Merge branch 'erikj/repl_tcp_server' of github.com:matrix-org/synapse into matrix-org-hotfixes-tcp-repl

Speed up cached function access

CHANGES.rst

@@ -1,3 +1,50 @@
+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>`_.
@@ -146,6 +146,7 @@ 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 +229,7 @@ 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
@@ -321,7 +323,7 @@ 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
 ------
@@ -808,7 +810,7 @@ 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

contrib/example_log_config.yaml

@@ -39,9 +39,11 @@ loggers:
     synapse:
         level: INFO
 
-    synapse.storage:
+    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:

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/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/tcp_replication.rst

@@ -0,0 +1,177 @@
+TCP Replication
+===============
+
+This describes the TCP replication protocol that replaces the HTTP protocol.
+
+Motivation
+----------
+
+The HTTP API used long poll from the workers to the master, this has the problem
+of causing a lot of duplicate work on the server. This TCP protocol aims to
+solve.
+
+Overview
+--------
+
+The protocol is based on fire and forget, line based commands. An example flow
+would be (where '>' indicates master->worker and '<' worker->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 consisdered 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``.

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/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"],
 }
 
 

synapse/__init__.py

@@ -16,4 +16,4 @@
 """ This is a reference implementation of a Matrix home server.
 """
 
-__version__ = "0.19.2"
+__version__ = "0.19.3"

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.
@@ -44,6 +45,7 @@ class JoinRules(object):
 class LoginType(object):
     PASSWORD = u"m.login.password"
     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 +51,35 @@ 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 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 +90,39 @@ 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 +148,11 @@ 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 +213,6 @@ 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 +282,19 @@ 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

@@ -29,7 +29,7 @@ from synapse.replication.slave.storage.registration import SlavedRegistrationSto
 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
@@ -157,7 +157,7 @@ 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 +187,11 @@ 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:

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.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
@@ -63,6 +64,7 @@ class ClientReaderSlavedStore(
     DirectoryStore,
     SlavedApplicationServiceStore,
     SlavedRegistrationStore,
+    TransactionStore,
     BaseSlavedStore,
     ClientIpStore,  # After BaseSlavedStore because the constructor is different
 ):
@@ -171,7 +173,7 @@ 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 +195,11 @@ 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:

synapse/app/federation_reader.py

@@ -31,7 +31,7 @@ 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
@@ -162,7 +162,7 @@ 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 +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:

synapse/app/federation_sender.py

@@ -35,7 +35,7 @@ from synapse.storage.engines import create_engine
 from synapse.storage.presence import UserPresenceState
 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
@@ -160,7 +160,7 @@ 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 +193,11 @@ 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:

synapse/app/homeserver.py

@@ -20,6 +20,8 @@ import gc
 import logging
 import os
 import sys
+
+import synapse.config.logger
 from synapse.config._base import ConfigError
 
 from synapse.python_dependencies import (
@@ -50,10 +52,11 @@ 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
@@ -220,6 +223,16 @@ 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 +299,7 @@ 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 +467,12 @@ 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

@@ -24,6 +24,7 @@ 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.rest.media.v0.content_repository import ContentRepoResource
 from synapse.rest.media.v1.media_repository import MediaRepositoryResource
 from synapse.server import HomeServer
@@ -32,7 +33,7 @@ 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
@@ -59,6 +60,7 @@ logger = logging.getLogger("synapse.app.media_repository")
 class MediaRepositorySlavedStore(
     SlavedApplicationServiceStore,
     SlavedRegistrationStore,
+    TransactionStore,
     BaseSlavedStore,
     MediaRepositoryStore,
     ClientIpStore,
@@ -168,7 +170,7 @@ 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 +192,11 @@ 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:

synapse/app/pusher.py

@@ -31,7 +31,8 @@ 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
@@ -245,7 +246,7 @@ 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 +276,11 @@ 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:

synapse/app/synchrotron.py

@@ -20,7 +20,6 @@ from synapse.api.constants import EventTypes, PresenceState
 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.http.site import SynapseSite
 from synapse.http.server import JsonResource
@@ -48,7 +47,8 @@ from synapse.storage.presence import PresenceStore, 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, preserve_fn, \
+    PreserveLoggingContext
 from synapse.util.manhole import manhole
 from synapse.util.rlimit import change_resource_limit
 from synapse.util.stringutils import random_string
@@ -399,8 +399,7 @@ class SynchrotronServer(HomeServer):
                 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]
+                room_ids = yield store.get_rooms_for_user(user_id)
 
                 notifier.on_new_event(
                     "device_list_key", position, rooms=room_ids,
@@ -411,11 +410,16 @@ class SynchrotronServer(HomeServer):
             stream = result.get("events")
             if stream:
                 max_position = stream["position"]
+
+                event_map = yield store.get_events([row[1] for row in stream["rows"]])
+
                 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)
+                    event_id = row[1]
+                    event = event_map.get(event_id, None)
+                    if not event:
+                        continue
+
                     extra_users = ()
                     if event.type == EventTypes.Member:
                         extra_users = (event.state_key,)
@@ -478,7 +482,7 @@ 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
 
@@ -497,7 +501,11 @@ 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:

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", [
@@ -190,7 +217,19 @@ 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:

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/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/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",

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
 

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
@@ -165,7 +164,7 @@ 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)
@@ -497,27 +496,16 @@ 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 +517,7 @@ 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

@@ -54,6 +54,7 @@ class FederationRemoteSendQueue(object):
     def __init__(self, hs):
         self.server_name = hs.hostname
         self.clock = hs.get_clock()
+        self.notifier = hs.get_notifier()
 
         self.presence_map = {}
         self.presence_changed = sorteddict()
@@ -186,6 +187,8 @@ class FederationRemoteSendQueue(object):
         else:
             self.edus[pos] = edu
 
+        self.notifier.on_new_replication_data()
+
     def send_presence(self, destination, states):
         """As per TransactionQueue"""
         pos = self._next_pos()
@@ -199,24 +202,33 @@ class FederationRemoteSendQueue(object):
             (destination, state.user_id) for state in 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
@@ -225,8 +237,8 @@ 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
 
         rows = []
 
@@ -237,10 +249,11 @@ class FederationRemoteSendQueue(object):
 
         # Fetch changed presence
         keys = self.presence_changed.keys()
-        i = keys.bisect_right(token)
+        i = keys.bisect_right(from_token)
+        j = keys.bisect_right(to_token) + 1
         dest_user_ids = set(
             (pos, dest_user_id)
-            for pos in keys[i:]
+            for pos in keys[i:j]
             for dest_user_id in self.presence_changed[pos]
         )
 
@@ -252,8 +265,9 @@ class FederationRemoteSendQueue(object):
 
         # 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
+        keyed_edus = set((k, self.keyed_edu_changed[k]) for k in keys[i:j])
 
         for (pos, (destination, edu_key)) in keyed_edus:
             rows.append(
@@ -265,16 +279,18 @@ class FederationRemoteSendQueue(object):
 
         # 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 = set((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())))
 
         # 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 = set((k, self.failures[k]) for k in keys[i:j])
 
         for (pos, (destination, failure)) in failures:
             rows.append((pos, FAILURE_TYPE, ujson.dumps({
@@ -284,8 +300,9 @@ class FederationRemoteSendQueue(object):
 
         # 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 = set((k, self.device_messages[k]) for k in keys[i:j])
 
         for (pos, destination) in device_messages:
             rows.append((pos, DEVICE_MESSAGE_TYPE, ujson.dumps({

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
 
@@ -22,9 +22,7 @@ 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.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
@@ -99,7 +97,12 @@ 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
@@ -300,20 +303,20 @@ 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:
-                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)
                 )
@@ -369,7 +372,6 @@ class TransactionQueue(object):
 
                 success = yield self._send_new_transaction(
                     destination, pending_pdus, pending_edus, pending_failures,
-                    limiter=limiter,
                 )
                 if success:
                     # Remove the acknowledged device messages from the database
@@ -387,12 +389,24 @@ 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)
@@ -432,7 +446,7 @@ 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])
@@ -442,132 +456,104 @@ 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)
@@ -242,6 +245,7 @@ class TransportLayerClient(object):
             destination=destination,
             path=path,
             data=content,
+            ignore_backoff=True,
         )
 
         defer.returnValue(response)
@@ -269,6 +273,7 @@ class TransportLayerClient(object):
             destination=remote_server,
             path=path,
             args=args,
+            ignore_backoff=True,
         )
 
         defer.returnValue(response)

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

@@ -16,6 +16,7 @@ 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.metrics import measure_func
 from synapse.types import get_domain_from_id, RoomStreamToken
 from twisted.internet import defer
@@ -34,10 +35,11 @@ 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,
@@ -167,6 +169,40 @@ 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
@@ -212,8 +248,7 @@ 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,
@@ -234,8 +269,7 @@ 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(
@@ -260,7 +294,7 @@ class DeviceHandler(BaseHandler):
                 # 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
@@ -299,58 +333,6 @@ 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 extremity and prev_ids[0] and int(extremity) >= int(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)
@@ -363,8 +345,8 @@ 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)
@@ -376,3 +358,129 @@ 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
+
+            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)
+                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.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

@@ -22,7 +22,7 @@ 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.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:
@@ -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
@@ -316,7 +308,7 @@ 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)
 

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
@@ -31,7 +32,7 @@ from synapse.util.logcontext import (
 )
 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 +80,216 @@ 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
+                        )
+
+            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.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 (str[]): List of event ids which we are missing
+            min_depth (int): Minimum depth of events to return.
+
+        Returns:
+            Deferred<dict(str, str?)>: updated have_seen dictionary
+        """
+        # 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:
+            # nothing left to do
+            defer.returnValue(have_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.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,
+        )
+
+        # 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.on_receive_pdu(
+                origin,
+                e,
+                get_missing=False
+            )
+
+        have_seen = yield self.store.have_events(
+            [ev for ev, _ in pdu.prev_events]
+        )
+        defer.returnValue(have_seen)
+
+    @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
@@ -670,8 +858,6 @@ 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 +866,15 @@ 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:
@@ -733,17 +927,36 @@ class FederationHandler(BaseHandler):
             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):
@@ -1096,7 +1309,7 @@ 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 +1538,17 @@ 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.
@@ -150,7 +151,7 @@ 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"
@@ -161,3 +162,37 @@ class IdentityHandler(BaseHandler):
         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 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/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
@@ -510,6 +513,71 @@ 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):
+        """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
+        """
+        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()
+            )
+            time_now_ms = self.clock.time_msec()
+
+            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=time_now_ms,
+                        last_user_sync_ts=time_now_ms,
+                    ))
+                else:
+                    updates.append(prev_state.copy_and_replace(
+                        last_user_sync_ts=time_now_ms,
+                    ))
+                process_presence.add(user_id)
+            elif user_id in process_presence:
+                updates.append(prev_state.copy_and_replace(
+                    last_user_sync_ts=time_now_ms,
+                ))
+                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.
@@ -559,9 +627,9 @@ class PresenceHandler(object):
         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)
+            room_ids = yield self.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 self.store.get_presence_list_observers_accepted(state.user_id)
             for u in plist:
@@ -577,8 +645,7 @@ class PresenceHandler(object):
                 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)
+                hosts = yield self.store.get_hosts_in_room(room_id)
 
                 for host in hosts:
                     hosts_to_states.setdefault(host, []).extend(local_states)
@@ -722,9 +789,7 @@ 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):
@@ -798,6 +863,9 @@ 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
         }
@@ -850,6 +918,7 @@ 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,
@@ -913,11 +982,12 @@ 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)
@@ -982,14 +1052,18 @@ 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:
@@ -1028,7 +1102,6 @@ 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)
 
