encode("ascii")
+        with limiter:
+            headers_dict[b"User-Agent"] = [self.version_string]
+            headers_dict[b"Host"] = [destination]
 
-        outbound_logger.info(
-            "{%s} [%s] Sending request: %s %s",
-            txn_id, destination, method, url_bytes
-        )
+            url_bytes = self._create_url(
+                destination, path_bytes, param_bytes, query_bytes
+            )
 
-        # XXX: Would be much nicer to retry only at the transaction-layer
-        # (once we have reliable transactions in place)
-        if long_retries:
-            retries_left = MAX_LONG_RETRIES
-        else:
-            retries_left = MAX_SHORT_RETRIES
+            txn_id = "%s-O-%s" % (method, self._next_id)
+            self._next_id = (self._next_id + 1) % (sys.maxint - 1)
 
-        http_url_bytes = urlparse.urlunparse(
-            ("", "", path_bytes, param_bytes, query_bytes, "")
-        )
+            outbound_logger.info(
+                "{%s} [%s] Sending request: %s %s",
+                txn_id, destination, method, url_bytes
+            )
 
-        log_result = None
-        try:
-            while True:
-                producer = None
-                if body_callback:
-                    producer = body_callback(method, http_url_bytes, headers_dict)
+            # XXX: Would be much nicer to retry only at the transaction-layer
+            # (once we have reliable transactions in place)
+            if long_retries:
+                retries_left = MAX_LONG_RETRIES
+            else:
+                retries_left = MAX_SHORT_RETRIES
 
-                try:
-                    def send_request():
-                        request_deferred = preserve_context_over_fn(
-                            self.agent.request,
+            http_url_bytes = urlparse.urlunparse(
+                ("", "", path_bytes, param_bytes, query_bytes, "")
+            )
+
+            log_result = None
+            try:
+                while True:
+                    producer = None
+                    if body_callback:
+                        producer = body_callback(method, http_url_bytes, headers_dict)
+
+                    try:
+                        def send_request():
+                            request_deferred = self.agent.request(
+                                method,
+                                url_bytes,
+                                Headers(headers_dict),
+                                producer
+                            )
+
+                            return self.clock.time_bound_deferred(
+                                request_deferred,
+                                time_out=timeout / 1000. if timeout else 60,
+                            )
+
+                        with logcontext.PreserveLoggingContext():
+                            response = yield send_request()
+
+                        log_result = "%d %s" % (response.code, response.phrase,)
+                        break
+                    except Exception as e:
+                        if not retry_on_dns_fail and isinstance(e, DNSLookupError):
+                            logger.warn(
+                                "DNS Lookup failed to %s with %s",
+                                destination,
+                                e
+                            )
+                            log_result = "DNS Lookup failed to %s with %s" % (
+                                destination, e
+                            )
+                            raise
+
+                        logger.warn(
+                            "{%s} Sending request failed to %s: %s %s: %s - %s",
+                            txn_id,
+                            destination,
                             method,
                             url_bytes,
-                            Headers(headers_dict),
-                            producer
+                            type(e).__name__,
+                            _flatten_response_never_received(e),
                         )
 
-                        return self.clock.time_bound_deferred(
-                            request_deferred,
-                            time_out=timeout / 1000. if timeout else 60,
+                        log_result = "%s - %s" % (
+                            type(e).__name__, _flatten_response_never_received(e),
                         )
 
-                    response = yield preserve_context_over_fn(send_request)
+                        if retries_left and not timeout:
+                            if long_retries:
+                                delay = 4 ** (MAX_LONG_RETRIES + 1 - retries_left)
+                                delay = min(delay, 60)
+                                delay *= random.uniform(0.8, 1.4)
+                            else:
+                                delay = 0.5 * 2 ** (MAX_SHORT_RETRIES - retries_left)
+                                delay = min(delay, 2)
+                                delay *= random.uniform(0.8, 1.4)
 
-                    log_result = "%d %s" % (response.code, response.phrase,)
-                    break
-                except Exception as e:
-                    if not retry_on_dns_fail and isinstance(e, DNSLookupError):
-                        logger.warn(
-                            "DNS Lookup failed to %s with %s",
-                            destination,
-                            e
-                        )
-                        log_result = "DNS Lookup failed to %s with %s" % (
-                            destination, e
-                        )
-                        raise
-
-                    logger.warn(
-                        "{%s} Sending request failed to %s: %s %s: %s - %s",
-                        txn_id,
-                        destination,
-                        method,
-                        url_bytes,
-                        type(e).__name__,
-                        _flatten_response_never_received(e),
-                    )
-
-                    log_result = "%s - %s" % (
-                        type(e).__name__, _flatten_response_never_received(e),
-                    )
-
-                    if retries_left and not timeout:
-                        if long_retries:
-                            delay = 4 ** (MAX_LONG_RETRIES + 1 - retries_left)
-                            delay = min(delay, 60)
-                            delay *= random.uniform(0.8, 1.4)
+                            yield sleep(delay)
+                            retries_left -= 1
                         else:
-                            delay = 0.5 * 2 ** (MAX_SHORT_RETRIES - retries_left)
-                            delay = min(delay, 2)
-                            delay *= random.uniform(0.8, 1.4)
+                            raise
+            finally:
+                outbound_logger.info(
+                    "{%s} [%s] Result: %s",
+                    txn_id,
+                    destination,
+                    log_result,
+                )
 
-                        yield sleep(delay)
-                        retries_left -= 1
-                    else:
-                        raise
-        finally:
-            outbound_logger.info(
-                "{%s} [%s] Result: %s",
-                txn_id,
-                destination,
-                log_result,
-            )
+            if 200 <= response.code < 300:
+                pass
+            else:
+                # :'(
+                # Update transactions table?
+                with logcontext.PreserveLoggingContext():
+                    body = yield readBody(response)
+                raise HttpResponseException(
+                    response.code, response.phrase, body
+                )
 
-        if 200 <= response.code < 300:
-            pass
-        else:
-            # :'(
-            # Update transactions table?
-            body = yield preserve_context_over_fn(readBody, response)
-            raise HttpResponseException(
-                response.code, response.phrase, body
-            )
-
-        defer.returnValue(response)
+            defer.returnValue(response)
 
     def sign_request(self, destination, method, url_bytes, headers_dict,
                      content=None):
@@ -248,7 +279,9 @@ class MatrixFederationHttpClient(object):
 
     @defer.inlineCallbacks
     def put_json(self, destination, path, data={}, json_data_callback=None,
-                 long_retries=False, timeout=None):
+                 long_retries=False, timeout=None,
+                 ignore_backoff=False,
+                 backoff_on_404=False):
         """ Sends the specifed json data using PUT
 
         Args:
@@ -263,11 +296,21 @@ class MatrixFederationHttpClient(object):
                 retry for a short or long time.
             timeout(int): How long to try (in ms) the destination for before
                 giving up. None indicates no timeout.
+            ignore_backoff (bool): true to ignore the historical backoff data
+                and try the request anyway.
+            backoff_on_404 (bool): True if we should count a 404 response as
+                a failure of the server (and should therefore back off future
+                requests)
 
         Returns:
             Deferred: Succeeds when we get a 2xx HTTP response. The result
-            will be the decoded JSON body. On a 4xx or 5xx error response a
-            CodeMessageException is raised.
+            will be the decoded JSON body.
+
+            Fails with ``HTTPRequestException`` if we get an HTTP response
+            code >= 300.
+
+            Fails with ``NotRetryingDestination`` if we are not yet ready
+            to retry this server.
         """
 
         if not json_data_callback:
@@ -282,26 +325,29 @@ class MatrixFederationHttpClient(object):
             producer = _JsonProducer(json_data)
             return producer
 
-        response = yield self._create_request(
-            destination.encode("ascii"),
+        response = yield self._request(
+            destination,
             "PUT",
-            path.encode("ascii"),
+            path,
             body_callback=body_callback,
             headers_dict={"Content-Type": ["application/json"]},
             long_retries=long_retries,
             timeout=timeout,
+            ignore_backoff=ignore_backoff,
+            backoff_on_404=backoff_on_404,
         )
 
         if 200 <= response.code < 300:
             # We need to update the transactions table to say it was sent?
             check_content_type_is_json(response.headers)
 
-        body = yield preserve_context_over_fn(readBody, response)
+        with logcontext.PreserveLoggingContext():
+            body = yield readBody(response)
         defer.returnValue(json.loads(body))
 
     @defer.inlineCallbacks
     def post_json(self, destination, path, data={}, long_retries=False,
-                  timeout=None):
+                  timeout=None, ignore_backoff=False):
         """ Sends the specifed json data using POST
 
         Args:
@@ -314,11 +360,17 @@ class MatrixFederationHttpClient(object):
                 retry for a short or long time.
             timeout(int): How long to try (in ms) the destination for before
                 giving up. None indicates no timeout.
-
+            ignore_backoff (bool): true to ignore the historical backoff data and
+                try the request anyway.
         Returns:
             Deferred: Succeeds when we get a 2xx HTTP response. The result
-            will be the decoded JSON body. On a 4xx or 5xx error response a
-            CodeMessageException is raised.
+            will be the decoded JSON body.
+
+            Fails with ``HTTPRequestException`` if we get an HTTP response
+            code >= 300.
+
+            Fails with ``NotRetryingDestination`` if we are not yet ready
+            to retry this server.
         """
 
         def body_callback(method, url_bytes, headers_dict):
@@ -327,27 +379,29 @@ class MatrixFederationHttpClient(object):
             )
             return _JsonProducer(data)
 
-        response = yield self._create_request(
-            destination.encode("ascii"),
+        response = yield self._request(
+            destination,
             "POST",
-            path.encode("ascii"),
+            path,
             body_callback=body_callback,
             headers_dict={"Content-Type": ["application/json"]},
             long_retries=long_retries,
             timeout=timeout,
+            ignore_backoff=ignore_backoff,
         )
 
         if 200 <= response.code < 300:
             # We need to update the transactions table to say it was sent?
             check_content_type_is_json(response.headers)
 
-        body = yield preserve_context_over_fn(readBody, response)
+        with logcontext.PreserveLoggingContext():
+            body = yield readBody(response)
 
         defer.returnValue(json.loads(body))
 
     @defer.inlineCallbacks
     def get_json(self, destination, path, args={}, retry_on_dns_fail=True,
-                 timeout=None):
+                 timeout=None, ignore_backoff=False):
         """ GETs some json from the given host homeserver and path
 
         Args:
@@ -359,11 +413,17 @@ class MatrixFederationHttpClient(object):
             timeout (int): How long to try (in ms) the destination for before
                 giving up. None indicates no timeout and that the request will
                 be retried.
+            ignore_backoff (bool): true to ignore the historical backoff data
+                and try the request anyway.
         Returns:
-            Deferred: Succeeds when we get *any* HTTP response.
+            Deferred: Succeeds when we get a 2xx HTTP response. The result
+            will be the decoded JSON body.
 
-            The result of the deferred is a tuple of `(code, response)`,
-            where `response` is a dict representing the decoded JSON body.
+            Fails with ``HTTPRequestException`` if we get an HTTP response
+            code >= 300.
+
+            Fails with ``NotRetryingDestination`` if we are not yet ready
+            to retry this server.
         """
         logger.debug("get_json args: %s", args)
 
@@ -380,36 +440,47 @@ class MatrixFederationHttpClient(object):
             self.sign_request(destination, method, url_bytes, headers_dict)
             return None
 
-        response = yield self._create_request(
-            destination.encode("ascii"),
+        response = yield self._request(
+            destination,
             "GET",
-            path.encode("ascii"),
+            path,
             query_bytes=query_bytes,
             body_callback=body_callback,
             retry_on_dns_fail=retry_on_dns_fail,
             timeout=timeout,
+            ignore_backoff=ignore_backoff,
         )
 
         if 200 <= response.code < 300:
             # We need to update the transactions table to say it was sent?
             check_content_type_is_json(response.headers)
 
-        body = yield preserve_context_over_fn(readBody, response)
+        with logcontext.PreserveLoggingContext():
+            body = yield readBody(response)
 
         defer.returnValue(json.loads(body))
 
     @defer.inlineCallbacks
     def get_file(self, destination, path, output_stream, args={},
-                 retry_on_dns_fail=True, max_size=None):
+                 retry_on_dns_fail=True, max_size=None,
+                 ignore_backoff=False):
         """GETs a file from a given homeserver
         Args:
             destination (str): The remote server to send the HTTP request to.
             path (str): The HTTP path to GET.
             output_stream (file): File to write the response body to.
             args (dict): Optional dictionary used to create the query string.
+            ignore_backoff (bool): true to ignore the historical backoff data
+                and try the request anyway.
         Returns:
-            A (int,dict) tuple of the file length and a dict of the response
-            headers.
+            Deferred: resolves with an (int,dict) tuple of the file length and
+            a dict of the response headers.
+
+            Fails with ``HTTPRequestException`` if we get an HTTP response code
+            >= 300
+
+            Fails with ``NotRetryingDestination`` if we are not yet ready
+            to retry this server.
         """
 
         encoded_args = {}
@@ -419,28 +490,29 @@ class MatrixFederationHttpClient(object):
             encoded_args[k] = [v.encode("UTF-8") for v in vs]
 
         query_bytes = urllib.urlencode(encoded_args, True)
-        logger.debug("Query bytes: %s Retry DNS: %s", args, retry_on_dns_fail)
+        logger.debug("Query bytes: %s Retry DNS: %s", query_bytes, retry_on_dns_fail)
 
         def body_callback(method, url_bytes, headers_dict):
             self.sign_request(destination, method, url_bytes, headers_dict)
             return None
 
-        response = yield self._create_request(
-            destination.encode("ascii"),
+        response = yield self._request(
+            destination,
             "GET",
-            path.encode("ascii"),
+            path,
             query_bytes=query_bytes,
             body_callback=body_callback,
-            retry_on_dns_fail=retry_on_dns_fail
+            retry_on_dns_fail=retry_on_dns_fail,
+            ignore_backoff=ignore_backoff,
         )
 
         headers = dict(response.headers.getAllRawHeaders())
 
         try:
-            length = yield preserve_context_over_fn(
-                _readBodyToFile,
-                response, output_stream, max_size
-            )
+            with logcontext.PreserveLoggingContext():
+                length = yield _readBodyToFile(
+                    response, output_stream, max_size
+                )
         except:
             logger.exception("Failed to download body")
             raise

synapse/http/servlet.py

@@ -192,6 +192,16 @@ def parse_json_object_from_request(request):
     return content
 
 
+def assert_params_in_request(body, required):
+    absent = []
+    for k in required:
+        if k not in body:
+            absent.append(k)
+
+    if len(absent) > 0:
+        raise SynapseError(400, "Missing params: %r" % absent, Codes.MISSING_PARAM)
+
+
 class RestServlet(object):
 
     """ A Synapse REST Servlet.

synapse/notifier.py

@@ -16,6 +16,7 @@
 from twisted.internet import defer
 from synapse.api.constants import EventTypes, Membership
 from synapse.api.errors import AuthError
+from synapse.handlers.presence import format_user_presence_state
 
 from synapse.util import DeferredTimedOutError
 from synapse.util.logutils import log_function
@@ -37,6 +38,10 @@ metrics = synapse.metrics.get_metrics_for(__name__)
 
 notified_events_counter = metrics.register_counter("notified_events")
 
+users_woken_by_stream_counter = metrics.register_counter(
+    "users_woken_by_stream", labels=["stream"]
+)
+
 
 # TODO(paul): Should be shared somewhere
 def count(func, l):
@@ -73,6 +78,13 @@ class _NotifierUserStream(object):
         self.user_id = user_id
         self.rooms = set(rooms)
         self.current_token = current_token
+
+        # The last token for which we should wake up any streams that have a
+        # token that comes before it. This gets updated everytime we get poked.
+        # We start it at the current token since if we get any streams
+        # that have a token from before we have no idea whether they should be
+        # woken up or not, so lets just wake them up.
+        self.last_notified_token = current_token
         self.last_notified_ms = time_now_ms
 
         with PreserveLoggingContext():
@@ -89,9 +101,12 @@ class _NotifierUserStream(object):
         self.current_token = self.current_token.copy_and_advance(
             stream_key, stream_id
         )
+        self.last_notified_token = self.current_token
         self.last_notified_ms = time_now_ms
         noify_deferred = self.notify_deferred
 
+        users_woken_by_stream_counter.inc(stream_key)
+
         with PreserveLoggingContext():
             self.notify_deferred = ObservableDeferred(defer.Deferred())
             noify_deferred.callback(self.current_token)
@@ -113,8 +128,14 @@ class _NotifierUserStream(object):
     def new_listener(self, token):
         """Returns a deferred that is resolved when there is a new token
         greater than the given token.
+
+        Args:
+            token: The token from which we are streaming from, i.e. we shouldn't
+                notify for things that happened before this.
         """
-        if self.current_token.is_after(token):
+        # Immediately wake up stream if something has already since happened
+        # since their last token.
+        if self.last_notified_token.is_after(token):
             return _NotificationListener(defer.succeed(self.current_token))
         else:
             return _NotificationListener(self.notify_deferred.observe())
@@ -142,6 +163,8 @@ class Notifier(object):
         self.store = hs.get_datastore()
         self.pending_new_room_events = []
 
+        self.replication_callbacks = []
+
         self.clock = hs.get_clock()
         self.appservice_handler = hs.get_application_service_handler()
 
@@ -181,7 +204,12 @@ class Notifier(object):
             lambda: len(self.user_to_user_stream),
         )
 
-    @preserve_fn
+    def add_replication_callback(self, cb):
+        """Add a callback that will be called when some new data is available.
+        Callback is not given any arguments.
+        """
+        self.replication_callbacks.append(cb)
+
     def on_new_room_event(self, event, room_stream_id, max_room_stream_id,
                           extra_users=[]):
         """ Used by handlers to inform the notifier something has happened
@@ -195,15 +223,13 @@ class Notifier(object):
         until all previous events have been persisted before notifying
         the client streams.
         """
-        with PreserveLoggingContext():
-            self.pending_new_room_events.append((
-                room_stream_id, event, extra_users
-            ))
-            self._notify_pending_new_room_events(max_room_stream_id)
+        self.pending_new_room_events.append((
+            room_stream_id, event, extra_users
+        ))
+        self._notify_pending_new_room_events(max_room_stream_id)
 
-            self.notify_replication()
+        self.notify_replication()
 
-    @preserve_fn
     def _notify_pending_new_room_events(self, max_room_stream_id):
         """Notify for the room events that were queued waiting for a previous
         event to be persisted.
@@ -221,14 +247,16 @@ class Notifier(object):
             else:
                 self._on_new_room_event(event, room_stream_id, extra_users)
 
-    @preserve_fn
     def _on_new_room_event(self, event, room_stream_id, extra_users=[]):
         """Notify any user streams that are interested in this room event"""
         # poke any interested application service.
-        self.appservice_handler.notify_interested_services(room_stream_id)
+        preserve_fn(self.appservice_handler.notify_interested_services)(
+            room_stream_id)
 
         if self.federation_sender:
-            self.federation_sender.notify_new_events(room_stream_id)
+            preserve_fn(self.federation_sender.notify_new_events)(
+                room_stream_id
+            )
 
         if event.type == EventTypes.Member and event.membership == Membership.JOIN:
             self._user_joined_room(event.state_key, event.room_id)
@@ -239,7 +267,6 @@ class Notifier(object):
             rooms=[event.room_id],
         )
 
-    @preserve_fn
     def on_new_event(self, stream_key, new_token, users=[], rooms=[]):
         """ Used to inform listeners that something has happend event wise.
 
@@ -266,7 +293,6 @@ class Notifier(object):
 
                 self.notify_replication()
 
-    @preserve_fn
     def on_new_replication_data(self):
         """Used to inform replication listeners that something has happend
         without waking up any of the normal user event streams"""
@@ -283,8 +309,7 @@ class Notifier(object):
         if user_stream is None:
             current_token = yield self.event_sources.get_current_token()
             if room_ids is None:
-                rooms = yield self.store.get_rooms_for_user(user_id)
-                room_ids = [room.room_id for room in rooms]
+                room_ids = yield self.store.get_rooms_for_user(user_id)
             user_stream = _NotifierUserStream(
                 user_id=user_id,
                 rooms=room_ids,
@@ -294,40 +319,44 @@ class Notifier(object):
             self._register_with_keys(user_stream)
 
         result = None
+        prev_token = from_token
         if timeout:
             end_time = self.clock.time_msec() + timeout
 
-            prev_token = from_token
             while not result:
                 try:
-                    current_token = user_stream.current_token
-
-                    result = yield callback(prev_token, current_token)
-                    if result:
-                        break
-
                     now = self.clock.time_msec()
                     if end_time <= now:
                         break
 
                     # Now we wait for the _NotifierUserStream to be told there
                     # is a new token.
-                    # We need to supply the token we supplied to callback so
-                    # that we don't miss any current_token updates.
-                    prev_token = current_token
                     listener = user_stream.new_listener(prev_token)
                     with PreserveLoggingContext():
                         yield self.clock.time_bound_deferred(
                             listener.deferred,
                             time_out=(end_time - now) / 1000.
                         )
+
+                    current_token = user_stream.current_token
+
+                    result = yield callback(prev_token, current_token)
+                    if result:
+                        break
+
+                    # Update the prev_token to the current_token since nothing
+                    # has happened between the old prev_token and the current_token
+                    prev_token = current_token
                 except DeferredTimedOutError:
                     break
                 except defer.CancelledError:
                     break
-        else:
+
+        if result is None:
+            # This happened if there was no timeout or if the timeout had
+            # already expired.
             current_token = user_stream.current_token
-            result = yield callback(from_token, current_token)
+            result = yield callback(prev_token, current_token)
 
         defer.returnValue(result)
 
@@ -388,6 +417,15 @@ class Notifier(object):
                         new_events,
                         is_peeking=is_peeking,
                     )
+                elif name == "presence":
+                    now = self.clock.time_msec()
+                    new_events[:] = [
+                        {
+                            "type": "m.presence",
+                            "content": format_user_presence_state(event, now),
+                        }
+                        for event in new_events
+                    ]
 
                 events.extend(new_events)
                 end_token = end_token.copy_and_replace(keyname, new_key)
@@ -420,8 +458,7 @@ class Notifier(object):
 
     @defer.inlineCallbacks
     def _get_room_ids(self, user, explicit_room_id):
-        joined_rooms = yield self.store.get_rooms_for_user(user.to_string())
-        joined_room_ids = map(lambda r: r.room_id, joined_rooms)
+        joined_room_ids = yield self.store.get_rooms_for_user(user.to_string())
         if explicit_room_id:
             if explicit_room_id in joined_room_ids:
                 defer.returnValue(([explicit_room_id], True))
@@ -478,6 +515,9 @@ class Notifier(object):
             self.replication_deferred = ObservableDeferred(defer.Deferred())
             deferred.callback(None)
 
+        for cb in self.replication_callbacks:
+            preserve_fn(cb)()
+
     @defer.inlineCallbacks
     def wait_for_replication(self, callback, timeout):
         """Wait for an event to happen.

synapse/push/bulk_push_rule_evaluator.py

@@ -87,7 +87,11 @@ class BulkPushRuleEvaluator:
         condition_cache = {}
 
         for uid, rules in self.rules_by_user.items():
-            display_name = room_members.get(uid, {}).get("display_name", None)
+            display_name = None
+            profile_info = room_members.get(uid)
+            if profile_info:
+                display_name = profile_info.display_name
+
             if not display_name:
                 # Handle the case where we are pushing a membership event to
                 # that user, as they might not be already joined.

synapse/push/emailpusher.py

@@ -218,7 +218,8 @@ class EmailPusher(object):
         )
 
     def seconds_until(self, ts_msec):
-        return (ts_msec - self.clock.time_msec()) / 1000
+        secs = (ts_msec - self.clock.time_msec()) / 1000
+        return max(secs, 0)
 
     def get_room_throttle_ms(self, room_id):
         if room_id in self.throttle_params:

synapse/push/mailer.py

@@ -139,7 +139,7 @@ class Mailer(object):
 
         @defer.inlineCallbacks
         def _fetch_room_state(room_id):
-            room_state = yield self.state_handler.get_current_state_ids(room_id)
+            room_state = yield self.store.get_current_state_ids(room_id)
             state_by_room[room_id] = room_state
 
         # Run at most 3 of these at once: sync does 10 at a time but email
@@ -200,7 +200,11 @@ class Mailer(object):
         yield sendmail(
             self.hs.config.email_smtp_host,
             raw_from, raw_to, multipart_msg.as_string(),
-            port=self.hs.config.email_smtp_port
+            port=self.hs.config.email_smtp_port,
+            requireAuthentication=self.hs.config.email_smtp_user is not None,
+            username=self.hs.config.email_smtp_user,
+            password=self.hs.config.email_smtp_pass,
+            requireTransportSecurity=self.hs.config.require_transport_security
         )
 
     @defer.inlineCallbacks

synapse/push/push_rule_evaluator.py

@@ -17,6 +17,7 @@ import logging
 import re
 
 from synapse.types import UserID
+from synapse.util.caches import CACHE_SIZE_FACTOR, register_cache
 from synapse.util.caches.lrucache import LruCache
 
 logger = logging.getLogger(__name__)
@@ -125,6 +126,11 @@ class PushRuleEvaluatorForEvent(object):
         return self._value_cache.get(dotted_key, None)
 
 
+# Caches (glob, word_boundary) -> regex for push. See _glob_matches
+regex_cache = LruCache(50000 * CACHE_SIZE_FACTOR)
+register_cache("regex_push_cache", regex_cache)
+
+
 def _glob_matches(glob, value, word_boundary=False):
     """Tests if value matches glob.
 
@@ -137,46 +143,63 @@ def _glob_matches(glob, value, word_boundary=False):
     Returns:
         bool
     """
+
     try:
-        if IS_GLOB.search(glob):
-            r = re.escape(glob)
-
-            r = r.replace(r'\*', '.*?')
-            r = r.replace(r'\?', '.')
-
-            # handle [abc], [a-z] and [!a-z] style ranges.
-            r = GLOB_REGEX.sub(
-                lambda x: (
-                    '[%s%s]' % (
-                        x.group(1) and '^' or '',
-                        x.group(2).replace(r'\\\-', '-')
-                    )
-                ),
-                r,
-            )
-            if word_boundary:
-                r = r"\b%s\b" % (r,)
-                r = _compile_regex(r)
-
-                return r.search(value)
-            else:
-                r = r + "$"
-                r = _compile_regex(r)
-
-                return r.match(value)
-        elif word_boundary:
-            r = re.escape(glob)
-            r = r"\b%s\b" % (r,)
-            r = _compile_regex(r)
-
-            return r.search(value)
-        else:
-            return value.lower() == glob.lower()
+        r = regex_cache.get((glob, word_boundary), None)
+        if not r:
+            r = _glob_to_re(glob, word_boundary)
+            regex_cache[(glob, word_boundary)] = r
+        return r.search(value)
     except re.error:
         logger.warn("Failed to parse glob to regex: %r", glob)
         return False
 
 
+def _glob_to_re(glob, word_boundary):
+    """Generates regex for a given glob.
+
+    Args:
+        glob (string)
+        word_boundary (bool): Whether to match against word boundaries or entire
+            string. Defaults to False.
+
+    Returns:
+        regex object
+    """
+    if IS_GLOB.search(glob):
+        r = re.escape(glob)
+
+        r = r.replace(r'\*', '.*?')
+        r = r.replace(r'\?', '.')
+
+        # handle [abc], [a-z] and [!a-z] style ranges.
+        r = GLOB_REGEX.sub(
+            lambda x: (
+                '[%s%s]' % (
+                    x.group(1) and '^' or '',
+                    x.group(2).replace(r'\\\-', '-')
+                )
+            ),
+            r,
+        )
+        if word_boundary:
+            r = r"\b%s\b" % (r,)
+
+            return re.compile(r, flags=re.IGNORECASE)
+        else:
+            r = "^" + r + "$"
+
+            return re.compile(r, flags=re.IGNORECASE)
+    elif word_boundary:
+        r = re.escape(glob)
+        r = r"\b%s\b" % (r,)
+
+        return re.compile(r, flags=re.IGNORECASE)
+    else:
+        r = "^" + re.escape(glob) + "$"
+        return re.compile(r, flags=re.IGNORECASE)
+
+
 def _flatten_dict(d, prefix=[], result={}):
     for key, value in d.items():
         if isinstance(value, basestring):
@@ -185,16 +208,3 @@ def _flatten_dict(d, prefix=[], result={}):
             _flatten_dict(value, prefix=(prefix + [key]), result=result)
 
     return result
-
-
-regex_cache = LruCache(5000)
-
-
-def _compile_regex(regex_str):
-    r = regex_cache.get(regex_str, None)
-    if r:
-        return r
-
-    r = re.compile(regex_str, flags=re.IGNORECASE)
-    regex_cache[regex_str] = r
-    return r

synapse/push/push_tools.py

@@ -17,15 +17,12 @@ from twisted.internet import defer
 from synapse.push.presentable_names import (
     calculate_room_name, name_from_member_event
 )
-from synapse.util.logcontext import preserve_fn, preserve_context_over_deferred
 