@@ -1037,18 +1110,7 @@ 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
@@ -1076,16 +1138,13 @@ 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()
@@ -1093,6 +1152,31 @@ 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
@@ -1160,7 +1244,10 @@ 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,

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:
@@ -156,11 +158,11 @@ class ProfileHandler(BaseHandler):
 
         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/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/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/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

@@ -293,6 +293,9 @@ class TypingHandler(object):
         rows.sort()
         return rows
 
+    def get_current_token(self):
+        return self._latest_room_serial
+
 
 class TypingNotificationEventSource(object):
     def __init__(self, hs):

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 +103,152 @@ 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.
         """
-        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 +277,9 @@ 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 +294,19 @@ 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.
+
+            Fails with ``NotRetryingDestination`` if we are not yet ready
+            to retry this server.
         """
 
         if not json_data_callback:
@@ -282,26 +321,29 @@ 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 +356,15 @@ 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.
+
+            Fails with ``NotRetryingDestination`` if we are not yet ready
+            to retry this server.
         """
 
         def body_callback(method, url_bytes, headers_dict):
@@ -327,27 +373,29 @@ 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 +407,16 @@ 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.
 
             The result of the deferred is a tuple of `(code, response)`,
             where `response` is a dict representing the decoded JSON body.
+
+            Fails with ``NotRetryingDestination`` if we are not yet ready
+            to retry this server.
         """
         logger.debug("get_json args: %s", args)
 
@@ -380,36 +433,47 @@ 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 +483,29 @@ 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())
@@ -283,8 +304,7 @@ 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 +314,44 @@ 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 +412,15 @@ 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 +453,7 @@ 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))
@@ -518,3 +550,9 @@ class Notifier(object):
                 break
 
         defer.returnValue(result)
+
+    def wait_once_for_replication(self):
+        """Returns a deferred which resolves when there is new data for
+        replication to handle.
+        """
+        return self.replication_deferred.observe()

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

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/resource.py

@@ -283,12 +283,12 @@ class ReplicationResource(Resource):
 
             if request_events != upto_events_token:
                 writer.write_header_and_rows("events", res.new_forward_events, (
-                    "position", "internal", "json", "state_group"
+                    "position", "event_id", "room_id", "type", "state_key",
                 ), 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", "event_id", "room_id", "type", "state_key", "redacts",
                 ), position=upto_backfill_token)
 
             writer.write_header_and_rows(
@@ -489,7 +489,7 @@ class ReplicationResource(Resource):
 
         if federation is not None and federation != current_position:
             federation_rows = self.federation_sender.get_replication_rows(
-                federation, limit, federation_ack=federation_ack,
+                federation, current_position, limit, federation_ack=federation_ack,
             )
             upto_token = _position_from_rows(federation_rows, current_position)
             writer.write_header_and_rows("federation", federation_rows, (
@@ -504,7 +504,7 @@ class ReplicationResource(Resource):
 
         if device_lists is not None and device_lists != current_position:
             changes = yield self.store.get_all_device_list_changes_for_remotes(
-                device_lists,
+                device_lists, current_position,
             )
             writer.write_header_and_rows("device_lists", changes, (
                 "position", "user_id", "destination",

synapse/replication/slave/storage/_base.py

@@ -54,7 +54,9 @@ class BaseSlavedStore(SQLBaseStore):
                 try:
                     getattr(self, cache_func).invalidate(tuple(keys))
                 except AttributeError:
-                    logger.info("Got unexpected cache_func: %r", cache_func)
+                    # We probably haven't pulled in the cache in this worker,
+                    # which is fine.
+                    pass
             self._cache_id_gen.advance(int(stream["position"]))
         return defer.succeed(None)
 

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/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__

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
 
 
@@ -109,6 +107,10 @@ class SlavedEventStore(BaseSlavedStore):
     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__
@@ -165,7 +167,6 @@ 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__
@@ -238,46 +239,32 @@ class SlavedEventStore(BaseSlavedStore):
         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)
+        stream_ordering = row[0] if not backfilled else -row[0]
         self.invalidate_caches_for_event(
-            event, backfilled,
+            stream_ordering, row[1], row[2], row[3], row[4], row[5],
+            backfilled=backfilled,
         )
 
-    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

@@ -57,5 +57,6 @@ class SlavedPresenceStore(BaseSlavedStore):
                 self.presence_stream_cache.entity_has_changed(
                     user_id, position
                 )
+                self._get_presence_for_user.invalidate((user_id,))
 
         return super(SlavedPresenceStore, self).process_replication(result)

synapse/replication/tcp/__init__.py

@@ -0,0 +1,46 @@
+# -*- 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).
+
+The protocol is based on fire and forget, line based commands. An example flow
+would be (where '>' indicates master->worker and '<' worker->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.
+
+
+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
+
+Further details can be found in docs/tcp_replication.rst
+"""

synapse/replication/tcp/commands.py

@@ -0,0 +1,341 @@
+# -*- 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.
+
+    Format::
+
+        USER_SYNC <user_id> <state>
+
+    Where <state> is either "start" or "stop"
+    """
+    NAME = "USER_SYNC"
+
+    def __init__(self, user_id, is_syncing):
+        self.user_id = user_id
+        self.is_syncing = is_syncing
+
+    @classmethod
+    def from_line(cls, line):
+        user_id, state = line.split(" ", 1)
+
+        if state not in ("start", "end"):
+            raise Exception("Invalid USER_SYNC state %r" % (state,))
+
+        return cls(user_id, state == "start")
+
+    def to_line(self):
+        return " ".join((self.user_id, "start" if self.is_syncing else "end"))
+
+
+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 +1,601 @@
+# -*- 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 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
+
+import logging
+import synapse.metrics
+import struct
+import fcntl
+
+
+metrics = synapse.metrics.get_metrics_for(__name__)
+
+inbound_commands_counter = metrics.register_counter(
+    "inbound_commands", labels=["command", "name", "conn_id"],
+)
+outbound_commands_counter = metrics.register_counter(
+    "outbound_commands", labels=["command", "name", "conn_id"],
+)
+
+
+# A list of all connected protocols. This allows us to send metrics about the
+# connections.
+connected_connections = []
+
+
+logger = logging.getLogger(__name__)
+
+
+PING_TIME = 5000
+
+
+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
+
+    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_TIME * 3:
+                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_TIME * 3:
+                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()
+
+        inbound_commands_counter.inc(cmd_name, self.name, self.conn_id)
+
+        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):
+        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
+
+        outbound_commands_counter.inc(cmd.NAME, self.name, self.conn_id)
+
+        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)
+
+        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):
+        self.name = cmd.data
+
+    def on_USER_SYNC(self, cmd):
+        self.streamer.on_user_sync(self.conn_id, cmd.user_id, cmd.is_syncing)
+
+    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 onINVALIDATE_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))
+
+            # 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:
+                self.send_command(RdataCommand(stream_name, token, update))
+
+            # 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))
+
+            # 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.info("[%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)
+        logger.info("[%s] Replication connection closed", self.id())
+        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()
+        for user_id in currently_syncing:
+            self.send_command(UserSyncCommand(user_id, True))
+
+        # 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.transport.abortConnection()
+
+    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"],
+)