 
 @defer.inlineCallbacks
 def get_badge_count(store, user_id):
-    invites, joins = yield preserve_context_over_deferred(defer.gatherResults([
-        preserve_fn(store.get_invited_rooms_for_user)(user_id),
-        preserve_fn(store.get_rooms_for_user)(user_id),
-    ], consumeErrors=True))
+    invites = yield store.get_invited_rooms_for_user(user_id)
+    joins = yield store.get_rooms_for_user(user_id)
 
     my_receipts_by_room = yield store.get_receipts_for_user(
         user_id, "m.read",
@@ -33,13 +30,13 @@ def get_badge_count(store, user_id):
 
     badge = len(invites)
 
-    for r in joins:
-        if r.room_id in my_receipts_by_room:
-            last_unread_event_id = my_receipts_by_room[r.room_id]
+    for room_id in joins:
+        if room_id in my_receipts_by_room:
+            last_unread_event_id = my_receipts_by_room[room_id]
 
             notifs = yield (
                 store.get_unread_event_push_actions_by_room_for_user(
-                    r.room_id, user_id, last_unread_event_id
+                    room_id, user_id, last_unread_event_id
                 )
             )
             # return one badge count per conversation, as count per

synapse/python_dependencies.py

@@ -1,4 +1,5 @@
 # Copyright 2015, 2016 OpenMarket Ltd
+# Copyright 2017 Vector Creations Ltd
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -18,6 +19,7 @@ from distutils.version import LooseVersion
 logger = logging.getLogger(__name__)
 
 REQUIREMENTS = {
+    "jsonschema>=2.5.1": ["jsonschema>=2.5.1"],
     "frozendict>=0.4": ["frozendict"],
     "unpaddedbase64>=1.1.0": ["unpaddedbase64>=1.1.0"],
     "canonicaljson>=1.0.0": ["canonicaljson>=1.0.0"],
@@ -37,6 +39,7 @@ REQUIREMENTS = {
     "pysaml2>=3.0.0,<4.0.0": ["saml2>=3.0.0,<4.0.0"],
     "pymacaroons-pynacl": ["pymacaroons"],
     "msgpack-python>=0.3.0": ["msgpack"],
+    "phonenumbers>=8.2.0": ["phonenumbers"],
 }
 CONDITIONAL_REQUIREMENTS = {
     "web_client": {

synapse/replication/expire_cache.py

@@ -1,60 +0,0 @@
-# Copyright 2016 OpenMarket Ltd
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-from synapse.http.server import respond_with_json_bytes, request_handler
-from synapse.http.servlet import parse_json_object_from_request
-
-from twisted.web.resource import Resource
-from twisted.web.server import NOT_DONE_YET
-
-
-class ExpireCacheResource(Resource):
-    """
-    HTTP endpoint for expiring storage caches.
-
-    POST /_synapse/replication/expire_cache HTTP/1.1
-    Content-Type: application/json
-
-    {
-        "invalidate": [
-            {
-                "name": "func_name",
-                "keys": ["key1", "key2"]
-            }
-        ]
-    }
-    """
-
-    def __init__(self, hs):
-        Resource.__init__(self)  # Resource is old-style, so no super()
-
-        self.store = hs.get_datastore()
-        self.version_string = hs.version_string
-        self.clock = hs.get_clock()
-
-    def render_POST(self, request):
-        self._async_render_POST(request)
-        return NOT_DONE_YET
-
-    @request_handler()
-    def _async_render_POST(self, request):
-        content = parse_json_object_from_request(request)
-
-        for row in content["invalidate"]:
-            name = row["name"]
-            keys = tuple(row["keys"])
-
-            getattr(self.store, name).invalidate(keys)
-
-        respond_with_json_bytes(request, 200, "{}")

synapse/replication/presence_resource.py

@@ -1,59 +0,0 @@
-# Copyright 2016 OpenMarket Ltd
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-from synapse.http.server import respond_with_json_bytes, request_handler
-from synapse.http.servlet import parse_json_object_from_request
-
-from twisted.web.resource import Resource
-from twisted.web.server import NOT_DONE_YET
-from twisted.internet import defer
-
-
-class PresenceResource(Resource):
-    """
-    HTTP endpoint for marking users as syncing.
-
-    POST /_synapse/replication/presence HTTP/1.1
-    Content-Type: application/json
-
-    {
-        "process_id": "<process_id>",
-        "syncing_users": ["<user_id>"]
-    }
-    """
-
-    def __init__(self, hs):
-        Resource.__init__(self)  # Resource is old-style, so no super()
-
-        self.version_string = hs.version_string
-        self.clock = hs.get_clock()
-        self.presence_handler = hs.get_presence_handler()
-
-    def render_POST(self, request):
-        self._async_render_POST(request)
-        return NOT_DONE_YET
-
-    @request_handler()
-    @defer.inlineCallbacks
-    def _async_render_POST(self, request):
-        content = parse_json_object_from_request(request)
-
-        process_id = content["process_id"]
-        syncing_user_ids = content["syncing_users"]
-
-        yield self.presence_handler.update_external_syncs(
-            process_id, set(syncing_user_ids)
-        )
-
-        respond_with_json_bytes(request, 200, "{}")

synapse/replication/pusher_resource.py

@@ -1,54 +0,0 @@
-# Copyright 2016 OpenMarket Ltd
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-from synapse.http.server import respond_with_json_bytes, request_handler
-from synapse.http.servlet import parse_json_object_from_request
-
-from twisted.web.resource import Resource
-from twisted.web.server import NOT_DONE_YET
-from twisted.internet import defer
-
-
-class PusherResource(Resource):
-    """
-    HTTP endpoint for deleting rejected pushers
-    """
-
-    def __init__(self, hs):
-        Resource.__init__(self)  # Resource is old-style, so no super()
-
-        self.version_string = hs.version_string
-        self.store = hs.get_datastore()
-        self.notifier = hs.get_notifier()
-        self.clock = hs.get_clock()
-
-    def render_POST(self, request):
-        self._async_render_POST(request)
-        return NOT_DONE_YET
-
-    @request_handler()
-    @defer.inlineCallbacks
-    def _async_render_POST(self, request):
-        content = parse_json_object_from_request(request)
-
-        for remove in content["remove"]:
-            yield self.store.delete_pusher_by_app_id_pushkey_user_id(
-                remove["app_id"],
-                remove["push_key"],
-                remove["user_id"],
-            )
-
-        self.notifier.on_new_replication_data()
-
-        respond_with_json_bytes(request, 200, "{}")

synapse/replication/resource.py

@@ -1,576 +0,0 @@
-# -*- coding: utf-8 -*-
-# Copyright 2015 OpenMarket Ltd
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-from synapse.http.servlet import parse_integer, parse_string
-from synapse.http.server import request_handler, finish_request
-from synapse.replication.pusher_resource import PusherResource
-from synapse.replication.presence_resource import PresenceResource
-from synapse.replication.expire_cache import ExpireCacheResource
-from synapse.api.errors import SynapseError
-
-from twisted.web.resource import Resource
-from twisted.web.server import NOT_DONE_YET
-from twisted.internet import defer
-
-import ujson as json
-
-import collections
-import logging
-
-logger = logging.getLogger(__name__)
-
-REPLICATION_PREFIX = "/_synapse/replication"
-
-STREAM_NAMES = (
-    ("events",),
-    ("presence",),
-    ("typing",),
-    ("receipts",),
-    ("user_account_data", "room_account_data", "tag_account_data",),
-    ("backfill",),
-    ("push_rules",),
-    ("pushers",),
-    ("caches",),
-    ("to_device",),
-    ("public_rooms",),
-    ("federation",),
-    ("device_lists",),
-)
-
-
-class ReplicationResource(Resource):
-    """
-    HTTP endpoint for extracting data from synapse.
-
-    The streams of data returned by the endpoint are controlled by the
-    parameters given to the API. To return a given stream pass a query
-    parameter with a position in the stream to return data from or the
-    special value "-1" to return data from the start of the stream.
-
-    If there is no data for any of the supplied streams after the given
-    position then the request will block until there is data for one
-    of the streams. This allows clients to long-poll this API.
-
-    The possible streams are:
-
-    * "streams": A special stream returing the positions of other streams.
-    * "events": The new events seen on the server.
-    * "presence": Presence updates.
-    * "typing": Typing updates.
-    * "receipts": Receipt updates.
-    * "user_account_data": Top-level per user account data.
-    * "room_account_data: Per room per user account data.
-    * "tag_account_data": Per room per user tags.
-    * "backfill": Old events that have been backfilled from other servers.
-    * "push_rules": Per user changes to push rules.
-    * "pushers": Per user changes to their pushers.
-    * "caches": Cache invalidations.
-
-    The API takes two additional query parameters:
-
-    * "timeout": How long to wait before returning an empty response.
-    * "limit": The maximum number of rows to return for the selected streams.
-
-    The response is a JSON object with keys for each stream with updates. Under
-    each key is a JSON object with:
-
-    * "position": The current position of the stream.
-    * "field_names": The names of the fields in each row.
-    * "rows": The updates as an array of arrays.
-
-    There are a number of ways this API could be used:
-
-    1) To replicate the contents of the backing database to another database.
-    2) To be notified when the contents of a shared backing database changes.
-    3) To "tail" the activity happening on a server for debugging.
-
-    In the first case the client would track all of the streams and store it's
-    own copy of the data.
-
-    In the second case the client might theoretically just be able to follow
-    the "streams" stream to track where the other streams are. However in
-    practise it will probably need to get the contents of the streams in
-    order to expire the any in-memory caches. Whether it gets the contents
-    of the streams from this replication API or directly from the backing
-    store is a matter of taste.
-
-    In the third case the client would use the "streams" stream to find what
-    streams are available and their current positions. Then it can start
-    long-polling this replication API for new data on those streams.
-    """
-
-    def __init__(self, hs):
-        Resource.__init__(self)  # Resource is old-style, so no super()
-
-        self.version_string = hs.version_string
-        self.store = hs.get_datastore()
-        self.sources = hs.get_event_sources()
-        self.presence_handler = hs.get_presence_handler()
-        self.typing_handler = hs.get_typing_handler()
-        self.federation_sender = hs.get_federation_sender()
-        self.notifier = hs.notifier
-        self.clock = hs.get_clock()
-        self.config = hs.get_config()
-
-        self.putChild("remove_pushers", PusherResource(hs))
-        self.putChild("syncing_users", PresenceResource(hs))
-        self.putChild("expire_cache", ExpireCacheResource(hs))
-
-    def render_GET(self, request):
-        self._async_render_GET(request)
-        return NOT_DONE_YET
-
-    @defer.inlineCallbacks
-    def current_replication_token(self):
-        stream_token = yield self.sources.get_current_token()
-        backfill_token = yield self.store.get_current_backfill_token()
-        push_rules_token, room_stream_token = self.store.get_push_rules_stream_token()
-        pushers_token = self.store.get_pushers_stream_token()
-        caches_token = self.store.get_cache_stream_token()
-        public_rooms_token = self.store.get_current_public_room_stream_id()
-        federation_token = self.federation_sender.get_current_token()
-        device_list_token = self.store.get_device_stream_token()
-
-        defer.returnValue(_ReplicationToken(
-            room_stream_token,
-            int(stream_token.presence_key),
-            int(stream_token.typing_key),
-            int(stream_token.receipt_key),
-            int(stream_token.account_data_key),
-            backfill_token,
-            push_rules_token,
-            pushers_token,
-            0,  # State stream is no longer a thing
-            caches_token,
-            int(stream_token.to_device_key),
-            int(public_rooms_token),
-            int(federation_token),
-            int(device_list_token),
-        ))
-
-    @request_handler()
-    @defer.inlineCallbacks
-    def _async_render_GET(self, request):
-        limit = parse_integer(request, "limit", 100)
-        timeout = parse_integer(request, "timeout", 10 * 1000)
-
-        request.setHeader(b"Content-Type", b"application/json")
-
-        request_streams = {
-            name: parse_integer(request, name)
-            for names in STREAM_NAMES for name in names
-        }
-        request_streams["streams"] = parse_string(request, "streams")
-
-        federation_ack = parse_integer(request, "federation_ack", None)
-
-        def replicate():
-            return self.replicate(
-                request_streams, limit,
-                federation_ack=federation_ack
-            )
-
-        writer = yield self.notifier.wait_for_replication(replicate, timeout)
-        result = writer.finish()
-
-        for stream_name, stream_content in result.items():
-            logger.info(
-                "Replicating %d rows of %s from %s -> %s",
-                len(stream_content["rows"]),
-                stream_name,
-                request_streams.get(stream_name),
-                stream_content["position"],
-            )
-
-        request.write(json.dumps(result, ensure_ascii=False))
-        finish_request(request)
-
-    @defer.inlineCallbacks
-    def replicate(self, request_streams, limit, federation_ack=None):
-        writer = _Writer()
-        current_token = yield self.current_replication_token()
-        logger.debug("Replicating up to %r", current_token)
-
-        if limit == 0:
-            raise SynapseError(400, "Limit cannot be 0")
-
-        yield self.account_data(writer, current_token, limit, request_streams)
-        yield self.events(writer, current_token, limit, request_streams)
-        # TODO: implement limit
-        yield self.presence(writer, current_token, request_streams)
-        yield self.typing(writer, current_token, request_streams)
-        yield self.receipts(writer, current_token, limit, request_streams)
-        yield self.push_rules(writer, current_token, limit, request_streams)
-        yield self.pushers(writer, current_token, limit, request_streams)
-        yield self.caches(writer, current_token, limit, request_streams)
-        yield self.to_device(writer, current_token, limit, request_streams)
-        yield self.public_rooms(writer, current_token, limit, request_streams)
-        yield self.device_lists(writer, current_token, limit, request_streams)
-        self.federation(writer, current_token, limit, request_streams, federation_ack)
-        self.streams(writer, current_token, request_streams)
-
-        logger.debug("Replicated %d rows", writer.total)
-        defer.returnValue(writer)
-
-    def streams(self, writer, current_token, request_streams):
-        request_token = request_streams.get("streams")
-
-        streams = []
-
-        if request_token is not None:
-            if request_token == "-1":
-                for names, position in zip(STREAM_NAMES, current_token):
-                    streams.extend((name, position) for name in names)
-            else:
-                items = zip(
-                    STREAM_NAMES,
-                    current_token,
-                    _ReplicationToken(request_token)
-                )
-                for names, current_id, last_id in items:
-                    if last_id < current_id:
-                        streams.extend((name, current_id) for name in names)
-
-            if streams:
-                writer.write_header_and_rows(
-                    "streams", streams, ("name", "position"),
-                    position=str(current_token)
-                )
-
-    @defer.inlineCallbacks
-    def events(self, writer, current_token, limit, request_streams):
-        request_events = request_streams.get("events")
-        request_backfill = request_streams.get("backfill")
-
-        if request_events is not None or request_backfill is not None:
-            if request_events is None:
-                request_events = current_token.events
-            if request_backfill is None:
-                request_backfill = current_token.backfill
-
-            no_new_tokens = (
-                request_events == current_token.events
-                and request_backfill == current_token.backfill
-            )
-            if no_new_tokens:
-                return
-
-            res = yield self.store.get_all_new_events(
-                request_backfill, request_events,
-                current_token.backfill, current_token.events,
-                limit
-            )
-
-            upto_events_token = _position_from_rows(
-                res.new_forward_events, current_token.events
-            )
-
-            upto_backfill_token = _position_from_rows(
-                res.new_backfill_events, current_token.backfill
-            )
-
-            if request_events != upto_events_token:
-                writer.write_header_and_rows("events", res.new_forward_events, (
-                    "position", "internal", "json", "state_group"
-                ), position=upto_events_token)
-
-            if request_backfill != upto_backfill_token:
-                writer.write_header_and_rows("backfill", res.new_backfill_events, (
-                    "position", "internal", "json", "state_group",
-                ), position=upto_backfill_token)
-
-            writer.write_header_and_rows(
-                "forward_ex_outliers", res.forward_ex_outliers,
-                ("position", "event_id", "state_group"),
-            )
-            writer.write_header_and_rows(
-                "backward_ex_outliers", res.backward_ex_outliers,
-                ("position", "event_id", "state_group"),
-            )
-
-    @defer.inlineCallbacks
-    def presence(self, writer, current_token, request_streams):
-        current_position = current_token.presence
-
-        request_presence = request_streams.get("presence")
-
-        if request_presence is not None and request_presence != current_position:
-            presence_rows = yield self.presence_handler.get_all_presence_updates(
-                request_presence, current_position
-            )
-            upto_token = _position_from_rows(presence_rows, current_position)
-            writer.write_header_and_rows("presence", presence_rows, (
-                "position", "user_id", "state", "last_active_ts",
-                "last_federation_update_ts", "last_user_sync_ts",
-                "status_msg", "currently_active",
-            ), position=upto_token)
-
-    @defer.inlineCallbacks
-    def typing(self, writer, current_token, request_streams):
-        current_position = current_token.typing
-
-        request_typing = request_streams.get("typing")
-
-        if request_typing is not None and request_typing != current_position:
-            # If they have a higher token than current max, we can assume that
-            # they had been talking to a previous instance of the master. Since
-            # we reset the token on restart, the best (but hacky) thing we can
-            # do is to simply resend down all the typing notifications.
-            if request_typing > current_position:
-                request_typing = 0
-
-            typing_rows = yield self.typing_handler.get_all_typing_updates(
-                request_typing, current_position
-            )
-            upto_token = _position_from_rows(typing_rows, current_position)
-            writer.write_header_and_rows("typing", typing_rows, (
-                "position", "room_id", "typing"
-            ), position=upto_token)
-
-    @defer.inlineCallbacks
-    def receipts(self, writer, current_token, limit, request_streams):
-        current_position = current_token.receipts
-
-        request_receipts = request_streams.get("receipts")
-
-        if request_receipts is not None and request_receipts != current_position:
-            receipts_rows = yield self.store.get_all_updated_receipts(
-                request_receipts, current_position, limit
-            )
-            upto_token = _position_from_rows(receipts_rows, current_position)
-            writer.write_header_and_rows("receipts", receipts_rows, (
-                "position", "room_id", "receipt_type", "user_id", "event_id", "data"
-            ), position=upto_token)
-
-    @defer.inlineCallbacks
-    def account_data(self, writer, current_token, limit, request_streams):
-        current_position = current_token.account_data
-
-        user_account_data = request_streams.get("user_account_data")
-        room_account_data = request_streams.get("room_account_data")
-        tag_account_data = request_streams.get("tag_account_data")
-
-        if user_account_data is not None or room_account_data is not None:
-            if user_account_data is None:
-                user_account_data = current_position
-            if room_account_data is None:
-                room_account_data = current_position
-
-            no_new_tokens = (
-                user_account_data == current_position
-                and room_account_data == current_position
-            )
-            if no_new_tokens:
-                return
-
-            user_rows, room_rows = yield self.store.get_all_updated_account_data(
-                user_account_data, room_account_data, current_position, limit
-            )
-
-            upto_users_token = _position_from_rows(user_rows, current_position)
-            upto_rooms_token = _position_from_rows(room_rows, current_position)
-
-            writer.write_header_and_rows("user_account_data", user_rows, (
-                "position", "user_id", "type", "content"
-            ), position=upto_users_token)
-            writer.write_header_and_rows("room_account_data", room_rows, (
-                "position", "user_id", "room_id", "type", "content"
-            ), position=upto_rooms_token)
-
-        if tag_account_data is not None:
-            tag_rows = yield self.store.get_all_updated_tags(
-                tag_account_data, current_position, limit
-            )
-            upto_tag_token = _position_from_rows(tag_rows, current_position)
-            writer.write_header_and_rows("tag_account_data", tag_rows, (
-                "position", "user_id", "room_id", "tags"
-            ), position=upto_tag_token)
-
-    @defer.inlineCallbacks
-    def push_rules(self, writer, current_token, limit, request_streams):
-        current_position = current_token.push_rules
-
-        push_rules = request_streams.get("push_rules")
-
-        if push_rules is not None and push_rules != current_position:
-            rows = yield self.store.get_all_push_rule_updates(
-                push_rules, current_position, limit
-            )
-            upto_token = _position_from_rows(rows, current_position)
-            writer.write_header_and_rows("push_rules", rows, (
-                "position", "event_stream_ordering", "user_id", "rule_id", "op",
-                "priority_class", "priority", "conditions", "actions"
-            ), position=upto_token)
-
-    @defer.inlineCallbacks
-    def pushers(self, writer, current_token, limit, request_streams):
-        current_position = current_token.pushers
-
-        pushers = request_streams.get("pushers")
-
-        if pushers is not None and pushers != current_position:
-            updated, deleted = yield self.store.get_all_updated_pushers(
-                pushers, current_position, limit
-            )
-            upto_token = _position_from_rows(updated, current_position)
-            writer.write_header_and_rows("pushers", updated, (
-                "position", "user_id", "access_token", "profile_tag", "kind",
-                "app_id", "app_display_name", "device_display_name", "pushkey",
-                "ts", "lang", "data"
-            ), position=upto_token)
-            writer.write_header_and_rows("deleted_pushers", deleted, (
-                "position", "user_id", "app_id", "pushkey"
-            ), position=upto_token)
-
-    @defer.inlineCallbacks
-    def caches(self, writer, current_token, limit, request_streams):
-        current_position = current_token.caches
-
-        caches = request_streams.get("caches")
-
-        if caches is not None and caches != current_position:
-            updated_caches = yield self.store.get_all_updated_caches(
-                caches, current_position, limit
-            )
-            upto_token = _position_from_rows(updated_caches, current_position)
-            writer.write_header_and_rows("caches", updated_caches, (
-                "position", "cache_func", "keys", "invalidation_ts"
-            ), position=upto_token)
-
-    @defer.inlineCallbacks
-    def to_device(self, writer, current_token, limit, request_streams):
-        current_position = current_token.to_device
-
-        to_device = request_streams.get("to_device")
-
-        if to_device is not None and to_device != current_position:
-            to_device_rows = yield self.store.get_all_new_device_messages(
-                to_device, current_position, limit
-            )
-            upto_token = _position_from_rows(to_device_rows, current_position)
-            writer.write_header_and_rows("to_device", to_device_rows, (
-                "position", "user_id", "device_id", "message_json"
-            ), position=upto_token)
-
-    @defer.inlineCallbacks
-    def public_rooms(self, writer, current_token, limit, request_streams):
-        current_position = current_token.public_rooms
-
-        public_rooms = request_streams.get("public_rooms")
-
-        if public_rooms is not None and public_rooms != current_position:
-            public_rooms_rows = yield self.store.get_all_new_public_rooms(
-                public_rooms, current_position, limit
-            )
-            upto_token = _position_from_rows(public_rooms_rows, current_position)
-            writer.write_header_and_rows("public_rooms", public_rooms_rows, (
-                "position", "room_id", "visibility", "appservice_id", "network_id",
-            ), position=upto_token)
-
-    def federation(self, writer, current_token, limit, request_streams, federation_ack):
-        if self.config.send_federation:
-            return
-
-        current_position = current_token.federation
-
-        federation = request_streams.get("federation")
-
-        if federation is not None and federation != current_position:
-            federation_rows = self.federation_sender.get_replication_rows(
-                federation, limit, federation_ack=federation_ack,
-            )
-            upto_token = _position_from_rows(federation_rows, current_position)
-            writer.write_header_and_rows("federation", federation_rows, (
-                "position", "type", "content",
-            ), position=upto_token)
-
-    @defer.inlineCallbacks
-    def device_lists(self, writer, current_token, limit, request_streams):
-        current_position = current_token.device_lists
-
-        device_lists = request_streams.get("device_lists")
-
-        if device_lists is not None and device_lists != current_position:
-            changes = yield self.store.get_all_device_list_changes_for_remotes(
-                device_lists,
-            )
-            writer.write_header_and_rows("device_lists", changes, (
-                "position", "user_id", "destination",
-            ), position=current_position)
-
-
-class _Writer(object):
-    """Writes the streams as a JSON object as the response to the request"""
-    def __init__(self):
-        self.streams = {}
-        self.total = 0
-
-    def write_header_and_rows(self, name, rows, fields, position=None):
-        if position is None:
-            if rows:
-                position = rows[-1][0]
-            else:
-                return
-
-        self.streams[name] = {
-            "position": position if type(position) is int else str(position),
-            "field_names": fields,
-            "rows": rows,
-        }
-
-        self.total += len(rows)
-
-    def __nonzero__(self):
-        return bool(self.total)
-
-    def finish(self):
-        return self.streams
-
-
-class _ReplicationToken(collections.namedtuple("_ReplicationToken", (
-    "events", "presence", "typing", "receipts", "account_data", "backfill",
-    "push_rules", "pushers", "state", "caches", "to_device", "public_rooms",
-    "federation", "device_lists",
-))):
-    __slots__ = []
-
-    def __new__(cls, *args):
-        if len(args) == 1:
-            streams = [int(value) for value in args[0].split("_")]
-            if len(streams) < len(cls._fields):
-                streams.extend([0] * (len(cls._fields) - len(streams)))
-            return cls(*streams)
-        else:
-            return super(_ReplicationToken, cls).__new__(cls, *args)
-
-    def __str__(self):
-        return "_".join(str(value) for value in self)
-
-
-def _position_from_rows(rows, current_position):
-    """Calculates a position to return for a stream. Ideally we want to return the
-    position of the last row, as that will be the most correct. However, if there
-    are no rows we fall back to using the current position to stop us from
-    repeatedly hitting the storage layer unncessarily thinking there are updates.
-    (Not all advances of the token correspond to an actual update)
-
-    We can't just always return the current position, as we often limit the
-    number of rows we replicate, and so the stream may lag. The assumption is
-    that if the storage layer returns no new rows then we are not lagging and
-    we are at the `current_position`.
-    """
-    if rows:
-        return rows[-1][0]
-    return current_position

synapse/replication/slave/storage/_base.py

@@ -15,7 +15,6 @@
 
 from synapse.storage._base import SQLBaseStore
 from synapse.storage.engines import PostgresEngine
-from twisted.internet import defer
 
 from ._slaved_id_tracker import SlavedIdTracker
 
@@ -34,8 +33,7 @@ class BaseSlavedStore(SQLBaseStore):
         else:
             self._cache_id_gen = None
 
-        self.expire_cache_url = hs.config.worker_replication_url + "/expire_cache"
-        self.http_client = hs.get_simple_http_client()
+        self.hs = hs
 
     def stream_positions(self):
         pos = {}
@@ -43,33 +41,20 @@ class BaseSlavedStore(SQLBaseStore):
             pos["caches"] = self._cache_id_gen.get_current_token()
         return pos
 
-    def process_replication(self, result):
-        stream = result.get("caches")
-        if stream:
-            for row in stream["rows"]:
-                (
-                    position, cache_func, keys, invalidation_ts,
-                ) = row
-
+    def process_replication_rows(self, stream_name, token, rows):
+        if stream_name == "caches":
+            self._cache_id_gen.advance(token)
+            for row in rows:
                 try:
-                    getattr(self, cache_func).invalidate(tuple(keys))
+                    getattr(self, row.cache_func).invalidate(tuple(row.keys))
                 except AttributeError:
-                    logger.info("Got unexpected cache_func: %r", cache_func)
-            self._cache_id_gen.advance(int(stream["position"]))
-        return defer.succeed(None)
+                    # We probably haven't pulled in the cache in this worker,
+                    # which is fine.
+                    pass
 
     def _invalidate_cache_and_stream(self, txn, cache_func, keys):
         txn.call_after(cache_func.invalidate, keys)
         txn.call_after(self._send_invalidation_poke, cache_func, keys)
 