synapse/replication/tcp/resource.py

@@ -0,0 +1,300 @@
+# -*- 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.
+"""The server side of the replication stream.
+"""
+
+from twisted.internet import defer, reactor
+from twisted.internet.protocol import Factory
+
+from streams import STREAMS_MAP, FederationStream
+from protocol import ServerReplicationStreamProtocol
+
+from synapse.util.logcontext import preserve_fn
+from synapse.util.metrics import Measure, measure_func
+
+import logging
+import synapse.metrics
+
+
+metrics = synapse.metrics.get_metrics_for(__name__)
+stream_updates_counter = metrics.register_counter(
+    "stream_updates", labels=["stream_name"]
+)
+user_sync_counter = metrics.register_counter("user_sync")
+federation_ack_counter = metrics.register_counter("federation_ack")
+remove_pusher_counter = metrics.register_counter("remove_pusher")
+invalidate_cache_counter = metrics.register_counter("invalidate_cache")
+
+logger = logging.getLogger(__name__)
+
+
+class ReplicationStreamProtocolFactory(Factory):
+    """Factory for new replication connections.
+    """
+    def __init__(self, hs):
+        self.streamer = ReplicationStreamer(hs)
+        self.clock = hs.get_clock()
+        self.server_name = hs.config.server_name
+
+    def buildProtocol(self, addr):
+        return ServerReplicationStreamProtocol(
+            self.server_name,
+            self.clock,
+            self.streamer,
+            addr
+        )
+
+
+class ReplicationStreamer(object):
+    """Handles replication connections.
+
+    This needs to be poked when new replication data may be available. When new
+    data is available it will propagate to all connected clients.
+    """
+
+    def __init__(self, hs):
+        self.store = hs.get_datastore()
+        self.notifier = hs.get_notifier()
+        self.presence_handler = hs.get_presence_handler()
+        self.clock = hs.get_clock()
+
+        # Current connections.
+        self.connections = []
+
+        metrics.register_callback("total_connections", lambda: len(self.connections))
+
+        # List of streams that clients can subscribe to.
+        # We only support federation stream if federation sending hase been
+        # disabled on the master.
+        self.streams = [
+            stream(hs) for stream in STREAMS_MAP.itervalues()
+            if stream != FederationStream or not hs.config.send_federation
+        ]
+
+        self.streams_by_name = {stream.NAME: stream for stream in self.streams}
+
+        metrics.register_callback(
+            "connections_per_stream",
+            lambda: {
+                (stream_name,): len([
+                    conn for conn in self.connections
+                    if stream_name in conn.replication_streams
+                ])
+                for stream_name in self.streams_by_name
+            },
+            labels=["stream_name"],
+        )
+
+        self.federation_sender = None
+        if not hs.config.send_federation:
+            self.federation_sender = hs.get_federation_sender()
+
+        # Start listening for updates from the notifier
+        preserve_fn(self.notifier_listener)()
+
+        # Keeps track of whether we are currently checking for updates
+        self.is_looping = False
+        self.pending_updates = False
+
+        reactor.addSystemEventTrigger("before", "shutdown", self.on_shutdown)
+
+    def on_shutdown(self):
+        # close all connections on shutdown
+        for conn in self.connections:
+            conn.send_error("server shutting down")
+
+    @defer.inlineCallbacks
+    def notifier_listener(self):
+        """Sits forever looping on the notifier waiting for new data.
+        """
+        while True:
+            yield self.notifier.wait_once_for_replication()
+            logger.debug("Woken up by notifier")
+            # We need to call this each time we get woken up, as per docstring
+            preserve_fn(self.on_notifier_poke)()
+
+    @defer.inlineCallbacks
+    def on_notifier_poke(self):
+        """Checks if there is actually any new data and sends it to the
+        connections if there are.
+
+        This should get called each time new data is available, even if it
+        is currently being executed, so that nothing gets missed
+        """
+        if not self.connections:
+            # Don't bother if nothing is listening
+            return
+
+        # If we're in the process of checking for new updates, mark that fact
+        # and return
+        if self.is_looping:
+            logger.debug("Noitifier poke loop already running")
+            self.pending_updates = True
+            return
+
+        self.pending_updates = True
+        self.is_looping = True
+
+        try:
+            # Keep looping while there have been pokes about potential updates.
+            # This protects against the race where a stream we already checked
+            # gets an update while we're handling other streams.
+            while self.pending_updates:
+                self.pending_updates = False
+
+                with Measure(self.clock, "repl.stream.get_updates"):
+                    # First we tell the streams that they should update their
+                    # current tokens.
+                    for stream in self.streams:
+                        stream.advance_current_token()
+
+                    for stream in self.streams:
+                        if stream.last_token == stream.upto_token:
+                            continue
+
+                        logger.debug(
+                            "Getting stream: %s: %s -> %s",
+                            stream.NAME, stream.last_token, stream.upto_token
+                        )
+                        updates, current_token = yield stream.get_updates()
+
+                        logger.debug(
+                            "Sending %d updates to %d connections",
+                            len(updates), len(self.connections),
+                        )
+
+                        if updates:
+                            logger.info(
+                                "Streaming: %s -> %s", stream.NAME, updates[-1][0]
+                            )
+                            stream_updates_counter.inc_by(len(updates), stream.NAME)
+
+                        # Some streams return multiple rows with the same stream IDs,
+                        # we need to make sure they get sent out in batches. We do
+                        # this by setting the current token to all but the last of
+                        # a series of updates with the same token to have a None
+                        # token. See RdataCommand for more details.
+                        batched_updates = _batch_updates(updates)
+
+                        for conn in self.connections:
+                            for token, row in batched_updates:
+                                try:
+                                    conn.stream_update(stream.NAME, token, row)
+                                except Exception:
+                                    logger.exception("Failed to replicate")
+
+            logger.debug("No more pending updates, breaking poke loop")
+        finally:
+            self.pending_updates = False
+            self.is_looping = False
+
+    @measure_func("repl.get_stream_updates")
+    def get_stream_updates(self, stream_name, token):
+        """For a given stream get all updates since token. This is called when
+        a client first subscribes to a stream.
+        """
+        stream = self.streams_by_name.get(stream_name, None)
+        if not stream:
+            raise Exception("unknown stream %s", stream_name)
+
+        return stream.get_updates_since(token)
+
+    @measure_func("repl.federation_ack")
+    def federation_ack(self, token):
+        """We've received an ack for federation stream from a client.
+        """
+        federation_ack_counter.inc()
+        if self.federation_sender:
+            self.federation_sender.federation_ack(token)
+
+    @measure_func("repl.on_user_sync")
+    def on_user_sync(self, conn_id, user_id, is_syncing):
+        """A client has started/stopped syncing on a worker.
+        """
+        user_sync_counter.inc()
+        self.presence_handler.update_external_syncs_row(
+            conn_id, user_id, is_syncing
+        )
+
+    @measure_func("repl.on_remove_pusher")
+    @defer.inlineCallbacks
+    def on_remove_pusher(self, app_id, push_key, user_id):
+        """A client has asked us to remove a pusher
+        """
+        remove_pusher_counter.inc()
+        yield self.store.delete_pusher_by_app_id_pushkey_user_id(
+            app_id=app_id, pushkey=push_key, user_id=user_id
+        )
+
+        self.notifier.on_new_replication_data()
+
+    @measure_func("repl.on_invalidate_cache")
+    def on_invalidate_cache(self, cache_func, keys):
+        """The client has asked us to invalidate a cache
+        """
+        invalidate_cache_counter.inc()
+        getattr(self.store, cache_func).invalidate(tuple(keys))
+
+    def send_sync_to_all_connections(self, data):
+        """Sends a SYNC command to all clients.
+
+        Used in tests.
+        """
+        for conn in self.connections:
+            conn.send_sync(data)
+
+    def new_connection(self, connection):
+        """A new client connection has been established
+        """
+        self.connections.append(connection)
+
+    def lost_connection(self, connection):
+        """A client connection has been lost
+        """
+        try:
+            self.connections.remove(connection)
+        except ValueError:
+            pass
+
+        # We need to tell the presence handler that the connection has been
+        # lost so that it can handle any ongoing syncs on that connection.
+        self.presence_handler.update_external_syncs_clear(connection.conn_id)
+
+
+def _batch_updates(updates):
+    """Takes a list of updates of form [(token, row)] and sets the token to
+    None for all rows where the next row has the same token. This is used to
+    implement batching.
+
+    For example:
+
+        [(1, _), (1, _), (2, _), (3, _), (3, _)]
+
+    becomes:
+
+        [(None, _), (1, _), (2, _), (None, _), (3, _)]
+    """
+    if not updates:
+        return []
+
+    new_updates = []
+    for i, update in enumerate(updates[:-1]):
+        if update[0] == updates[i + 1][0]:
+            new_updates.append((None, update[1]))
+        else:
+            new_updates.append(update)
+
+    new_updates.append(updates[-1])
+    return new_updates