-    @defer.inlineCallbacks
     def _send_invalidation_poke(self, cache_func, keys):
-        try:
-            yield self.http_client.post_json_get_json(self.expire_cache_url, {
-                "invalidate": [{
-                    "name": cache_func.__name__,
-                    "keys": list(keys),
-                }]
-            })
-        except:
-            logger.exception("Failed to poke on expire_cache")
+        self.hs.get_tcp_replication().send_invalidate_cache(cache_func, keys)

synapse/replication/slave/storage/_slaved_id_tracker.py

@@ -27,4 +27,9 @@ class SlavedIdTracker(object):
         self._current = (max if self.step > 0 else min)(self._current, new_id)
 
     def get_current_token(self):
+        """
+
+        Returns:
+            int
+        """
         return self._current

synapse/replication/slave/storage/account_data.py

@@ -46,6 +46,12 @@ class SlavedAccountDataStore(BaseSlavedStore):
     )
 
     get_tags_for_user = TagsStore.__dict__["get_tags_for_user"]
+    get_tags_for_room = (
+        DataStore.get_tags_for_room.__func__
+    )
+    get_account_data_for_room = (
+        DataStore.get_account_data_for_room.__func__
+    )
 
     get_updated_tags = DataStore.get_updated_tags.__func__
     get_updated_account_data_for_user = (
@@ -63,38 +69,25 @@ class SlavedAccountDataStore(BaseSlavedStore):
         result["tag_account_data"] = position
         return result
 
-    def process_replication(self, result):
-        stream = result.get("user_account_data")
-        if stream:
-            self._account_data_id_gen.advance(int(stream["position"]))
-            for row in stream["rows"]:
-                position, user_id, data_type = row[:3]
-                self.get_global_account_data_by_type_for_user.invalidate(
-                    (data_type, user_id,)
-                )
-                self.get_account_data_for_user.invalidate((user_id,))
+    def process_replication_rows(self, stream_name, token, rows):
+        if stream_name == "tag_account_data":
+            self._account_data_id_gen.advance(token)
+            for row in rows:
+                self.get_tags_for_user.invalidate((row.user_id,))
                 self._account_data_stream_cache.entity_has_changed(
-                    user_id, position
+                    row.user_id, token
                 )
-
-        stream = result.get("room_account_data")
-        if stream:
-            self._account_data_id_gen.advance(int(stream["position"]))
-            for row in stream["rows"]:
-                position, user_id = row[:2]
-                self.get_account_data_for_user.invalidate((user_id,))
+        elif stream_name == "account_data":
+            self._account_data_id_gen.advance(token)
+            for row in rows:
+                if not row.room_id:
+                    self.get_global_account_data_by_type_for_user.invalidate(
+                        (row.data_type, row.user_id,)
+                    )
+                self.get_account_data_for_user.invalidate((row.user_id,))
                 self._account_data_stream_cache.entity_has_changed(
-                    user_id, position
+                    row.user_id, token
                 )
-
-        stream = result.get("tag_account_data")
-        if stream:
-            self._account_data_id_gen.advance(int(stream["position"]))
-            for row in stream["rows"]:
-                position, user_id = row[:2]
-                self.get_tags_for_user.invalidate((user_id,))
-                self._account_data_stream_cache.entity_has_changed(
-                    user_id, position
-                )
-
-        return super(SlavedAccountDataStore, self).process_replication(result)
+        return super(SlavedAccountDataStore, self).process_replication_rows(
+            stream_name, token, rows
+        )

synapse/replication/slave/storage/deviceinbox.py

@@ -17,6 +17,7 @@ from ._base import BaseSlavedStore
 from ._slaved_id_tracker import SlavedIdTracker
 from synapse.storage import DataStore
 from synapse.util.caches.stream_change_cache import StreamChangeCache
+from synapse.util.caches.expiringcache import ExpiringCache
 
 
 class SlavedDeviceInboxStore(BaseSlavedStore):
@@ -34,6 +35,13 @@ class SlavedDeviceInboxStore(BaseSlavedStore):
             self._device_inbox_id_gen.get_current_token()
         )
 
+        self._last_device_delete_cache = ExpiringCache(
+            cache_name="last_device_delete_cache",
+            clock=self._clock,
+            max_len=10000,
+            expiry_ms=30 * 60 * 1000,
+        )
+
     get_to_device_stream_token = DataStore.get_to_device_stream_token.__func__
     get_new_messages_for_device = DataStore.get_new_messages_for_device.__func__
     get_new_device_msgs_for_remote = DataStore.get_new_device_msgs_for_remote.__func__
@@ -45,21 +53,18 @@ class SlavedDeviceInboxStore(BaseSlavedStore):
         result["to_device"] = self._device_inbox_id_gen.get_current_token()
         return result
 
-    def process_replication(self, result):
-        stream = result.get("to_device")
-        if stream:
-            self._device_inbox_id_gen.advance(int(stream["position"]))
-            for row in stream["rows"]:
-                stream_id = row[0]
-                entity = row[1]
-
-                if entity.startswith("@"):
+    def process_replication_rows(self, stream_name, token, rows):
+        if stream_name == "to_device":
+            self._device_inbox_id_gen.advance(token)
+            for row in rows:
+                if row.entity.startswith("@"):
                     self._device_inbox_stream_cache.entity_has_changed(
-                        entity, stream_id
+                        row.entity, token
                     )
                 else:
                     self._device_federation_outbox_stream_cache.entity_has_changed(
-                        entity, stream_id
+                        row.entity, token
                     )
-
-        return super(SlavedDeviceInboxStore, self).process_replication(result)
+        return super(SlavedDeviceInboxStore, self).process_replication_rows(
+            stream_name, token, rows
+        )

synapse/replication/slave/storage/devices.py

@@ -51,22 +51,18 @@ class SlavedDeviceStore(BaseSlavedStore):
         result["device_lists"] = self._device_list_id_gen.get_current_token()
         return result
 
-    def process_replication(self, result):
-        stream = result.get("device_lists")
-        if stream:
-            self._device_list_id_gen.advance(int(stream["position"]))
-            for row in stream["rows"]:
-                stream_id = row[0]
-                user_id = row[1]
-                destination = row[2]
-
+    def process_replication_rows(self, stream_name, token, rows):
+        if stream_name == "device_lists":
+            self._device_list_id_gen.advance(token)
+            for row in rows:
                 self._device_list_stream_cache.entity_has_changed(
-                    user_id, stream_id
+                    row.user_id, token
                 )
 
-                if destination:
+                if row.destination:
                     self._device_list_federation_stream_cache.entity_has_changed(
-                        destination, stream_id
+                        row.destination, token
                     )
-
-        return super(SlavedDeviceStore, self).process_replication(result)
+        return super(SlavedDeviceStore, self).process_replication_rows(
+            stream_name, token, rows
+        )

synapse/replication/slave/storage/events.py

@@ -16,7 +16,6 @@ from ._base import BaseSlavedStore
 from ._slaved_id_tracker import SlavedIdTracker
 
 from synapse.api.constants import EventTypes
-from synapse.events import FrozenEvent
 from synapse.storage import DataStore
 from synapse.storage.roommember import RoomMemberStore
 from synapse.storage.event_federation import EventFederationStore
@@ -25,7 +24,6 @@ from synapse.storage.state import StateStore
 from synapse.storage.stream import StreamStore
 from synapse.util.caches.stream_change_cache import StreamChangeCache
 
-import ujson as json
 import logging
 
 
@@ -73,6 +71,7 @@ class SlavedEventStore(BaseSlavedStore):
     # to reach inside the __dict__ to extract them.
     get_rooms_for_user = RoomMemberStore.__dict__["get_rooms_for_user"]
     get_users_in_room = RoomMemberStore.__dict__["get_users_in_room"]
+    get_hosts_in_room = RoomMemberStore.__dict__["get_hosts_in_room"]
     get_users_who_share_room_with_user = (
         RoomMemberStore.__dict__["get_users_who_share_room_with_user"]
     )
@@ -85,6 +84,12 @@ class SlavedEventStore(BaseSlavedStore):
     get_unread_event_push_actions_by_room_for_user = (
         EventPushActionsStore.__dict__["get_unread_event_push_actions_by_room_for_user"]
     )
+    _get_unread_counts_by_receipt_txn = (
+        DataStore._get_unread_counts_by_receipt_txn.__func__
+    )
+    _get_unread_counts_by_pos_txn = (
+        DataStore._get_unread_counts_by_pos_txn.__func__
+    )
     _get_state_group_for_events = (
         StateStore.__dict__["_get_state_group_for_events"]
     )
@@ -97,12 +102,13 @@ class SlavedEventStore(BaseSlavedStore):
     _get_state_groups_from_groups_txn = (
         DataStore._get_state_groups_from_groups_txn.__func__
     )
-    _get_state_group_from_group = (
-        StateStore.__dict__["_get_state_group_from_group"]
-    )
     get_recent_event_ids_for_room = (
         StreamStore.__dict__["get_recent_event_ids_for_room"]
     )
+    get_current_state_ids = (
+        StateStore.__dict__["get_current_state_ids"]
+    )
+    has_room_changed_since = DataStore.has_room_changed_since.__func__
 
     get_unread_push_actions_for_user_in_range_for_http = (
         DataStore.get_unread_push_actions_for_user_in_range_for_http.__func__
@@ -138,6 +144,9 @@ class SlavedEventStore(BaseSlavedStore):
         RoomMemberStore.__dict__["_get_joined_users_from_context"]
     )
 
+    get_joined_hosts = DataStore.get_joined_hosts.__func__
+    _get_joined_hosts = RoomMemberStore.__dict__["_get_joined_hosts"]
+
     get_recent_events_for_room = DataStore.get_recent_events_for_room.__func__
     get_room_events_stream_for_rooms = (
         DataStore.get_room_events_stream_for_rooms.__func__
@@ -159,7 +168,6 @@ class SlavedEventStore(BaseSlavedStore):
     _get_rooms_for_user_where_membership_is_txn = (
         DataStore._get_rooms_for_user_where_membership_is_txn.__func__
     )
-    _get_members_rows_txn = DataStore._get_members_rows_txn.__func__
     _get_state_for_groups = DataStore._get_state_for_groups.__func__
     _get_all_state_from_cache = DataStore._get_all_state_from_cache.__func__
     _get_events_around_txn = DataStore._get_events_around_txn.__func__
@@ -194,84 +202,47 @@ class SlavedEventStore(BaseSlavedStore):
         result["backfill"] = -self._backfill_id_gen.get_current_token()
         return result
 
-    def process_replication(self, result):
-        stream = result.get("events")
-        if stream:
-            self._stream_id_gen.advance(int(stream["position"]))
-
-            if stream["rows"]:
-                logger.info("Got %d event rows", len(stream["rows"]))
-
-            for row in stream["rows"]:
-                self._process_replication_row(
-                    row, backfilled=False,
+    def process_replication_rows(self, stream_name, token, rows):
+        if stream_name == "events":
+            self._stream_id_gen.advance(token)
+            for row in rows:
+                self.invalidate_caches_for_event(
+                    token, row.event_id, row.room_id, row.type, row.state_key,
+                    row.redacts,
+                    backfilled=False,
                 )
-
-        stream = result.get("backfill")
-        if stream:
-            self._backfill_id_gen.advance(-int(stream["position"]))
-            for row in stream["rows"]:
-                self._process_replication_row(
-                    row, backfilled=True,
+        elif stream_name == "backfill":
+            self._backfill_id_gen.advance(-token)
+            for row in rows:
+                self.invalidate_caches_for_event(
+                    -token, row.event_id, row.room_id, row.type, row.state_key,
+                    row.redacts,
+                    backfilled=True,
                 )
-
-        stream = result.get("forward_ex_outliers")
-        if stream:
-            self._stream_id_gen.advance(int(stream["position"]))
-            for row in stream["rows"]:
-                event_id = row[1]
-                self._invalidate_get_event_cache(event_id)
-
-        stream = result.get("backward_ex_outliers")
-        if stream:
-            self._backfill_id_gen.advance(-int(stream["position"]))
-            for row in stream["rows"]:
-                event_id = row[1]
-                self._invalidate_get_event_cache(event_id)
-
-        return super(SlavedEventStore, self).process_replication(result)
-
-    def _process_replication_row(self, row, backfilled):
-        internal = json.loads(row[1])
-        event_json = json.loads(row[2])
-        event = FrozenEvent(event_json, internal_metadata_dict=internal)
-        self.invalidate_caches_for_event(
-            event, backfilled,
+        return super(SlavedEventStore, self).process_replication_rows(
+            stream_name, token, rows
         )
 
-    def invalidate_caches_for_event(self, event, backfilled):
-        self._invalidate_get_event_cache(event.event_id)
+    def invalidate_caches_for_event(self, stream_ordering, event_id, room_id,
+                                    etype, state_key, redacts, backfilled):
+        self._invalidate_get_event_cache(event_id)
 
-        self.get_latest_event_ids_in_room.invalidate((event.room_id,))
+        self.get_latest_event_ids_in_room.invalidate((room_id,))
 
         self.get_unread_event_push_actions_by_room_for_user.invalidate_many(
-            (event.room_id,)
+            (room_id,)
         )
 
         if not backfilled:
             self._events_stream_cache.entity_has_changed(
-                event.room_id, event.internal_metadata.stream_ordering
+                room_id, stream_ordering
             )
 
-        # self.get_unread_event_push_actions_by_room_for_user.invalidate_many(
-        #     (event.room_id,)
-        # )
+        if redacts:
+            self._invalidate_get_event_cache(redacts)
 
-        if event.type == EventTypes.Redaction:
-            self._invalidate_get_event_cache(event.redacts)
-
-        if event.type == EventTypes.Member:
+        if etype == EventTypes.Member:
             self._membership_stream_cache.entity_has_changed(
-                event.state_key, event.internal_metadata.stream_ordering
+                state_key, stream_ordering
             )
-            self.get_invited_rooms_for_user.invalidate((event.state_key,))
-
-        if not event.is_state():
-            return
-
-        if backfilled:
-            return
-
-        if (not event.internal_metadata.is_invite_from_remote()
-                and event.internal_metadata.is_outlier()):
-            return
+            self.get_invited_rooms_for_user.invalidate((state_key,))

synapse/replication/slave/storage/presence.py

@@ -18,6 +18,7 @@ from ._slaved_id_tracker import SlavedIdTracker
 
 from synapse.util.caches.stream_change_cache import StreamChangeCache
 from synapse.storage import DataStore
+from synapse.storage.presence import PresenceStore
 
 
 class SlavedPresenceStore(BaseSlavedStore):
@@ -35,7 +36,18 @@ class SlavedPresenceStore(BaseSlavedStore):
 