synapse/replication/tcp/streams.py

@@ -0,0 +1,409 @@
+# -*- 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 all the valid streams that clients can subscribe to, and the format
+of the rows returned by each stream.
+
+Each stream is defined by the following information:
+
+    stream name:        The name of the stream
+    row type:           The type that is used to serialise/deserialse the row
+    current_token:      The function that returns the current token for the stream
+    update_function:    The function that returns a list of updates between two tokens
+"""
+
+from twisted.internet import defer
+from collections import namedtuple
+
+import logging
+
+
+logger = logging.getLogger(__name__)
+
+
+MAX_EVENTS_BEHIND = 10000
+
+
+EventStreamRow = namedtuple("EventStreamRow",
+                            ("event_id", "room_id", "type", "state_key", "redacts"))
+BackfillStreamRow = namedtuple("BackfillStreamRow",
+                               ("event_id", "room_id", "type", "state_key", "redacts"))
+PresenceStreamRow = namedtuple("PresenceStreamRow",
+                               ("user_id", "state", "last_active_ts",
+                                "last_federation_update_ts", "last_user_sync_ts",
+                                "status_msg", "currently_active"))
+TypingStreamRow = namedtuple("TypingStreamRow",
+                             ("room_id", "user_ids"))
+ReceiptsStreamRow = namedtuple("ReceiptsStreamRow",
+                               ("room_id", "receipt_type", "user_id", "event_id",
+                                "data"))
+PushRulesStreamRow = namedtuple("PushRulesStreamRow", ("user_id",))
+PushersStreamRow = namedtuple("PushersStreamRow",
+                              ("user_id", "app_id", "pushkey", "deleted",))
+CachesStreamRow = namedtuple("CachesStreamRow",
+                             ("cache_func", "keys", "invalidation_ts",))
+PublicRoomsStreamRow = namedtuple("PublicRoomsStreamRow",
+                                  ("room_id", "visibility", "appservice_id",
+                                   "network_id",))
+DeviceListsStreamRow = namedtuple("DeviceListsStreamRow", ("user_id", "destination",))
+ToDeviceStreamRow = namedtuple("ToDeviceStreamRow", ("entity",))
+FederationStreamRow = namedtuple("FederationStreamRow", ("type", "data",))
+TagAccountDataStreamRow = namedtuple("TagAccountDataStreamRow",
+                                     ("user_id", "room_id", "data"))
+AccountDataStreamRow = namedtuple("AccountDataStream",
+                                  ("user_id", "room_id", "data_type", "data"))
+
+
+class Stream(object):
+    """Base class for the streams.
+
+    Provides a `get_updates()` function that returns new updates since the last
+    time it was called up until the point `advance_current_token` was called.
+    """
+    NAME = None  # The name of the stream
+    ROW_TYPE = None  # The type of the row
+    _LIMITED = True  # Whether the update function takes a limit
+
+    def __init__(self, hs):
+        # The token from which we last asked for updates
+        self.last_token = self.current_token()
+
+        # The token that we will get updates up to
+        self.upto_token = self.current_token()
+
+    def advance_current_token(self):
+        """Updates `upto_token` to "now", which updates up until which point
+        get_updates[_since] will fetch rows till.
+        """
+        self.upto_token = self.current_token()
+
+    @defer.inlineCallbacks
+    def get_updates(self):
+        """Gets all updates since the last time this function was called (or
+        since the stream was constructed if it hadn't been called before),
+        until the `upto_token`
+
+        Returns:
+            (list(ROW_TYPE), int): list of updates plus the token used as an
+                upper bound of the updates (i.e. the "current token")
+        """
+        updates, current_token = yield self.get_updates_since(self.last_token)
+        self.last_token = current_token
+
+        defer.returnValue((updates, current_token))
+
+    @defer.inlineCallbacks
+    def get_updates_since(self, from_token):
+        """Like get_updates except allows specifying from when we should
+        stream updates
+
+        Returns:
+            (list(ROW_TYPE), int): list of updates plus the token used as an
+                upper bound of the updates (i.e. the "current token")
+        """
+        if from_token in ("NOW", "now"):
+            defer.returnValue(([], self.upto_token))
+
+        current_token = self.upto_token
+
+        from_token = int(from_token)
+
+        if from_token == current_token:
+            defer.returnValue(([], current_token))
+
+        if self._LIMITED:
+            rows = yield self.update_function(
+                from_token, current_token,
+                limit=MAX_EVENTS_BEHIND + 1,
+            )
+
+            if len(rows) >= MAX_EVENTS_BEHIND:
+                raise Exception("stream %s has fallen behined" % (self.NAME))
+        else:
+            rows = yield self.update_function(
+                from_token, current_token,
+            )
+
+        updates = [(row[0], self.ROW_TYPE(*row[1:])) for row in rows]
+
+        defer.returnValue((updates, current_token))
+
+    def current_token(self):
+        """Gets the current token of the underlying streams. Should be provided
+        by the sub classes
+
+        Returns:
+            int
+        """
+        raise NotImplementedError()
+
+    def update_function(self, from_token, current_token, limit=None):
+        """Get updates between from_token and to_token. If Stream._LIMITED is
+        True then limit is provided, otherwise it's not.
+
+        Returns:
+            list(tuple): the first entry in the tuple is the token for that
+                update, and the rest of the tuple gets used to construct
+                a ``ROW_TYPE`` instance
+        """
+        raise NotImplementedError()
+
+
+class EventsStream(Stream):
+    """We received a new event, or an event went from being an outlier to not
+    """
+    NAME = "events"
+    ROW_TYPE = EventStreamRow
+
+    def __init__(self, hs):
+        store = hs.get_datastore()
+        self.current_token = store.get_current_events_token
+        self.update_function = store.get_all_new_forward_event_rows
+
+        super(EventsStream, self).__init__(hs)
+
+
+class BackfillStream(Stream):
+    """We fetched some old events and either we had never seen that event before
+    or it went from being an outlier to not.
+    """
+    NAME = "backfill"
+    ROW_TYPE = BackfillStreamRow
+
+    def __init__(self, hs):
+        store = hs.get_datastore()
+        self.current_token = store.get_current_backfill_token
+        self.update_function = store.get_all_new_backfill_event_rows
+
+        super(BackfillStream, self).__init__(hs)
+
+
+class PresenceStream(Stream):
+    NAME = "presence"
+    _LIMITED = False
+    ROW_TYPE = PresenceStreamRow
+
+    def __init__(self, hs):
+        store = hs.get_datastore()
+        presence_handler = hs.get_presence_handler()
+
+        self.current_token = store.get_current_presence_token
+        self.update_function = presence_handler.get_all_presence_updates
+
+        super(PresenceStream, self).__init__(hs)
+
+
+class TypingStream(Stream):
+    NAME = "typing"
+    _LIMITED = False
+    ROW_TYPE = TypingStreamRow
+
+    def __init__(self, hs):
+        typing_handler = hs.get_typing_handler()
+
+        self.current_token = typing_handler.get_current_token
+        self.update_function = typing_handler.get_all_typing_updates
+
+        super(TypingStream, self).__init__(hs)
+
+
+class ReceiptsStream(Stream):
+    NAME = "receipts"
+    ROW_TYPE = ReceiptsStreamRow
+
+    def __init__(self, hs):
+        store = hs.get_datastore()
+
+        self.current_token = store.get_max_receipt_stream_id
+        self.update_function = store.get_all_updated_receipts
+
+        super(ReceiptsStream, self).__init__(hs)
+
+
+class PushRulesStream(Stream):
+    """A user has changed their push rules
+    """
+    NAME = "push_rules"
+    ROW_TYPE = PushRulesStreamRow
+
+    def __init__(self, hs):
+        self.store = hs.get_datastore()
+        super(PushRulesStream, self).__init__(hs)
+
+    def current_token(self):
+        push_rules_token, _ = self.store.get_push_rules_stream_token()
+        return push_rules_token
+
+    @defer.inlineCallbacks
+    def update_function(self, from_token, to_token, limit):
+        rows = yield self.store.get_all_push_rule_updates(from_token, to_token, limit)
+        defer.returnValue([(row[0], row[2]) for row in rows])
+
+
+class PushersStream(Stream):
+    """A user has added/changed/removed a pusher
+    """
+    NAME = "pushers"
+    ROW_TYPE = PushersStreamRow
+
+    def __init__(self, hs):
+        store = hs.get_datastore()
+
+        self.current_token = store.get_pushers_stream_token
+        self.update_function = store.get_all_updated_pushers_rows
+
+        super(PushersStream, self).__init__(hs)
+
+
+class CachesStream(Stream):
+    """A cache was invalidated on the master and no other stream would invalidate
+    the cache on the workers
+    """
+    NAME = "caches"
+    ROW_TYPE = CachesStreamRow
+
+    def __init__(self, hs):
+        store = hs.get_datastore()
+
+        self.current_token = store.get_cache_stream_token
+        self.update_function = store.get_all_updated_caches
+
+        super(CachesStream, self).__init__(hs)
+
+
+class PublicRoomsStream(Stream):
+    """The public rooms list changed
+    """
+    NAME = "public_rooms"
+    ROW_TYPE = PublicRoomsStreamRow
+
+    def __init__(self, hs):
+        store = hs.get_datastore()
+
+        self.current_token = store.get_current_public_room_stream_id
+        self.update_function = store.get_all_new_public_rooms
+
+        super(PublicRoomsStream, self).__init__(hs)
+
+
+class DeviceListsStream(Stream):
+    """Someone added/changed/removed a device
+    """
+    NAME = "device_lists"
+    _LIMITED = False
+    ROW_TYPE = DeviceListsStreamRow
+
+    def __init__(self, hs):
+        store = hs.get_datastore()
+
+        self.current_token = store.get_device_stream_token
+        self.update_function = store.get_all_device_list_changes_for_remotes
+
+        super(DeviceListsStream, self).__init__(hs)
+
+
+class ToDeviceStream(Stream):
+    """New to_device messages for a client
+    """
+    NAME = "to_device"
+    ROW_TYPE = ToDeviceStreamRow
+
+    def __init__(self, hs):
+        store = hs.get_datastore()
+
+        self.current_token = store.get_to_device_stream_token
+        self.update_function = store.get_all_new_device_messages
+
+        super(ToDeviceStream, self).__init__(hs)
+
+
+class FederationStream(Stream):
+    """Data to be sent over federation. Only available when master has federation
+    sending disabled.
+    """
+    NAME = "federation"
+    ROW_TYPE = FederationStreamRow
+
+    def __init__(self, hs):
+        federation_sender = hs.get_federation_sender()
+
+        self.current_token = federation_sender.get_current_token
+        self.update_function = federation_sender.get_replication_rows
+
+        super(FederationStream, self).__init__(hs)
+
+
+class TagAccountDataStream(Stream):
+    """Someone added/removed a tag for a room
+    """
+    NAME = "tag_account_data"
+    ROW_TYPE = TagAccountDataStreamRow
+
+    def __init__(self, hs):
+        store = hs.get_datastore()
+
+        self.current_token = store.get_max_account_data_stream_id
+        self.update_function = store.get_all_updated_tags
+
+        super(TagAccountDataStream, self).__init__(hs)
+
+
+class AccountDataStream(Stream):
+    """Global or per room account data was changed
+    """
+    NAME = "account_data"
+    ROW_TYPE = AccountDataStreamRow
+
+    def __init__(self, hs):
+        self.store = hs.get_datastore()
+
+        self.current_token = self.store.get_max_account_data_stream_id
+
+        super(AccountDataStream, self).__init__(hs)
+
+    @defer.inlineCallbacks
+    def update_function(self, from_token, to_token, limit):
+        global_results, room_results = yield self.store.get_all_updated_account_data(
+            from_token, from_token, to_token, limit
+        )
+
+        results = list(room_results)
+        results.extend(
+            (stream_id, user_id, None, account_data_type, content,)
+            for stream_id, user_id, account_data_type, content in global_results
+        )
+
+        defer.returnValue(results)
+
+
+STREAMS_MAP = {
+    stream.NAME: stream
+    for stream in (
+        EventsStream,
+        BackfillStream,
+        PresenceStream,
+        TypingStream,
+        ReceiptsStream,
+        PushRulesStream,
+        PushersStream,
+        CachesStream,
+        PublicRoomsStream,
+        DeviceListsStream,
+        ToDeviceStream,
+        FederationStream,
+        TagAccountDataStream,
+        AccountDataStream,
+    )
+}

synapse/rest/client/v1/login.py

@@ -19,6 +19,7 @@ from synapse.api.errors import SynapseError, LoginError, Codes
 from synapse.types import UserID
 from synapse.http.server import finish_request
 from synapse.http.servlet import parse_json_object_from_request
+from synapse.util.msisdn import phone_number_to_msisdn
 
 from .base import ClientV1RestServlet, client_path_patterns
 
@@ -33,10 +34,55 @@ from saml2.client import Saml2Client
 
 import xml.etree.ElementTree as ET
 
+from twisted.web.client import PartialDownloadError
+
 
 logger = logging.getLogger(__name__)
 
 
+def login_submission_legacy_convert(submission):
+    """
+    If the input login submission is an old style object
+    (ie. with top-level user / medium / address) convert it
+    to a typed object.
+    """
+    if "user" in submission:
+        submission["identifier"] = {
+            "type": "m.id.user",
+            "user": submission["user"],
+        }
+        del submission["user"]
+
+    if "medium" in submission and "address" in submission:
+        submission["identifier"] = {
+            "type": "m.id.thirdparty",
+            "medium": submission["medium"],
+            "address": submission["address"],
+        }
+        del submission["medium"]
+        del submission["address"]
+
+
+def login_id_thirdparty_from_phone(identifier):
+    """
+    Convert a phone login identifier type to a generic threepid identifier
+    Args:
+        identifier(dict): Login identifier dict of type 'm.id.phone'
+
+    Returns: Login identifier dict of type 'm.id.threepid'
+    """
+    if "country" not in identifier or "number" not in identifier:
+        raise SynapseError(400, "Invalid phone-type identifier")
+
+    msisdn = phone_number_to_msisdn(identifier["country"], identifier["number"])
+
+    return {
+        "type": "m.id.thirdparty",
+        "medium": "msisdn",
+        "address": msisdn,
+    }
+
+
 class LoginRestServlet(ClientV1RestServlet):
     PATTERNS = client_path_patterns("/login$")
     PASS_TYPE = "m.login.password"
@@ -117,20 +163,52 @@ class LoginRestServlet(ClientV1RestServlet):
 
     @defer.inlineCallbacks
     def do_password_login(self, login_submission):
-        if 'medium' in login_submission and 'address' in login_submission:
-            address = login_submission['address']
-            if login_submission['medium'] == 'email':
+        if "password" not in login_submission:
+            raise SynapseError(400, "Missing parameter: password")
+
+        login_submission_legacy_convert(login_submission)
+
+        if "identifier" not in login_submission:
+            raise SynapseError(400, "Missing param: identifier")
+
+        identifier = login_submission["identifier"]
+        if "type" not in identifier:
+            raise SynapseError(400, "Login identifier has no type")
+
+        # convert phone type identifiers to generic threepids
+        if identifier["type"] == "m.id.phone":
+            identifier = login_id_thirdparty_from_phone(identifier)
+
+        # convert threepid identifiers to user IDs
+        if identifier["type"] == "m.id.thirdparty":
+            if 'medium' not in identifier or 'address' not in identifier:
+                raise SynapseError(400, "Invalid thirdparty identifier")
+
+            address = identifier['address']
+            if identifier['medium'] == 'email':
                 # For emails, transform the address to lowercase.
                 # We store all email addreses as lowercase in the DB.
                 # (See add_threepid in synapse/handlers/auth.py)
                 address = address.lower()
             user_id = yield self.hs.get_datastore().get_user_id_by_threepid(
-                login_submission['medium'], address
+                identifier['medium'], address
             )
             if not user_id:
                 raise LoginError(403, "", errcode=Codes.FORBIDDEN)
-        else:
-            user_id = login_submission['user']
+
+            identifier = {
+                "type": "m.id.user",
+                "user": user_id,
+            }
+
+        # by this point, the identifier should be an m.id.user: if it's anything
+        # else, we haven't understood it.
+        if identifier["type"] != "m.id.user":
+            raise SynapseError(400, "Unknown login identifier type")
+        if "user" not in identifier:
+            raise SynapseError(400, "User identifier is missing 'user' key")
+
+        user_id = identifier["user"]
 
         if not user_id.startswith('@'):
             user_id = UserID.create(
@@ -341,7 +419,12 @@ class CasTicketServlet(ClientV1RestServlet):
             "ticket": request.args["ticket"],
             "service": self.cas_service_url
         }
-        body = yield http_client.get_raw(uri, args)
+        try:
+            body = yield http_client.get_raw(uri, args)
+        except PartialDownloadError as pde:
+            # Twisted raises this error if the connection is closed,
+            # even if that's being used old-http style to signal end-of-data
+            body = pde.response
         result = yield self.handle_cas_response(request, body, client_redirect_url)
         defer.returnValue(result)
 

synapse/rest/client/v1/presence.py

@@ -19,6 +19,7 @@ from twisted.internet import defer
 
 from synapse.api.errors import SynapseError, AuthError
 from synapse.types import UserID
+from synapse.handlers.presence import format_user_presence_state
 from synapse.http.servlet import parse_json_object_from_request
 from .base import ClientV1RestServlet, client_path_patterns
 
@@ -33,6 +34,7 @@ class PresenceStatusRestServlet(ClientV1RestServlet):
     def __init__(self, hs):
         super(PresenceStatusRestServlet, self).__init__(hs)
         self.presence_handler = hs.get_presence_handler()
+        self.clock = hs.get_clock()
 
     @defer.inlineCallbacks
     def on_GET(self, request, user_id):
@@ -48,6 +50,7 @@ class PresenceStatusRestServlet(ClientV1RestServlet):
                 raise AuthError(403, "You are not allowed to see their presence.")
 
         state = yield self.presence_handler.get_state(target_user=user)
+        state = format_user_presence_state(state, self.clock.time_msec())
 
         defer.returnValue((200, state))
 

synapse/rest/client/v1/room.py

@@ -749,8 +749,7 @@ class JoinedRoomsRestServlet(ClientV1RestServlet):
     def on_GET(self, request):
         requester = yield self.auth.get_user_by_req(request, allow_guest=True)
 
-        rooms = yield self.store.get_rooms_for_user(requester.user.to_string())
-        room_ids = set(r.room_id for r in rooms)  # Ensure they're unique.
+        room_ids = yield self.store.get_rooms_for_user(requester.user.to_string())
         defer.returnValue((200, {"joined_rooms": list(room_ids)}))
 
 

synapse/rest/client/v2_alpha/account.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.
@@ -17,8 +18,11 @@ from twisted.internet import defer
 
 from synapse.api.constants import LoginType
 from synapse.api.errors import LoginError, SynapseError, Codes
-from synapse.http.servlet import RestServlet, parse_json_object_from_request
+from synapse.http.servlet import (
+    RestServlet, parse_json_object_from_request, assert_params_in_request
+)
 from synapse.util.async import run_on_reactor
+from synapse.util.msisdn import phone_number_to_msisdn
 
 from ._base import client_v2_patterns
 
@@ -28,11 +32,11 @@ import logging
 logger = logging.getLogger(__name__)
 
 
-class PasswordRequestTokenRestServlet(RestServlet):
+class EmailPasswordRequestTokenRestServlet(RestServlet):
     PATTERNS = client_v2_patterns("/account/password/email/requestToken$")
 
     def __init__(self, hs):
-        super(PasswordRequestTokenRestServlet, self).__init__()
+        super(EmailPasswordRequestTokenRestServlet, self).__init__()
         self.hs = hs
         self.identity_handler = hs.get_handlers().identity_handler
 
@@ -40,14 +44,9 @@ class PasswordRequestTokenRestServlet(RestServlet):
     def on_POST(self, request):
         body = parse_json_object_from_request(request)
 
-        required = ['id_server', 'client_secret', 'email', 'send_attempt']
-        absent = []
-        for k in required:
-            if k not in body:
-                absent.append(k)
-
-        if absent:
-            raise SynapseError(400, "Missing params: %r" % absent, Codes.MISSING_PARAM)
+        assert_params_in_request(body, [
+            'id_server', 'client_secret', 'email', 'send_attempt'
+        ])
 
         existingUid = yield self.hs.get_datastore().get_user_id_by_threepid(
             'email', body['email']
@@ -60,6 +59,37 @@ class PasswordRequestTokenRestServlet(RestServlet):
         defer.returnValue((200, ret))
 
 
+class MsisdnPasswordRequestTokenRestServlet(RestServlet):
+    PATTERNS = client_v2_patterns("/account/password/msisdn/requestToken$")
+
+    def __init__(self, hs):
+        super(MsisdnPasswordRequestTokenRestServlet, self).__init__()
+        self.hs = hs
+        self.datastore = self.hs.get_datastore()
+        self.identity_handler = hs.get_handlers().identity_handler
+
+    @defer.inlineCallbacks
+    def on_POST(self, request):
+        body = parse_json_object_from_request(request)
+
+        assert_params_in_request(body, [
+            'id_server', 'client_secret',
+            'country', 'phone_number', 'send_attempt',
+        ])
+
+        msisdn = phone_number_to_msisdn(body['country'], body['phone_number'])
+
+        existingUid = yield self.datastore.get_user_id_by_threepid(
+            'msisdn', msisdn
+        )
+
+        if existingUid is None:
+            raise SynapseError(400, "MSISDN not found", Codes.THREEPID_NOT_FOUND)
+
+        ret = yield self.identity_handler.requestMsisdnToken(**body)
+        defer.returnValue((200, ret))
+
+
 class PasswordRestServlet(RestServlet):
     PATTERNS = client_v2_patterns("/account/password$")
 
@@ -68,6 +98,7 @@ class PasswordRestServlet(RestServlet):
         self.hs = hs
         self.auth = hs.get_auth()
         self.auth_handler = hs.get_auth_handler()
+        self.datastore = self.hs.get_datastore()
 
     @defer.inlineCallbacks
     def on_POST(self, request):
@@ -77,7 +108,8 @@ class PasswordRestServlet(RestServlet):
 
         authed, result, params, _ = yield self.auth_handler.check_auth([
             [LoginType.PASSWORD],
-            [LoginType.EMAIL_IDENTITY]
+            [LoginType.EMAIL_IDENTITY],
+            [LoginType.MSISDN],
         ], body, self.hs.get_ip_from_request(request))
 
         if not authed:
@@ -102,7 +134,7 @@ class PasswordRestServlet(RestServlet):
                 # (See add_threepid in synapse/handlers/auth.py)
                 threepid['address'] = threepid['address'].lower()
             # if using email, we must know about the email they're authing with!
-            threepid_user_id = yield self.hs.get_datastore().get_user_id_by_threepid(
+            threepid_user_id = yield self.datastore.get_user_id_by_threepid(
                 threepid['medium'], threepid['address']
             )
             if not threepid_user_id:
@@ -169,13 +201,14 @@ class DeactivateAccountRestServlet(RestServlet):
         defer.returnValue((200, {}))
 
 
-class ThreepidRequestTokenRestServlet(RestServlet):
+class EmailThreepidRequestTokenRestServlet(RestServlet):
     PATTERNS = client_v2_patterns("/account/3pid/email/requestToken$")
 
     def __init__(self, hs):
         self.hs = hs
-        super(ThreepidRequestTokenRestServlet, self).__init__()
+        super(EmailThreepidRequestTokenRestServlet, self).__init__()
         self.identity_handler = hs.get_handlers().identity_handler
+        self.datastore = self.hs.get_datastore()
 
     @defer.inlineCallbacks
     def on_POST(self, request):
@@ -190,7 +223,7 @@ class ThreepidRequestTokenRestServlet(RestServlet):
         if absent:
             raise SynapseError(400, "Missing params: %r" % absent, Codes.MISSING_PARAM)
 
-        existingUid = yield self.hs.get_datastore().get_user_id_by_threepid(
+        existingUid = yield self.datastore.get_user_id_by_threepid(
             'email', body['email']
         )
 
@@ -201,6 +234,44 @@ class ThreepidRequestTokenRestServlet(RestServlet):
         defer.returnValue((200, ret))
 
 
+class MsisdnThreepidRequestTokenRestServlet(RestServlet):
+    PATTERNS = client_v2_patterns("/account/3pid/msisdn/requestToken$")
+
+    def __init__(self, hs):
+        self.hs = hs
+        super(MsisdnThreepidRequestTokenRestServlet, self).__init__()
+        self.identity_handler = hs.get_handlers().identity_handler
+        self.datastore = self.hs.get_datastore()
+
+    @defer.inlineCallbacks
+    def on_POST(self, request):
+        body = parse_json_object_from_request(request)
+
+        required = [
+            'id_server', 'client_secret',
+            'country', 'phone_number', 'send_attempt',
+        ]
+        absent = []
+        for k in required:
+            if k not in body:
+                absent.append(k)
+
+        if absent:
+            raise SynapseError(400, "Missing params: %r" % absent, Codes.MISSING_PARAM)
+
+        msisdn = phone_number_to_msisdn(body['country'], body['phone_number'])
+
+        existingUid = yield self.datastore.get_user_id_by_threepid(
+            'msisdn', msisdn
+        )
+
+        if existingUid is not None:
+            raise SynapseError(400, "MSISDN is already in use", Codes.THREEPID_IN_USE)
+
+        ret = yield self.identity_handler.requestMsisdnToken(**body)
+        defer.returnValue((200, ret))
+
+
 class ThreepidRestServlet(RestServlet):
     PATTERNS = client_v2_patterns("/account/3pid$")
 
@@ -210,6 +281,7 @@ class ThreepidRestServlet(RestServlet):
         self.identity_handler = hs.get_handlers().identity_handler
         self.auth = hs.get_auth()
         self.auth_handler = hs.get_auth_handler()
+        self.datastore = self.hs.get_datastore()
 
     @defer.inlineCallbacks
     def on_GET(self, request):
@@ -217,7 +289,7 @@ class ThreepidRestServlet(RestServlet):
 
         requester = yield self.auth.get_user_by_req(request)
 
-        threepids = yield self.hs.get_datastore().user_get_threepids(
+        threepids = yield self.datastore.user_get_threepids(
             requester.user.to_string()
         )
 
@@ -258,7 +330,7 @@ class ThreepidRestServlet(RestServlet):
 
         if 'bind' in body and body['bind']:
             logger.debug(
-                "Binding emails %s to %s",
+                "Binding threepid %s to %s",
                 threepid, user_id
             )
             yield self.identity_handler.bind_threepid(
@@ -302,9 +374,11 @@ class ThreepidDeleteRestServlet(RestServlet):
 
 
 def register_servlets(hs, http_server):
-    PasswordRequestTokenRestServlet(hs).register(http_server)
+    EmailPasswordRequestTokenRestServlet(hs).register(http_server)
+    MsisdnPasswordRequestTokenRestServlet(hs).register(http_server)
     PasswordRestServlet(hs).register(http_server)
     DeactivateAccountRestServlet(hs).register(http_server)
-    ThreepidRequestTokenRestServlet(hs).register(http_server)
+    EmailThreepidRequestTokenRestServlet(hs).register(http_server)
+    MsisdnThreepidRequestTokenRestServlet(hs).register(http_server)
     ThreepidRestServlet(hs).register(http_server)
     ThreepidDeleteRestServlet(hs).register(http_server)

synapse/rest/client/v2_alpha/devices.py

@@ -46,6 +46,52 @@ class DevicesRestServlet(servlet.RestServlet):
         defer.returnValue((200, {"devices": devices}))
 