     _get_active_presence = DataStore._get_active_presence.__func__
     take_presence_startup_info = DataStore.take_presence_startup_info.__func__
-    get_presence_for_users = DataStore.get_presence_for_users.__func__
+    _get_presence_for_user = PresenceStore.__dict__["_get_presence_for_user"]
+    get_presence_for_users = PresenceStore.__dict__["get_presence_for_users"]
+
+    # XXX: This is a bit broken because we don't persist the accepted list in a
+    # way that can be replicated. This means that we don't have a way to
+    # invalidate the cache correctly.
+    get_presence_list_accepted = PresenceStore.__dict__[
+        "get_presence_list_accepted"
+    ]
+    get_presence_list_observers_accepted = PresenceStore.__dict__[
+        "get_presence_list_observers_accepted"
+    ]
 
     def get_current_presence_token(self):
         return self._presence_id_gen.get_current_token()
@@ -46,14 +58,14 @@ class SlavedPresenceStore(BaseSlavedStore):
         result["presence"] = position
         return result
 
-    def process_replication(self, result):
-        stream = result.get("presence")
-        if stream:
-            self._presence_id_gen.advance(int(stream["position"]))
-            for row in stream["rows"]:
-                position, user_id = row[:2]
+    def process_replication_rows(self, stream_name, token, rows):
+        if stream_name == "presence":
+            self._presence_id_gen.advance(token)
+            for row in rows:
                 self.presence_stream_cache.entity_has_changed(
-                    user_id, position
+                    row.user_id, token
                 )
-
-        return super(SlavedPresenceStore, self).process_replication(result)
+                self._get_presence_for_user.invalidate((row.user_id,))
+        return super(SlavedPresenceStore, self).process_replication_rows(
+            stream_name, token, rows
+        )

synapse/replication/slave/storage/push_rule.py

@@ -50,18 +50,15 @@ class SlavedPushRuleStore(SlavedEventStore):
         result["push_rules"] = self._push_rules_stream_id_gen.get_current_token()
         return result
 
-    def process_replication(self, result):
-        stream = result.get("push_rules")
-        if stream:
-            for row in stream["rows"]:
-                position = row[0]
-                user_id = row[2]
-                self.get_push_rules_for_user.invalidate((user_id,))
-                self.get_push_rules_enabled_for_user.invalidate((user_id,))
+    def process_replication_rows(self, stream_name, token, rows):
+        if stream_name == "push_rules":
+            self._push_rules_stream_id_gen.advance(token)
+            for row in rows:
+                self.get_push_rules_for_user.invalidate((row.user_id,))
+                self.get_push_rules_enabled_for_user.invalidate((row.user_id,))
                 self.push_rules_stream_cache.entity_has_changed(
-                    user_id, position
+                    row.user_id, token
                 )
-
-            self._push_rules_stream_id_gen.advance(int(stream["position"]))
-
-        return super(SlavedPushRuleStore, self).process_replication(result)
+        return super(SlavedPushRuleStore, self).process_replication_rows(
+            stream_name, token, rows
+        )

synapse/replication/slave/storage/pushers.py

@@ -40,13 +40,9 @@ class SlavedPusherStore(BaseSlavedStore):
         result["pushers"] = self._pushers_id_gen.get_current_token()
         return result
 
-    def process_replication(self, result):
-        stream = result.get("pushers")
-        if stream:
-            self._pushers_id_gen.advance(int(stream["position"]))
-
-        stream = result.get("deleted_pushers")
-        if stream:
-            self._pushers_id_gen.advance(int(stream["position"]))
-
-        return super(SlavedPusherStore, self).process_replication(result)
+    def process_replication_rows(self, stream_name, token, rows):
+        if stream_name == "pushers":
+            self._pushers_id_gen.advance(token)
+        return super(SlavedPusherStore, self).process_replication_rows(
+            stream_name, token, rows
+        )

synapse/replication/slave/storage/receipts.py

@@ -65,20 +65,22 @@ class SlavedReceiptsStore(BaseSlavedStore):
         result["receipts"] = self._receipts_id_gen.get_current_token()
         return result
 
-    def process_replication(self, result):
-        stream = result.get("receipts")
-        if stream:
-            self._receipts_id_gen.advance(int(stream["position"]))
-            for row in stream["rows"]:
-                position, room_id, receipt_type, user_id = row[:4]
-                self.invalidate_caches_for_receipt(room_id, receipt_type, user_id)
-                self._receipts_stream_cache.entity_has_changed(room_id, position)
-
-        return super(SlavedReceiptsStore, self).process_replication(result)
-
     def invalidate_caches_for_receipt(self, room_id, receipt_type, user_id):
         self.get_receipts_for_user.invalidate((user_id, receipt_type))
         self.get_linearized_receipts_for_room.invalidate_many((room_id,))
         self.get_last_receipt_event_id_for_user.invalidate(
             (user_id, room_id, receipt_type)
         )
+
+    def process_replication_rows(self, stream_name, token, rows):
+        if stream_name == "receipts":
+            self._receipts_id_gen.advance(token)
+            for row in rows:
+                self.invalidate_caches_for_receipt(
+                    row.room_id, row.receipt_type, row.user_id
+                )
+                self._receipts_stream_cache.entity_has_changed(row.room_id, token)
+
+        return super(SlavedReceiptsStore, self).process_replication_rows(
+            stream_name, token, rows
+        )

synapse/replication/slave/storage/room.py

@@ -46,9 +46,10 @@ class RoomStore(BaseSlavedStore):
         result["public_rooms"] = self._public_room_id_gen.get_current_token()
         return result
 
-    def process_replication(self, result):
-        stream = result.get("public_rooms")
-        if stream:
-            self._public_room_id_gen.advance(int(stream["position"]))
+    def process_replication_rows(self, stream_name, token, rows):
+        if stream_name == "public_rooms":
+            self._public_room_id_gen.advance(token)
 
-        return super(RoomStore, self).process_replication(result)
+        return super(RoomStore, self).process_replication_rows(
+            stream_name, token, rows
+        )

synapse/replication/tcp/__init__.py

@@ -0,0 +1,30 @@
+# -*- coding: utf-8 -*-
+# Copyright 2017 Vector Creations Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""This module implements the TCP replication protocol used by synapse to
+communicate between the master process and its workers (when they're enabled).
+
+Further details can be found in docs/tcp_replication.rst
+
+
+Structure of the module:
+ * client.py   - the client classes used for workers to connect to master
+ * command.py  - the definitions of all the valid commands
+ * protocol.py - contains bot the client and server protocol implementations,
+                 these should not be used directly
+ * resource.py - the server classes that accepts and handle client connections
+ * streams.py  - the definitons of all the valid streams
+
+"""

synapse/replication/tcp/client.py

@@ -0,0 +1,196 @@
+# -*- coding: utf-8 -*-
+# Copyright 2017 Vector Creations Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""A replication client for use by synapse workers.
+"""
+
+from twisted.internet import reactor, defer
+from twisted.internet.protocol import ReconnectingClientFactory
+
+from .commands import (
+    FederationAckCommand, UserSyncCommand, RemovePusherCommand, InvalidateCacheCommand,
+)
+from .protocol import ClientReplicationStreamProtocol
+
+import logging
+
+logger = logging.getLogger(__name__)
+
+
+class ReplicationClientFactory(ReconnectingClientFactory):
+    """Factory for building connections to the master. Will reconnect if the
+    connection is lost.
+
+    Accepts a handler that will be called when new data is available or data
+    is required.
+    """
+    maxDelay = 5  # Try at least once every N seconds
+
+    def __init__(self, hs, client_name, handler):
+        self.client_name = client_name
+        self.handler = handler
+        self.server_name = hs.config.server_name
+        self._clock = hs.get_clock()  # As self.clock is defined in super class
+
+        reactor.addSystemEventTrigger("before", "shutdown", self.stopTrying)
+
+    def startedConnecting(self, connector):
+        logger.info("Connecting to replication: %r", connector.getDestination())
+
+    def buildProtocol(self, addr):
+        logger.info("Connected to replication: %r", addr)
+        self.resetDelay()
+        return ClientReplicationStreamProtocol(
+            self.client_name, self.server_name, self._clock, self.handler
+        )
+
+    def clientConnectionLost(self, connector, reason):
+        logger.error("Lost replication conn: %r", reason)
+        ReconnectingClientFactory.clientConnectionLost(self, connector, reason)
+
+    def clientConnectionFailed(self, connector, reason):
+        logger.error("Failed to connect to replication: %r", reason)
+        ReconnectingClientFactory.clientConnectionFailed(
+            self, connector, reason
+        )
+
+
+class ReplicationClientHandler(object):
+    """A base handler that can be passed to the ReplicationClientFactory.
+
+    By default proxies incoming replication data to the SlaveStore.
+    """
+    def __init__(self, store):
+        self.store = store
+
+        # The current connection. None if we are currently (re)connecting
+        self.connection = None
+
+        # Any pending commands to be sent once a new connection has been
+        # established
+        self.pending_commands = []
+
+        # Map from string -> deferred, to wake up when receiveing a SYNC with
+        # the given string.
+        # Used for tests.
+        self.awaiting_syncs = {}
+
+    def start_replication(self, hs):
+        """Helper method to start a replication connection to the remote server
+        using TCP.
+        """
+        client_name = hs.config.worker_name
+        factory = ReplicationClientFactory(hs, client_name, self)
+        host = hs.config.worker_replication_host
+        port = hs.config.worker_replication_port
+        reactor.connectTCP(host, port, factory)
+
+    def on_rdata(self, stream_name, token, rows):
+        """Called when we get new replication data. By default this just pokes
+        the slave store.
+
+        Can be overriden in subclasses to handle more.
+        """
+        logger.info("Received rdata %s -> %s", stream_name, token)
+        self.store.process_replication_rows(stream_name, token, rows)
+
+    def on_position(self, stream_name, token):
+        """Called when we get new position data. By default this just pokes
+        the slave store.
+
+        Can be overriden in subclasses to handle more.
+        """
+        self.store.process_replication_rows(stream_name, token, [])
+
+    def on_sync(self, data):
+        """When we received a SYNC we wake up any deferreds that were waiting
+        for the sync with the given data.
+
+        Used by tests.
+        """
+        d = self.awaiting_syncs.pop(data, None)
+        if d:
+            d.callback(data)
+
+    def get_streams_to_replicate(self):
+        """Called when a new connection has been established and we need to
+        subscribe to streams.
+
+        Returns a dictionary of stream name to token.
+        """
+        args = self.store.stream_positions()
+        user_account_data = args.pop("user_account_data", None)
+        room_account_data = args.pop("room_account_data", None)
+        if user_account_data:
+            args["account_data"] = user_account_data
+        elif room_account_data:
+            args["account_data"] = room_account_data
+        return args
+
+    def get_currently_syncing_users(self):
+        """Get the list of currently syncing users (if any). This is called
+        when a connection has been established and we need to send the
+        currently syncing users. (Overriden by the synchrotron's only)
+        """
+        return []
+
+    def send_command(self, cmd):
+        """Send a command to master (when we get establish a connection if we
+        don't have one already.)
+        """
+        if self.connection:
+            self.connection.send_command(cmd)
+        else:
+            logger.warn("Queuing command as not connected: %r", cmd.NAME)
+            self.pending_commands.append(cmd)
+
+    def send_federation_ack(self, token):
+        """Ack data for the federation stream. This allows the master to drop
+        data stored purely in memory.
+        """
+        self.send_command(FederationAckCommand(token))
+
+    def send_user_sync(self, user_id, is_syncing, last_sync_ms):
+        """Poke the master that a user has started/stopped syncing.
+        """
+        self.send_command(UserSyncCommand(user_id, is_syncing, last_sync_ms))
+
+    def send_remove_pusher(self, app_id, push_key, user_id):
+        """Poke the master to remove a pusher for a user
+        """
+        cmd = RemovePusherCommand(app_id, push_key, user_id)
+        self.send_command(cmd)
+
+    def send_invalidate_cache(self, cache_func, keys):
+        """Poke the master to invalidate a cache.
+        """
+        cmd = InvalidateCacheCommand(cache_func.__name__, keys)
+        self.send_command(cmd)
+
+    def await_sync(self, data):
+        """Returns a deferred that is resolved when we receive a SYNC command
+        with given data.
+
+        Used by tests.
+        """
+        return self.awaiting_syncs.setdefault(data, defer.Deferred())
+
+    def update_connection(self, connection):
+        """Called when a connection has been established (or lost with None).
+        """
+        self.connection = connection
+        if connection:
+            for cmd in self.pending_commands:
+                connection.send_command(cmd)
+            self.pending_commands = []

synapse/replication/tcp/commands.py

@@ -0,0 +1,346 @@
+# -*- coding: utf-8 -*-
+# Copyright 2017 Vector Creations Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Defines the various valid commands
+
+The VALID_SERVER_COMMANDS and VALID_CLIENT_COMMANDS define which commands are
+allowed to be sent by which side.
+"""
+
+import logging
+import ujson as json
+
+
+logger = logging.getLogger(__name__)
+
+
+class Command(object):
+    """The base command class.
+
+    All subclasses must set the NAME variable which equates to the name of the
+    command on the wire.
+
+    A full command line on the wire is constructed from `NAME + " " + to_line()`
+
+    The default implementation creates a command of form `<NAME> <data>`
+    """
+    NAME = None
+
+    def __init__(self, data):
+        self.data = data
+
+    @classmethod
+    def from_line(cls, line):
+        """Deserialises a line from the wire into this command. `line` does not
+        include the command.
+        """
+        return cls(line)
+
+    def to_line(self):
+        """Serialises the comamnd for the wire. Does not include the command
+        prefix.
+        """
+        return self.data
+
+
+class ServerCommand(Command):
+    """Sent by the server on new connection and includes the server_name.
+
+    Format::
+
+        SERVER <server_name>
+    """
+    NAME = "SERVER"
+
+
+class RdataCommand(Command):
+    """Sent by server when a subscribed stream has an update.
+
+    Format::
+
+        RDATA <stream_name> <token> <row_json>
+
+    The `<token>` may either be a numeric stream id OR "batch". The latter case
+    is used to support sending multiple updates with the same stream ID. This
+    is done by sending an RDATA for each row, with all but the last RDATA having
+    a token of "batch" and the last having the final stream ID.
+
+    The client should batch all incoming RDATA with a token of "batch" (per
+    stream_name) until it sees an RDATA with a numeric stream ID.
+
+    `<token>` of "batch" maps to the instance variable `token` being None.
+
+    An example of a batched series of RDATA::
+
+        RDATA presence batch ["@foo:example.com", "online", ...]
+        RDATA presence batch ["@bar:example.com", "online", ...]
+        RDATA presence 59 ["@baz:example.com", "online", ...]
+    """
+    NAME = "RDATA"
+
+    def __init__(self, stream_name, token, row):
+        self.stream_name = stream_name
+        self.token = token
+        self.row = row
+
+    @classmethod
+    def from_line(cls, line):
+        stream_name, token, row_json = line.split(" ", 2)
+        return cls(
+            stream_name,
+            None if token == "batch" else int(token),
+            json.loads(row_json)
+        )
+
+    def to_line(self):
+        return " ".join((
+            self.stream_name,
+            str(self.token) if self.token is not None else "batch",
+            json.dumps(self.row),
+        ))
+
+
+class PositionCommand(Command):
+    """Sent by the client to tell the client the stream postition without
+    needing to send an RDATA.
+    """
+    NAME = "POSITION"
+
+    def __init__(self, stream_name, token):
+        self.stream_name = stream_name
+        self.token = token
+
+    @classmethod
+    def from_line(cls, line):
+        stream_name, token = line.split(" ", 1)
+        return cls(stream_name, int(token))
+
+    def to_line(self):
+        return " ".join((self.stream_name, str(self.token),))
+
+
+class ErrorCommand(Command):
+    """Sent by either side if there was an ERROR. The data is a string describing
+    the error.
+    """
+    NAME = "ERROR"
+
+
+class PingCommand(Command):
+    """Sent by either side as a keep alive. The data is arbitary (often timestamp)
+    """
+    NAME = "PING"
+
+
+class NameCommand(Command):
+    """Sent by client to inform the server of the client's identity. The data
+    is the name
+    """
+    NAME = "NAME"
+
+
+class ReplicateCommand(Command):
+    """Sent by the client to subscribe to the stream.
+
+    Format::
+
+        REPLICATE <stream_name> <token>
+
+    Where <token> may be either:
+        * a numeric stream_id to stream updates from
+        * "NOW" to stream all subsequent updates.
+
+    The <stream_name> can be "ALL" to subscribe to all known streams, in which
+    case the <token> must be set to "NOW", i.e.::
+
+        REPLICATE ALL NOW
+    """
+    NAME = "REPLICATE"
+
+    def __init__(self, stream_name, token):
+        self.stream_name = stream_name
+        self.token = token
+
+    @classmethod
+    def from_line(cls, line):
+        stream_name, token = line.split(" ", 1)
+        if token in ("NOW", "now"):
+            token = "NOW"
+        else:
+            token = int(token)
+        return cls(stream_name, token)
+
+    def to_line(self):
+        return " ".join((self.stream_name, str(self.token),))
+
+
+class UserSyncCommand(Command):
+    """Sent by the client to inform the server that a user has started or
+    stopped syncing. Used to calculate presence on the master.
+
+    Includes a timestamp of when the last user sync was.
+
+    Format::
+
+        USER_SYNC <user_id> <state> <last_sync_ms>
+
+    Where <state> is either "start" or "stop"
+    """
+    NAME = "USER_SYNC"
+
+    def __init__(self, user_id, is_syncing, last_sync_ms):
+        self.user_id = user_id
+        self.is_syncing = is_syncing
+        self.last_sync_ms = last_sync_ms
+
+    @classmethod
+    def from_line(cls, line):
+        user_id, state, last_sync_ms = line.split(" ", 2)
+
+        if state not in ("start", "end"):
+            raise Exception("Invalid USER_SYNC state %r" % (state,))
+
+        return cls(user_id, state == "start", int(last_sync_ms))
+
+    def to_line(self):
+        return " ".join((
+            self.user_id, "start" if self.is_syncing else "end", str(self.last_sync_ms),
+        ))
+
+
+class FederationAckCommand(Command):
+    """Sent by the client when it has processed up to a given point in the
+    federation stream. This allows the master to drop in-memory caches of the
+    federation stream.
+
+    This must only be sent from one worker (i.e. the one sending federation)
+
+    Format::
+
+        FEDERATION_ACK <token>
+    """
+    NAME = "FEDERATION_ACK"
+
+    def __init__(self, token):
+        self.token = token
+
+    @classmethod
+    def from_line(cls, line):
+        return cls(int(line))
+
+    def to_line(self):
+        return str(self.token)
+
+
+class SyncCommand(Command):
+    """Used for testing. The client protocol implementation allows waiting
+    on a SYNC command with a specified data.
+    """
+    NAME = "SYNC"
+
+
+class RemovePusherCommand(Command):
+    """Sent by the client to request the master remove the given pusher.
+
+    Format::
+
+        REMOVE_PUSHER <app_id> <push_key> <user_id>
+    """
+    NAME = "REMOVE_PUSHER"
+
+    def __init__(self, app_id, push_key, user_id):
+        self.user_id = user_id
+        self.app_id = app_id
+        self.push_key = push_key
+
+    @classmethod
+    def from_line(cls, line):
+        app_id, push_key, user_id = line.split(" ", 2)
+
+        return cls(app_id, push_key, user_id)
+
+    def to_line(self):
+        return " ".join((self.app_id, self.push_key, self.user_id))
+
+
+class InvalidateCacheCommand(Command):
+    """Sent by the client to invalidate an upstream cache.
+
+    THIS IS NOT RELIABLE, AND SHOULD *NOT* BE USED ACCEPT FOR THINGS THAT ARE
+    NOT DISASTROUS IF WE DROP ON THE FLOOR.
+
+    Mainly used to invalidate destination retry timing caches.
+
+    Format::
+
+        INVALIDATE_CACHE <cache_func> <keys_json>
+
+    Where <keys_json> is a json list.
+    """
+    NAME = "INVALIDATE_CACHE"
+
+    def __init__(self, cache_func, keys):
+        self.cache_func = cache_func
+        self.keys = keys
+
+    @classmethod
+    def from_line(cls, line):
+        cache_func, keys_json = line.split(" ", 1)
+
+        return cls(cache_func, json.loads(keys_json))
+
+    def to_line(self):
+        return " ".join((self.cache_func, json.dumps(self.keys)))
+
+
+# Map of command name to command type.
+COMMAND_MAP = {
+    cmd.NAME: cmd
+    for cmd in (
+        ServerCommand,
+        RdataCommand,
+        PositionCommand,
+        ErrorCommand,
+        PingCommand,
+        NameCommand,
+        ReplicateCommand,
+        UserSyncCommand,
+        FederationAckCommand,
+        SyncCommand,
+        RemovePusherCommand,
+        InvalidateCacheCommand,
+    )
+}
+
+# The commands the server is allowed to send
+VALID_SERVER_COMMANDS = (
+    ServerCommand.NAME,
+    RdataCommand.NAME,
+    PositionCommand.NAME,
+    ErrorCommand.NAME,
+    PingCommand.NAME,
+    SyncCommand.NAME,
+)
+
+# The commands the client is allowed to send
+VALID_CLIENT_COMMANDS = (
+    NameCommand.NAME,
+    ReplicateCommand.NAME,
+    PingCommand.NAME,
+    UserSyncCommand.NAME,
+    FederationAckCommand.NAME,
+    RemovePusherCommand.NAME,
+    InvalidateCacheCommand.NAME,
+    ErrorCommand.NAME,
+)