 
+class DeleteDevicesRestServlet(servlet.RestServlet):
+    """
+    API for bulk deletion of devices. Accepts a JSON object with a devices
+    key which lists the device_ids to delete. Requires user interactive auth.
+    """
+    PATTERNS = client_v2_patterns("/delete_devices", releases=[], v2_alpha=False)
+
+    def __init__(self, hs):
+        super(DeleteDevicesRestServlet, self).__init__()
+        self.hs = hs
+        self.auth = hs.get_auth()
+        self.device_handler = hs.get_device_handler()
+        self.auth_handler = hs.get_auth_handler()
+
+    @defer.inlineCallbacks
+    def on_POST(self, request):
+        try:
+            body = servlet.parse_json_object_from_request(request)
+        except errors.SynapseError as e:
+            if e.errcode == errors.Codes.NOT_JSON:
+                # deal with older clients which didn't pass a J*DELETESON dict
+                # the same as those that pass an empty dict
+                body = {}
+            else:
+                raise e
+
+        if 'devices' not in body:
+            raise errors.SynapseError(
+                400, "No devices supplied", errcode=errors.Codes.MISSING_PARAM
+            )
+
+        authed, result, params, _ = yield self.auth_handler.check_auth([
+            [constants.LoginType.PASSWORD],
+        ], body, self.hs.get_ip_from_request(request))
+
+        if not authed:
+            defer.returnValue((401, result))
+
+        requester = yield self.auth.get_user_by_req(request)
+        yield self.device_handler.delete_devices(
+            requester.user.to_string(),
+            body['devices'],
+        )
+        defer.returnValue((200, {}))
+
+
 class DeviceRestServlet(servlet.RestServlet):
     PATTERNS = client_v2_patterns("/devices/(?P<device_id>[^/]*)$",
                                   releases=[], v2_alpha=False)
@@ -111,5 +157,6 @@ class DeviceRestServlet(servlet.RestServlet):
 
 
 def register_servlets(hs, http_server):
+    DeleteDevicesRestServlet(hs).register(http_server)
     DevicesRestServlet(hs).register(http_server)
     DeviceRestServlet(hs).register(http_server)

synapse/rest/client/v2_alpha/register.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.
@@ -19,7 +20,10 @@ import synapse
 from synapse.api.auth import get_access_token_from_request, has_access_token
 from synapse.api.constants import LoginType
 from synapse.api.errors import SynapseError, Codes, UnrecognizedRequestError
-from synapse.http.servlet import RestServlet, parse_json_object_from_request
+from synapse.http.servlet import (
+    RestServlet, parse_json_object_from_request, assert_params_in_request
+)
+from synapse.util.msisdn import phone_number_to_msisdn
 
 from ._base import client_v2_patterns
 
@@ -43,7 +47,7 @@ else:
 logger = logging.getLogger(__name__)
 
 
-class RegisterRequestTokenRestServlet(RestServlet):
+class EmailRegisterRequestTokenRestServlet(RestServlet):
     PATTERNS = client_v2_patterns("/register/email/requestToken$")
 
     def __init__(self, hs):
@@ -51,7 +55,7 @@ class RegisterRequestTokenRestServlet(RestServlet):
         Args:
             hs (synapse.server.HomeServer): server
         """
-        super(RegisterRequestTokenRestServlet, self).__init__()
+        super(EmailRegisterRequestTokenRestServlet, self).__init__()
         self.hs = hs
         self.identity_handler = hs.get_handlers().identity_handler
 
@@ -59,14 +63,9 @@ class RegisterRequestTokenRestServlet(RestServlet):
     def on_POST(self, request):
         body = parse_json_object_from_request(request)
 
-        required = ['id_server', 'client_secret', 'email', 'send_attempt']
-        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)
+        assert_params_in_request(body, [
+            'id_server', 'client_secret', 'email', 'send_attempt'
+        ])
 
         existingUid = yield self.hs.get_datastore().get_user_id_by_threepid(
             'email', body['email']
@@ -79,6 +78,43 @@ class RegisterRequestTokenRestServlet(RestServlet):
         defer.returnValue((200, ret))
 
 
+class MsisdnRegisterRequestTokenRestServlet(RestServlet):
+    PATTERNS = client_v2_patterns("/register/msisdn/requestToken$")
+
+    def __init__(self, hs):
+        """
+        Args:
+            hs (synapse.server.HomeServer): server
+        """
+        super(MsisdnRegisterRequestTokenRestServlet, self).__init__()
+        self.hs = hs
+        self.identity_handler = hs.get_handlers().identity_handler
+
+    @defer.inlineCallbacks
+    def on_POST(self, request):
+        body = parse_json_object_from_request(request)
+
+        assert_params_in_request(body, [
+            'id_server', 'client_secret',
+            'country', 'phone_number',
+            'send_attempt',
+        ])
+
+        msisdn = phone_number_to_msisdn(body['country'], body['phone_number'])
+
+        existingUid = yield self.hs.get_datastore().get_user_id_by_threepid(
+            'msisdn', msisdn
+        )
+
+        if existingUid is not None:
+            raise SynapseError(
+                400, "Phone number is already in use", Codes.THREEPID_IN_USE
+            )
+
+        ret = yield self.identity_handler.requestMsisdnToken(**body)
+        defer.returnValue((200, ret))
+
+
 class RegisterRestServlet(RestServlet):
     PATTERNS = client_v2_patterns("/register$")
 
@@ -200,16 +236,37 @@ class RegisterRestServlet(RestServlet):
                 assigned_user_id=registered_user_id,
             )
 
+        # Only give msisdn flows if the x_show_msisdn flag is given:
+        # this is a hack to work around the fact that clients were shipped
+        # that use fallback registration if they see any flows that they don't
+        # recognise, which means we break registration for these clients if we
+        # advertise msisdn flows. Once usage of Riot iOS <=0.3.9 and Riot
+        # Android <=0.6.9 have fallen below an acceptable threshold, this
+        # parameter should go away and we should always advertise msisdn flows.
+        show_msisdn = False
+        if 'x_show_msisdn' in body and body['x_show_msisdn']:
+            show_msisdn = True
+
         if self.hs.config.enable_registration_captcha:
             flows = [
                 [LoginType.RECAPTCHA],
-                [LoginType.EMAIL_IDENTITY, LoginType.RECAPTCHA]
+                [LoginType.EMAIL_IDENTITY, LoginType.RECAPTCHA],
             ]
+            if show_msisdn:
+                flows.extend([
+                    [LoginType.MSISDN, LoginType.RECAPTCHA],
+                    [LoginType.MSISDN, LoginType.EMAIL_IDENTITY, LoginType.RECAPTCHA],
+                ])
         else:
             flows = [
                 [LoginType.DUMMY],
-                [LoginType.EMAIL_IDENTITY]
+                [LoginType.EMAIL_IDENTITY],
             ]
+            if show_msisdn:
+                flows.extend([
+                    [LoginType.MSISDN],
+                    [LoginType.MSISDN, LoginType.EMAIL_IDENTITY],
+                ])
 
         authed, auth_result, params, session_id = yield self.auth_handler.check_auth(
             flows, body, self.hs.get_ip_from_request(request)
@@ -224,8 +281,9 @@ class RegisterRestServlet(RestServlet):
                 "Already registered user ID %r for this session",
                 registered_user_id
             )
-            # don't re-register the email address
+            # don't re-register the threepids
             add_email = False
+            add_msisdn = False
         else:
             # NB: This may be from the auth handler and NOT from the POST
             if 'password' not in params:
@@ -250,6 +308,7 @@ class RegisterRestServlet(RestServlet):
             )
 
             add_email = True
+            add_msisdn = True
 
         return_dict = yield self._create_registration_details(
             registered_user_id, params
@@ -262,6 +321,13 @@ class RegisterRestServlet(RestServlet):
                 params.get("bind_email")
             )
 
+        if add_msisdn and auth_result and LoginType.MSISDN in auth_result:
+            threepid = auth_result[LoginType.MSISDN]
+            yield self._register_msisdn_threepid(
+                registered_user_id, threepid, return_dict["access_token"],
+                params.get("bind_msisdn")
+            )
+
         defer.returnValue((200, return_dict))
 
     def on_OPTIONS(self, _):
@@ -323,8 +389,9 @@ class RegisterRestServlet(RestServlet):
         """
         reqd = ('medium', 'address', 'validated_at')
         if any(x not in threepid for x in reqd):
+            # This will only happen if the ID server returns a malformed response
             logger.info("Can't add incomplete 3pid")
-            defer.returnValue()
+            return
 
         yield self.auth_handler.add_threepid(
             user_id,
@@ -371,6 +438,43 @@ class RegisterRestServlet(RestServlet):
         else:
             logger.info("bind_email not specified: not binding email")
 
+    @defer.inlineCallbacks
+    def _register_msisdn_threepid(self, user_id, threepid, token, bind_msisdn):
+        """Add a phone number as a 3pid identifier
+
+        Also optionally binds msisdn to the given user_id on the identity server
+
+        Args:
+            user_id (str): id of user
+            threepid (object): m.login.msisdn auth response
+            token (str): access_token for the user
+            bind_email (bool): true if the client requested the email to be
+                bound at the identity server
+        Returns:
+            defer.Deferred:
+        """
+        reqd = ('medium', 'address', 'validated_at')
+        if any(x not in threepid for x in reqd):
+            # This will only happen if the ID server returns a malformed response
+            logger.info("Can't add incomplete 3pid")
+            defer.returnValue()
+
+        yield self.auth_handler.add_threepid(
+            user_id,
+            threepid['medium'],
+            threepid['address'],
+            threepid['validated_at'],
+        )
+
+        if bind_msisdn:
+            logger.info("bind_msisdn specified: binding")
+            logger.debug("Binding msisdn %s to %s", threepid, user_id)
+            yield self.identity_handler.bind_threepid(
+                threepid['threepid_creds'], user_id
+            )
+        else:
+            logger.info("bind_msisdn not specified: not binding msisdn")
+
     @defer.inlineCallbacks
     def _create_registration_details(self, user_id, params):
         """Complete registration of newly-registered user
@@ -433,7 +537,7 @@ class RegisterRestServlet(RestServlet):
         # we have nowhere to store it.
         device_id = synapse.api.auth.GUEST_DEVICE_ID
         initial_display_name = params.get("initial_device_display_name")
-        self.device_handler.check_device_registered(
+        yield self.device_handler.check_device_registered(
             user_id, device_id, initial_display_name
         )
 
@@ -449,5 +553,6 @@ class RegisterRestServlet(RestServlet):
 
 
 def register_servlets(hs, http_server):
-    RegisterRequestTokenRestServlet(hs).register(http_server)
+    EmailRegisterRequestTokenRestServlet(hs).register(http_server)
+    MsisdnRegisterRequestTokenRestServlet(hs).register(http_server)
     RegisterRestServlet(hs).register(http_server)

synapse/rest/client/v2_alpha/sync.py

@@ -18,6 +18,7 @@ from twisted.internet import defer
 from synapse.http.servlet import (
     RestServlet, parse_string, parse_integer, parse_boolean
 )
+from synapse.handlers.presence import format_user_presence_state
 from synapse.handlers.sync import SyncConfig
 from synapse.types import StreamToken
 from synapse.events.utils import (
@@ -28,7 +29,6 @@ from synapse.api.errors import SynapseError
 from synapse.api.constants import PresenceState
 from ._base import client_v2_patterns
 
-import copy
 import itertools
 import logging
 
@@ -194,12 +194,18 @@ class SyncRestServlet(RestServlet):
         defer.returnValue((200, response_content))
 
     def encode_presence(self, events, time_now):
-        formatted = []
-        for event in events:
-            event = copy.deepcopy(event)
-            event['sender'] = event['content'].pop('user_id')
-            formatted.append(event)
-        return {"events": formatted}
+        return {
+            "events": [
+                {
+                    "type": "m.presence",
+                    "sender": event.user_id,
+                    "content": format_user_presence_state(
+                        event, time_now, include_user_id=False
+                    ),
+                }
+                for event in events
+            ]
+        }
 
     def encode_joined(self, rooms, time_now, token_id, event_fields):
         """

synapse/rest/media/v1/download_resource.py

@@ -12,6 +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.http.servlet
 
 from ._base import parse_media_id, respond_with_file, respond_404
 from twisted.web.resource import Resource
@@ -81,6 +82,17 @@ class DownloadResource(Resource):
 
     @defer.inlineCallbacks
     def _respond_remote_file(self, request, server_name, media_id, name):
+        # don't forward requests for remote media if allow_remote is false
+        allow_remote = synapse.http.servlet.parse_boolean(
+            request, "allow_remote", default=True)
+        if not allow_remote:
+            logger.info(
+                "Rejecting request for remote media %s/%s due to allow_remote",
+                server_name, media_id,
+            )
+            respond_404(request)
+            return
+
         media_info = yield self.media_repo.get_remote_media(server_name, media_id)
 
         media_type = media_info["media_type"]

synapse/rest/media/v1/media_repository.py

@@ -13,22 +13,23 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+from twisted.internet import defer, threads
+import twisted.internet.error
+import twisted.web.http
+from twisted.web.resource import Resource
+
 from .upload_resource import UploadResource
 from .download_resource import DownloadResource
 from .thumbnail_resource import ThumbnailResource
 from .identicon_resource import IdenticonResource
 from .preview_url_resource import PreviewUrlResource
 from .filepath import MediaFilePaths
-
-from twisted.web.resource import Resource
-
 from .thumbnailer import Thumbnailer
 
 from synapse.http.matrixfederationclient import MatrixFederationHttpClient
 from synapse.util.stringutils import random_string
-from synapse.api.errors import SynapseError
-
-from twisted.internet import defer, threads
+from synapse.api.errors import SynapseError, HttpResponseException, \
+    NotFoundError
 
 from synapse.util.async import Linearizer
 from synapse.util.stringutils import is_ascii
@@ -157,11 +158,34 @@ class MediaRepository(object):
                 try:
                     length, headers = yield self.client.get_file(
                         server_name, request_path, output_stream=f,
-                        max_size=self.max_upload_size,
+                        max_size=self.max_upload_size, args={
+                            # tell the remote server to 404 if it doesn't
+                            # recognise the server_name, to make sure we don't
+                            # end up with a routing loop.
+                            "allow_remote": "false",
+                        }
                     )
-                except Exception as e:
-                    logger.warn("Failed to fetch remoted media %r", e)
-                    raise SynapseError(502, "Failed to fetch remoted media")
+                except twisted.internet.error.DNSLookupError as e:
+                    logger.warn("HTTP error fetching remote media %s/%s: %r",
+                                server_name, media_id, e)
+                    raise NotFoundError()
+
+                except HttpResponseException as e:
+                    logger.warn("HTTP error fetching remote media %s/%s: %s",
+                                server_name, media_id, e.response)
+                    if e.code == twisted.web.http.NOT_FOUND:
+                        raise SynapseError.from_http_response_exception(e)
+                    raise SynapseError(502, "Failed to fetch remote media")
+
+                except SynapseError:
+                    logger.exception("Failed to fetch remote media %s/%s",
+                                     server_name, media_id)
+                    raise
+
+                except Exception:
+                    logger.exception("Failed to fetch remote media %s/%s",
+                                     server_name, media_id)
+                    raise SynapseError(502, "Failed to fetch remote media")
 
             media_type = headers["Content-Type"][0]
             time_now_ms = self.clock.time_msec()

synapse/state.py

@@ -177,17 +177,12 @@ class StateHandler(object):
 
     @defer.inlineCallbacks
     def compute_event_context(self, event, old_state=None):
-        """ Fills out the context with the `current state` of the graph. The
-        `current state` here is defined to be the state of the event graph
-        just before the event - i.e. it never includes `event`
-
-        If `event` has `auth_events` then this will also fill out the
-        `auth_events` field on `context` from the `current_state`.
+        """Build an EventContext structure for the event.
 
         Args:
-            event (EventBase)
+            event (synapse.events.EventBase):
         Returns:
-            an EventContext
+            synapse.events.snapshot.EventContext:
         """
         context = EventContext()
 
@@ -200,11 +195,11 @@ class StateHandler(object):
                     (s.type, s.state_key): s.event_id for s in old_state
                 }
                 if event.is_state():
-                    context.current_state_events = dict(context.prev_state_ids)
+                    context.current_state_ids = dict(context.prev_state_ids)
                     key = (event.type, event.state_key)
-                    context.current_state_events[key] = event.event_id
+                    context.current_state_ids[key] = event.event_id
                 else:
-                    context.current_state_events = context.prev_state_ids
+                    context.current_state_ids = context.prev_state_ids
             else:
                 context.current_state_ids = {}
                 context.prev_state_ids = {}

synapse/storage/_base.py

@@ -18,7 +18,6 @@ from synapse.api.errors import StoreError
 from synapse.util.logcontext import LoggingContext, PreserveLoggingContext
 from synapse.util.caches.dictionary_cache import DictionaryCache
 from synapse.util.caches.descriptors import Cache
-from synapse.util.caches import intern_dict
 from synapse.storage.engines import PostgresEngine
 import synapse.metrics
 
@@ -74,6 +73,9 @@ class LoggingTransaction(object):
     def __setattr__(self, name, value):
         setattr(self.txn, name, value)
 
+    def __iter__(self):
+        return self.txn.__iter__()
+
     def execute(self, sql, *args):
         self._do_execute(self.txn.execute, sql, *args)
 
@@ -133,7 +135,7 @@ class PerformanceCounters(object):
 
     def interval(self, interval_duration, limit=3):
         counters = []
-        for name, (count, cum_time) in self.current_counters.items():
+        for name, (count, cum_time) in self.current_counters.iteritems():
             prev_count, prev_time = self.previous_counters.get(name, (0, 0))
             counters.append((
                 (cum_time - prev_time) / interval_duration,
@@ -356,9 +358,9 @@ class SQLBaseStore(object):
         Returns:
             A list of dicts where the key is the column header.
         """
-        col_headers = list(column[0] for column in cursor.description)
+        col_headers = list(intern(column[0]) for column in cursor.description)
         results = list(
-            intern_dict(dict(zip(col_headers, row))) for row in cursor.fetchall()
+            dict(zip(col_headers, row)) for row in cursor
         )
         return results
 
@@ -489,10 +491,6 @@ class SQLBaseStore(object):
             " AND ".join("%s = ?" % (k,) for k in keyvalues)
         )
         sqlargs = values.values() + keyvalues.values()
-        logger.debug(
-            "[SQL] %s Args=%s",
-            sql, sqlargs,
-        )
 
         txn.execute(sql, sqlargs)
         if txn.rowcount == 0:
@@ -507,10 +505,6 @@ class SQLBaseStore(object):
                 ", ".join(k for k in allvalues),
                 ", ".join("?" for _ in allvalues)
             )
-            logger.debug(
-                "[SQL] %s Args=%s",
-                sql, keyvalues.values(),
-            )
             txn.execute(sql, allvalues.values())
 
             return True
@@ -574,7 +568,7 @@ class SQLBaseStore(object):
     @staticmethod
     def _simple_select_onecol_txn(txn, table, keyvalues, retcol):
         if keyvalues:
-            where = "WHERE %s" % " AND ".join("%s = ?" % k for k in keyvalues.keys())
+            where = "WHERE %s" % " AND ".join("%s = ?" % k for k in keyvalues.iterkeys())
         else:
             where = ""
 
@@ -588,7 +582,7 @@ class SQLBaseStore(object):
 
         txn.execute(sql, keyvalues.values())
 
-        return [r[0] for r in txn.fetchall()]
+        return [r[0] for r in txn]
 
     def _simple_select_onecol(self, table, keyvalues, retcol,
                               desc="_simple_select_onecol"):
@@ -721,7 +715,7 @@ class SQLBaseStore(object):
         )
         values.extend(iterable)
 
-        for key, value in keyvalues.items():
+        for key, value in keyvalues.iteritems():
             clauses.append("%s = ?" % (key,))
             values.append(value)
 
@@ -762,7 +756,7 @@ class SQLBaseStore(object):
     @staticmethod
     def _simple_update_one_txn(txn, table, keyvalues, updatevalues):
         if keyvalues:
-            where = "WHERE %s" % " AND ".join("%s = ?" % k for k in keyvalues.keys())
+            where = "WHERE %s" % " AND ".join("%s = ?" % k for k in keyvalues.iterkeys())
         else:
             where = ""
 
@@ -849,6 +843,47 @@ class SQLBaseStore(object):
 
         return txn.execute(sql, keyvalues.values())
 
+    def _simple_delete_many(self, table, column, iterable, keyvalues, desc):
+        return self.runInteraction(
+            desc, self._simple_delete_many_txn, table, column, iterable, keyvalues
+        )
+
+    @staticmethod
+    def _simple_delete_many_txn(txn, table, column, iterable, keyvalues):
+        """Executes a DELETE query on the named table.
+
+        Filters rows by if value of `column` is in `iterable`.
+
+        Args:
+            txn : Transaction object
+            table : string giving the table name
+            column : column name to test for inclusion against `iterable`
+            iterable : list
+            keyvalues : dict of column names and values to select the rows with
+        """
+        if not iterable:
+            return
+
+        sql = "DELETE FROM %s" % table
+
+        clauses = []
+        values = []
+        clauses.append(
+            "%s IN (%s)" % (column, ",".join("?" for _ in iterable))
+        )
+        values.extend(iterable)
+
+        for key, value in keyvalues.iteritems():
+            clauses.append("%s = ?" % (key,))
+            values.append(value)
+
+        if clauses:
+            sql = "%s WHERE %s" % (
+                sql,
+                " AND ".join(clauses),
+            )
+        return txn.execute(sql, values)
+
     def _get_cache_dict(self, db_conn, table, entity_column, stream_column,
                         max_value, limit=100000):
         # Fetch a mapping of room_id -> max stream position for "recent" rooms.
@@ -869,16 +904,16 @@ class SQLBaseStore(object):
 
         txn = db_conn.cursor()
         txn.execute(sql, (int(max_value),))
-        rows = txn.fetchall()
-        txn.close()
 
         cache = {
             row[0]: int(row[1])
-            for row in rows
+            for row in txn
         }
 