synapse/replication/tcp/protocol.py

@@ -0,0 +1,640 @@
+# -*- coding: utf-8 -*-
+# Copyright 2017 Vector Creations Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""This module contains the implementation of both the client and server
+protocols.
+
+The basic structure of the protocol is line based, where the initial word of
+each line specifies the command. The rest of the line is parsed based on the
+command. For example, the `RDATA` command is defined as::
+
+    RDATA <stream_name> <token> <row_json>
+
+(Note that `<row_json>` may contains spaces, but cannot contain newlines.)
+
+Blank lines are ignored.
+
+# Example
+
+An example iteraction is shown below. Each line is prefixed with '>' or '<' to
+indicate which side is sending, these are *not* included on the wire::
+
+    * connection established *
+    > SERVER localhost:8823
+    > PING 1490197665618
+    < NAME synapse.app.appservice
+    < PING 1490197665618
+    < REPLICATE events 1
+    < REPLICATE backfill 1
+    < REPLICATE caches 1
+    > POSITION events 1
+    > POSITION backfill 1
+    > POSITION caches 1
+    > RDATA caches 2 ["get_user_by_id",["@01register-user:localhost:8823"],1490197670513]
+    > RDATA events 14 ["$149019767112vOHxz:localhost:8823",
+        "!AFDCvgApUmpdfVjIXm:localhost:8823","m.room.guest_access","",null]
+    < PING 1490197675618
+    > ERROR server stopping
+    * connection closed by server *
+"""
+
+from twisted.internet import defer
+from twisted.protocols.basic import LineOnlyReceiver
+from twisted.python.failure import Failure
+
+from commands import (
+    COMMAND_MAP, VALID_CLIENT_COMMANDS, VALID_SERVER_COMMANDS,
+    ErrorCommand, ServerCommand, RdataCommand, PositionCommand, PingCommand,
+    NameCommand, ReplicateCommand, UserSyncCommand, SyncCommand,
+)
+from streams import STREAMS_MAP
+
+from synapse.util.stringutils import random_string
+from synapse.metrics.metric import CounterMetric
+
+import logging
+import synapse.metrics
+import struct
+import fcntl
+
+
+metrics = synapse.metrics.get_metrics_for(__name__)
+
+connection_close_counter = metrics.register_counter(
+    "close_reason", labels=["reason_type"],
+)
+
+
+# A list of all connected protocols. This allows us to send metrics about the
+# connections.
+connected_connections = []
+
+
+logger = logging.getLogger(__name__)
+
+
+PING_TIME = 5000
+PING_TIMEOUT_MULTIPLIER = 5
+PING_TIMEOUT_MS = PING_TIME * PING_TIMEOUT_MULTIPLIER
+
+
+class ConnectionStates(object):
+    CONNECTING = "connecting"
+    ESTABLISHED = "established"
+    PAUSED = "paused"
+    CLOSED = "closed"
+
+
+class BaseReplicationStreamProtocol(LineOnlyReceiver):
+    """Base replication protocol shared between client and server.
+
+    Reads lines (ignoring blank ones) and parses them into command classes,
+    asserting that they are valid for the given direction, i.e. server commands
+    are only sent by the server.
+
+    On receiving a new command it calls `on_<COMMAND_NAME>` with the parsed
+    command.
+
+    It also sends `PING` periodically, and correctly times out remote connections
+    (if they send a `PING` command)
+    """
+    delimiter = b'\n'
+
+    VALID_INBOUND_COMMANDS = []  # Valid commands we expect to receive
+    VALID_OUTBOUND_COMMANDS = []  # Valid commans we can send
+
+    max_line_buffer = 10000
+
+    def __init__(self, clock):
+        self.clock = clock
+
+        self.last_received_command = self.clock.time_msec()
+        self.last_sent_command = 0
+        self.time_we_closed = None  # When we requested the connection be closed
+
+        self.received_ping = False  # Have we reecived a ping from the other side
+
+        self.state = ConnectionStates.CONNECTING
+
+        self.name = "anon"  # The name sent by a client.
+        self.conn_id = random_string(5)  # To dedupe in case of name clashes.
+
+        # List of pending commands to send once we've established the connection
+        self.pending_commands = []
+
+        # The LoopingCall for sending pings.
+        self._send_ping_loop = None
+
+        self.inbound_commands_counter = CounterMetric(
+            "inbound_commands", labels=["command"],
+        )
+        self.outbound_commands_counter = CounterMetric(
+            "outbound_commands", labels=["command"],
+        )
+
+    def connectionMade(self):
+        logger.info("[%s] Connection established", self.id())
+
+        self.state = ConnectionStates.ESTABLISHED
+
+        connected_connections.append(self)  # Register connection for metrics
+
+        self.transport.registerProducer(self, True)  # For the *Producing callbacks
+
+        self._send_pending_commands()
+
+        # Starts sending pings
+        self._send_ping_loop = self.clock.looping_call(self.send_ping, 5000)
+
+        # Always send the initial PING so that the other side knows that they
+        # can time us out.
+        self.send_command(PingCommand(self.clock.time_msec()))
+
+    def send_ping(self):
+        """Periodically sends a ping and checks if we should close the connection
+        due to the other side timing out.
+        """
+        now = self.clock.time_msec()
+
+        if self.time_we_closed:
+            if now - self.time_we_closed > PING_TIMEOUT_MS:
+                logger.info(
+                    "[%s] Failed to close connection gracefully, aborting", self.id()
+                )
+                self.transport.abortConnection()
+        else:
+            if now - self.last_sent_command >= PING_TIME:
+                self.send_command(PingCommand(now))
+
+            if self.received_ping and now - self.last_received_command > PING_TIMEOUT_MS:
+                logger.info(
+                    "[%s] Connection hasn't received command in %r ms. Closing.",
+                    self.id(), now - self.last_received_command
+                )
+                self.send_error("ping timeout")
+
+    def lineReceived(self, line):
+        """Called when we've received a line
+        """
+        if line.strip() == "":
+            # Ignore blank lines
+            return
+
+        line = line.decode("utf-8")
+        cmd_name, rest_of_line = line.split(" ", 1)
+
+        if cmd_name not in self.VALID_INBOUND_COMMANDS:
+            logger.error("[%s] invalid command %s", self.id(), cmd_name)
+            self.send_error("invalid command: %s", cmd_name)
+            return
+
+        self.last_received_command = self.clock.time_msec()
+
+        self.inbound_commands_counter.inc(cmd_name)
+
+        cmd_cls = COMMAND_MAP[cmd_name]
+        try:
+            cmd = cmd_cls.from_line(rest_of_line)
+        except Exception as e:
+            logger.exception(
+                "[%s] failed to parse line %r: %r", self.id(), cmd_name, rest_of_line
+            )
+            self.send_error(
+                "failed to parse line for  %r: %r (%r):" % (cmd_name, e, rest_of_line)
+            )
+            return
+
+        # Now lets try and call on_<CMD_NAME> function
+        try:
+            getattr(self, "on_%s" % (cmd_name,))(cmd)
+        except Exception:
+            logger.exception("[%s] Failed to handle line: %r", self.id(), line)
+
+    def close(self):
+        logger.warn("[%s] Closing connection", self.id())
+        self.time_we_closed = self.clock.time_msec()
+        self.transport.loseConnection()
+        self.on_connection_closed()
+
+    def send_error(self, error_string, *args):
+        """Send an error to remote and close the connection.
+        """
+        self.send_command(ErrorCommand(error_string % args))
+        self.close()
+
+    def send_command(self, cmd, do_buffer=True):
+        """Send a command if connection has been established.
+
+        Args:
+            cmd (Command)
+            do_buffer (bool): Whether to buffer the message or always attempt
+                to send the command. This is mostly used to send an error
+                message if we're about to close the connection due our buffers
+                becoming full.
+        """
+        if self.state == ConnectionStates.CLOSED:
+            logger.info("[%s] Not sending, connection closed", self.id())
+            return
+
+        if do_buffer and self.state != ConnectionStates.ESTABLISHED:
+            self._queue_command(cmd)
+            return
+
+        self.outbound_commands_counter.inc(cmd.NAME)
+
+        string = "%s %s" % (cmd.NAME, cmd.to_line(),)
+        if "\n" in string:
+            raise Exception("Unexpected newline in command: %r", string)
+
+        self.sendLine(string.encode("utf-8"))
+
+        self.last_sent_command = self.clock.time_msec()
+
+    def _queue_command(self, cmd):
+        """Queue the command until the connection is ready to write to again.
+        """
+        logger.info("[%s] Queing as conn %r, cmd: %r", self.id(), self.state, cmd)
+        self.pending_commands.append(cmd)
+
+        if len(self.pending_commands) > self.max_line_buffer:
+            # The other side is failing to keep up and out buffers are becoming
+            # full, so lets close the connection.
+            # XXX: should we squawk more loudly?
+            logger.error("[%s] Remote failed to keep up", self.id())
+            self.send_command(ErrorCommand("Failed to keep up"), do_buffer=False)
+            self.close()
+
+    def _send_pending_commands(self):
+        """Send any queued commandes
+        """
+        pending = self.pending_commands
+        self.pending_commands = []
+        for cmd in pending:
+            self.send_command(cmd)
+
+    def on_PING(self, line):
+        self.received_ping = True
+
+    def on_ERROR(self, cmd):
+        logger.error("[%s] Remote reported error: %r", self.id(), cmd.data)
+
+    def pauseProducing(self):
+        """This is called when both the kernel send buffer and the twisted
+        tcp connection send buffers have become full.
+
+        We don't actually have any control over those sizes, so we buffer some
+        commands ourselves before knifing the connection due to the remote
+        failing to keep up.
+        """
+        logger.info("[%s] Pause producing", self.id())
+        self.state = ConnectionStates.PAUSED
+
+    def resumeProducing(self):
+        """The remote has caught up after we started buffering!
+        """
+        logger.info("[%s] Resume producing", self.id())
+        self.state = ConnectionStates.ESTABLISHED
+        self._send_pending_commands()
+
+    def stopProducing(self):
+        """We're never going to send any more data (normally because either
+        we or the remote has closed the connection)
+        """
+        logger.info("[%s] Stop producing", self.id())
+        self.on_connection_closed()
+
+    def connectionLost(self, reason):
+        logger.info("[%s] Replication connection closed: %r", self.id(), reason)
+        if isinstance(reason, Failure):
+            connection_close_counter.inc(reason.type.__name__)
+        else:
+            connection_close_counter.inc(reason.__class__.__name__)
+
+        try:
+            # Remove us from list of connections to be monitored
+            connected_connections.remove(self)
+        except ValueError:
+            pass
+
+        # Stop the looping call sending pings.
+        if self._send_ping_loop and self._send_ping_loop.running:
+            self._send_ping_loop.stop()
+
+        self.on_connection_closed()
+
+    def on_connection_closed(self):
+        logger.info("[%s] Connection was closed", self.id())
+
+        self.state = ConnectionStates.CLOSED
+        self.pending_commands = []
+
+        if self.transport:
+            self.transport.unregisterProducer()
+
+    def __str__(self):
+        return "ReplicationConnection<name=%s,conn_id=%s,addr=%s>" % (
+            self.name, self.conn_id, self.addr,
+        )
+
+    def id(self):
+        return "%s-%s" % (self.name, self.conn_id)
+
+
+class ServerReplicationStreamProtocol(BaseReplicationStreamProtocol):
+    VALID_INBOUND_COMMANDS = VALID_CLIENT_COMMANDS
+    VALID_OUTBOUND_COMMANDS = VALID_SERVER_COMMANDS
+
+    def __init__(self, server_name, clock, streamer, addr):
+        BaseReplicationStreamProtocol.__init__(self, clock)  # Old style class
+
+        self.server_name = server_name
+        self.streamer = streamer
+        self.addr = addr
+
+        # The streams the client has subscribed to and is up to date with
+        self.replication_streams = set()
+
+        # The streams the client is currently subscribing to.
+        self.connecting_streams = set()
+
+        # Map from stream name to list of updates to send once we've finished
+        # subscribing the client to the stream.
+        self.pending_rdata = {}
+
+    def connectionMade(self):
+        self.send_command(ServerCommand(self.server_name))
+        BaseReplicationStreamProtocol.connectionMade(self)
+        self.streamer.new_connection(self)
+
+    def on_NAME(self, cmd):
+        logger.info("[%s] Renamed to %r", self.id(), cmd.data)
+        self.name = cmd.data
+
+    def on_USER_SYNC(self, cmd):
+        self.streamer.on_user_sync(
+            self.conn_id, cmd.user_id, cmd.is_syncing, cmd.last_sync_ms,
+        )
+
+    def on_REPLICATE(self, cmd):
+        stream_name = cmd.stream_name
+        token = cmd.token
+
+        if stream_name == "ALL":
+            # Subscribe to all streams we're publishing to.
+            for stream in self.streamer.streams_by_name.iterkeys():
+                self.subscribe_to_stream(stream, token)
+        else:
+            self.subscribe_to_stream(stream_name, token)
+
+    def on_FEDERATION_ACK(self, cmd):
+        self.streamer.federation_ack(cmd.token)
+
+    def on_REMOVE_PUSHER(self, cmd):
+        self.streamer.on_remove_pusher(cmd.app_id, cmd.push_key, cmd.user_id)
+
+    def on_INVALIDATE_CACHE(self, cmd):
+        self.streamer.on_invalidate_cache(cmd.cache_func, cmd.keys)
+
+    @defer.inlineCallbacks
+    def subscribe_to_stream(self, stream_name, token):
+        """Subscribe the remote to a streams.
+
+        This invloves checking if they've missed anything and sending those
+        updates down if they have. During that time new updates for the stream
+        are queued and sent once we've sent down any missed updates.
+        """
+        self.replication_streams.discard(stream_name)
+        self.connecting_streams.add(stream_name)
+
+        try:
+            # Get missing updates
+            updates, current_token = yield self.streamer.get_stream_updates(
+                stream_name, token,
+            )
+
+            # Send all the missing updates
+            for update in updates:
+                token, row = update[0], update[1]
+                self.send_command(RdataCommand(stream_name, token, row))
+
+            # We send a POSITION command to ensure that they have an up to
+            # date token (especially useful if we didn't send any updates
+            # above)
+            self.send_command(PositionCommand(stream_name, current_token))
+
+            # Now we can send any updates that came in while we were subscribing
+            pending_rdata = self.pending_rdata.pop(stream_name, [])
+            for token, update in pending_rdata:
+                # Only send updates newer than the current token
+                if token > current_token:
+                    self.send_command(RdataCommand(stream_name, token, update))
+
+            # They're now fully subscribed
+            self.replication_streams.add(stream_name)
+        except Exception as e:
+            logger.exception("[%s] Failed to handle REPLICATE command", self.id())
+            self.send_error("failed to handle replicate: %r", e)
+        finally:
+            self.connecting_streams.discard(stream_name)
+
+    def stream_update(self, stream_name, token, data):
+        """Called when a new update is available to stream to clients.
+
+        We need to check if the client is interested in the stream or not
+        """
+        if stream_name in self.replication_streams:
+            # The client is subscribed to the stream
+            self.send_command(RdataCommand(stream_name, token, data))
+        elif stream_name in self.connecting_streams:
+            # The client is being subscribed to the stream
+            logger.debug("[%s] Queuing RDATA %r %r", self.id(), stream_name, token)
+            self.pending_rdata.setdefault(stream_name, []).append((token, data))
+        else:
+            # The client isn't subscribed
+            logger.debug("[%s] Dropping RDATA %r %r", self.id(), stream_name, token)
+
+    def send_sync(self, data):
+        self.send_command(SyncCommand(data))
+
+    def on_connection_closed(self):
+        BaseReplicationStreamProtocol.on_connection_closed(self)
+        self.streamer.lost_connection(self)
+
+
+class ClientReplicationStreamProtocol(BaseReplicationStreamProtocol):
+    VALID_INBOUND_COMMANDS = VALID_SERVER_COMMANDS
+    VALID_OUTBOUND_COMMANDS = VALID_CLIENT_COMMANDS
+
+    def __init__(self, client_name, server_name, clock, handler):
+        BaseReplicationStreamProtocol.__init__(self, clock)
+
+        self.client_name = client_name
+        self.server_name = server_name
+        self.handler = handler
+
+        # Map of stream to batched updates. See RdataCommand for info on how
+        # batching works.
+        self.pending_batches = {}
+
+    def connectionMade(self):
+        self.send_command(NameCommand(self.client_name))
+        BaseReplicationStreamProtocol.connectionMade(self)
+
+        # Once we've connected subscribe to the necessary streams
+        for stream_name, token in self.handler.get_streams_to_replicate().iteritems():
+            self.replicate(stream_name, token)
+
+        # Tell the server if we have any users currently syncing (should only
+        # happen on synchrotrons)
+        currently_syncing = self.handler.get_currently_syncing_users()
+        now = self.clock.time_msec()
+        for user_id in currently_syncing:
+            self.send_command(UserSyncCommand(user_id, True, now))
+
+        # We've now finished connecting to so inform the client handler
+        self.handler.update_connection(self)
+
+    def on_SERVER(self, cmd):
+        if cmd.data != self.server_name:
+            logger.error("[%s] Connected to wrong remote: %r", self.id(), cmd.data)
+            self.send_error("Wrong remote")
+
+    def on_RDATA(self, cmd):
+        try:
+            row = STREAMS_MAP[cmd.stream_name].ROW_TYPE(*cmd.row)
+        except Exception:
+            logger.exception(
+                "[%s] Failed to parse RDATA: %r %r",
+                self.id(), cmd.stream_name, cmd.row
+            )
+            raise
+
+        if cmd.token is None:
+            # I.e. this is part of a batch of updates for this stream. Batch
+            # until we get an update for the stream with a non None token
+            self.pending_batches.setdefault(cmd.stream_name, []).append(row)
+        else:
+            # Check if this is the last of a batch of updates
+            rows = self.pending_batches.pop(cmd.stream_name, [])
+            rows.append(row)
+
+            self.handler.on_rdata(cmd.stream_name, cmd.token, rows)
+
+    def on_POSITION(self, cmd):
+        self.handler.on_position(cmd.stream_name, cmd.token)
+
+    def on_SYNC(self, cmd):
+        self.handler.on_sync(cmd.data)
+
+    def replicate(self, stream_name, token):
+        """Send the subscription request to the server
+        """
+        if stream_name not in STREAMS_MAP:
+            raise Exception("Invalid stream name %r" % (stream_name,))
+
+        logger.info(
+            "[%s] Subscribing to replication stream: %r from %r",
+            self.id(), stream_name, token
+        )
+
+        self.send_command(ReplicateCommand(stream_name, token))
+
+    def on_connection_closed(self):
+        BaseReplicationStreamProtocol.on_connection_closed(self)
+        self.handler.update_connection(None)
+
+
+# The following simply registers metrics for the replication connections
+
+metrics.register_callback(
+    "pending_commands",
+    lambda: {
+        (p.name, p.conn_id): len(p.pending_commands)
+        for p in connected_connections
+    },
+    labels=["name", "conn_id"],
+)
+
+
+def transport_buffer_size(protocol):
+    if protocol.transport:
+        size = len(protocol.transport.dataBuffer) + protocol.transport._tempDataLen
+        return size
+    return 0
+
+
+metrics.register_callback(
+    "transport_send_buffer",
+    lambda: {
+        (p.name, p.conn_id): transport_buffer_size(p)
+        for p in connected_connections
+    },
+    labels=["name", "conn_id"],
+)
+
+
+def transport_kernel_read_buffer_size(protocol, read=True):
+    SIOCINQ = 0x541B
+    SIOCOUTQ = 0x5411
+
+    if protocol.transport:
+        fileno = protocol.transport.getHandle().fileno()
+        if read:
+            op = SIOCINQ
+        else:
+            op = SIOCOUTQ
+        size = struct.unpack("I", fcntl.ioctl(fileno, op, '\0\0\0\0'))[0]
+        return size
+    return 0
+
+
+metrics.register_callback(
+    "transport_kernel_send_buffer",
+    lambda: {
+        (p.name, p.conn_id): transport_kernel_read_buffer_size(p, False)
+        for p in connected_connections
+    },
+    labels=["name", "conn_id"],
+)
+
+
+metrics.register_callback(
+    "transport_kernel_read_buffer",
+    lambda: {
+        (p.name, p.conn_id): transport_kernel_read_buffer_size(p, True)
+        for p in connected_connections
+    },
+    labels=["name", "conn_id"],
+)
+
+
+metrics.register_callback(
+    "inbound_commands",
+    lambda: {
+        (k[0], p.name, p.conn_id): count
+        for p in connected_connections
+        for k, count in p.inbound_commands_counter.counts.iteritems()
+    },
+    labels=["command", "name", "conn_id"],
+)
+
+metrics.register_callback(
+    "outbound_commands",
+    lambda: {
+        (k[0], p.name, p.conn_id): count
+        for p in connected_connections
+        for k, count in p.outbound_commands_counter.counts.iteritems()
+    },
+    labels=["command", "name", "conn_id"],
+)