+        txn.close()
+
         if cache:
-            min_val = min(cache.values())
+            min_val = min(cache.itervalues())
         else:
             min_val = max_value
 

synapse/storage/account_data.py

@@ -182,7 +182,7 @@ class AccountDataStore(SQLBaseStore):
             txn.execute(sql, (user_id, stream_id))
 
             global_account_data = {
-                row[0]: json.loads(row[1]) for row in txn.fetchall()
+                row[0]: json.loads(row[1]) for row in txn
             }
 
             sql = (
@@ -193,7 +193,7 @@ class AccountDataStore(SQLBaseStore):
             txn.execute(sql, (user_id, stream_id))
 
             account_data_by_room = {}
-            for row in txn.fetchall():
+            for row in txn:
                 room_account_data = account_data_by_room.setdefault(row[0], {})
                 room_account_data[row[1]] = json.loads(row[2])
 

synapse/storage/background_updates.py

@@ -12,6 +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.async
 
 from ._base import SQLBaseStore
 from . import engines
@@ -84,24 +85,14 @@ class BackgroundUpdateStore(SQLBaseStore):
         self._background_update_performance = {}
         self._background_update_queue = []
         self._background_update_handlers = {}
-        self._background_update_timer = None
 
     @defer.inlineCallbacks
     def start_doing_background_updates(self):
-        assert self._background_update_timer is None, \
-            "background updates already running"
-
         logger.info("Starting background schema updates")
 
         while True:
-            sleep = defer.Deferred()
-            self._background_update_timer = self._clock.call_later(
-                self.BACKGROUND_UPDATE_INTERVAL_MS / 1000., sleep.callback, None
-            )
-            try:
-                yield sleep
-            finally:
-                self._background_update_timer = None
+            yield synapse.util.async.sleep(
+                self.BACKGROUND_UPDATE_INTERVAL_MS / 1000.)
 
             try:
                 result = yield self.do_next_background_update(

synapse/storage/deviceinbox.py

@@ -20,6 +20,8 @@ from twisted.internet import defer
 
 from .background_updates import BackgroundUpdateStore
 
+from synapse.util.caches.expiringcache import ExpiringCache
+
 
 logger = logging.getLogger(__name__)
 
@@ -42,6 +44,15 @@ class DeviceInboxStore(BackgroundUpdateStore):
             self._background_drop_index_device_inbox,
         )
 
+        # Map of (user_id, device_id) to the last stream_id that has been
+        # deleted up to. This is so that we can no op deletions.
+        self._last_device_delete_cache = ExpiringCache(
+            cache_name="last_device_delete_cache",
+            clock=self._clock,
+            max_len=10000,
+            expiry_ms=30 * 60 * 1000,
+        )
+
     @defer.inlineCallbacks
     def add_messages_to_device_inbox(self, local_messages_by_user_then_device,
                                      remote_messages_by_destination):
@@ -167,7 +178,7 @@ class DeviceInboxStore(BackgroundUpdateStore):
                 )
                 txn.execute(sql, (user_id,))
                 message_json = ujson.dumps(messages_by_device["*"])
-                for row in txn.fetchall():
+                for row in txn:
                     # Add the message for all devices for this user on this
                     # server.
                     device = row[0]
@@ -184,7 +195,7 @@ class DeviceInboxStore(BackgroundUpdateStore):
                 # TODO: Maybe this needs to be done in batches if there are
                 # too many local devices for a given user.
                 txn.execute(sql, [user_id] + devices)
-                for row in txn.fetchall():
+                for row in txn:
                     # Only insert into the local inbox if the device exists on
                     # this server
                     device = row[0]
@@ -240,7 +251,7 @@ class DeviceInboxStore(BackgroundUpdateStore):
                 user_id, device_id, last_stream_id, current_stream_id, limit
             ))
             messages = []
-            for row in txn.fetchall():
+            for row in txn:
                 stream_pos = row[0]
                 messages.append(ujson.loads(row[1]))
             if len(messages) < limit:
@@ -251,6 +262,7 @@ class DeviceInboxStore(BackgroundUpdateStore):
             "get_new_messages_for_device", get_new_messages_for_device_txn,
         )
 
+    @defer.inlineCallbacks
     def delete_messages_for_device(self, user_id, device_id, up_to_stream_id):
         """
         Args:
@@ -260,6 +272,18 @@ class DeviceInboxStore(BackgroundUpdateStore):
         Returns:
             A deferred that resolves to the number of messages deleted.
         """
+        # If we have cached the last stream id we've deleted up to, we can
+        # check if there is likely to be anything that needs deleting
+        last_deleted_stream_id = self._last_device_delete_cache.get(
+            (user_id, device_id), None
+        )
+        if last_deleted_stream_id:
+            has_changed = self._device_inbox_stream_cache.has_entity_changed(
+                user_id, last_deleted_stream_id
+            )
+            if not has_changed:
+                defer.returnValue(0)
+
         def delete_messages_for_device_txn(txn):
             sql = (
                 "DELETE FROM device_inbox"
@@ -269,10 +293,20 @@ class DeviceInboxStore(BackgroundUpdateStore):
             txn.execute(sql, (user_id, device_id, up_to_stream_id))
             return txn.rowcount
 
-        return self.runInteraction(
+        count = yield self.runInteraction(
             "delete_messages_for_device", delete_messages_for_device_txn
         )
 
+        # Update the cache, ensuring that we only ever increase the value
+        last_deleted_stream_id = self._last_device_delete_cache.get(
+            (user_id, device_id), 0
+        )
+        self._last_device_delete_cache[(user_id, device_id)] = max(
+            last_deleted_stream_id, up_to_stream_id
+        )
+
+        defer.returnValue(count)
+
     def get_all_new_device_messages(self, last_pos, current_pos, limit):
         """
         Args:
@@ -306,7 +340,7 @@ class DeviceInboxStore(BackgroundUpdateStore):
                 " ORDER BY stream_id ASC"
             )
             txn.execute(sql, (last_pos, upper_pos))
-            rows.extend(txn.fetchall())
+            rows.extend(txn)
 
             return rows
 
@@ -323,12 +357,12 @@ class DeviceInboxStore(BackgroundUpdateStore):
         """
         Args:
             destination(str): The name of the remote server.
-            last_stream_id(int): The last position of the device message stream
+            last_stream_id(int|long): The last position of the device message stream
                 that the server sent up to.
-            current_stream_id(int): The current position of the device
+            current_stream_id(int|long): The current position of the device
                 message stream.
         Returns:
-            Deferred ([dict], int): List of messages for the device and where
+            Deferred ([dict], int|long): List of messages for the device and where
                 in the stream the messages got to.
         """
 
@@ -350,7 +384,7 @@ class DeviceInboxStore(BackgroundUpdateStore):
                 destination, last_stream_id, current_stream_id, limit
             ))
             messages = []
-            for row in txn.fetchall():
+            for row in txn:
                 stream_pos = row[0]
                 messages.append(ujson.loads(row[1]))
             if len(messages) < limit:

synapse/storage/devices.py

@@ -33,6 +33,13 @@ class DeviceStore(SQLBaseStore):
             self._prune_old_outbound_device_pokes, 60 * 60 * 1000
         )
 
+        self.register_background_index_update(
+            "device_lists_stream_idx",
+            index_name="device_lists_stream_user_id",
+            table="device_lists_stream",
+            columns=["user_id", "device_id"],
+        )
+
     @defer.inlineCallbacks
     def store_device(self, user_id, device_id,
                      initial_device_display_name):
@@ -101,6 +108,23 @@ class DeviceStore(SQLBaseStore):
             desc="delete_device",
         )
 
+    def delete_devices(self, user_id, device_ids):
+        """Deletes several devices.
+
+        Args:
+            user_id (str): The ID of the user which owns the devices
+            device_ids (list): The IDs of the devices to delete
+        Returns:
+            defer.Deferred
+        """
+        return self._simple_delete_many(
+            table="devices",
+            column="device_id",
+            iterable=device_ids,
+            keyvalues={"user_id": user_id},
+            desc="delete_devices",
+        )
+
     def update_device(self, user_id, device_id, new_display_name=None):
         """Update a device.
 
@@ -284,7 +308,7 @@ class DeviceStore(SQLBaseStore):
         """Get stream of updates to send to remote servers
 
         Returns:
-            (now_stream_id, [ { updates }, .. ])
+            (int, list[dict]): current stream id and list of updates
         """
         now_stream_id = self._device_list_id_gen.get_current_token()
 
@@ -305,17 +329,20 @@ class DeviceStore(SQLBaseStore):
             SELECT user_id, device_id, max(stream_id) FROM device_lists_outbound_pokes
             WHERE destination = ? AND ? < stream_id AND stream_id <= ? AND sent = ?
             GROUP BY user_id, device_id
+            LIMIT 20
         """
         txn.execute(
             sql, (destination, from_stream_id, now_stream_id, False)
         )
-        rows = txn.fetchall()
-
-        if not rows:
-            return (now_stream_id, [])
 
         # maps (user_id, device_id) -> stream_id
-        query_map = {(r[0], r[1]): r[2] for r in rows}
+        query_map = {(r[0], r[1]): r[2] for r in txn}
+        if not query_map:
+            return (now_stream_id, [])
+
+        if len(query_map) >= 20:
+            now_stream_id = max(stream_id for stream_id in query_map.itervalues())
+
         devices = self._get_e2e_device_keys_txn(
             txn, query_map.keys(), include_all_devices=True
         )
@@ -501,12 +528,12 @@ class DeviceStore(SQLBaseStore):
             defer.returnValue(set(changed))
 
         sql = """
-            SELECT user_id FROM device_lists_stream WHERE stream_id > ?
+            SELECT DISTINCT user_id FROM device_lists_stream WHERE stream_id > ?
         """
         rows = yield self._execute("get_user_whose_devices_changed", None, sql, from_key)
         defer.returnValue(set(row[0] for row in rows))
 
-    def get_all_device_list_changes_for_remotes(self, from_key):
+    def get_all_device_list_changes_for_remotes(self, from_key, to_key):
         """Return a list of `(stream_id, user_id, destination)` which is the
         combined list of changes to devices, and which destinations need to be
         poked. `destination` may be None if no destinations need to be poked.
@@ -514,11 +541,11 @@ class DeviceStore(SQLBaseStore):
         sql = """
             SELECT stream_id, user_id, destination FROM device_lists_stream
             LEFT JOIN device_lists_outbound_pokes USING (stream_id, user_id, device_id)
-            WHERE stream_id > ?
+            WHERE ? < stream_id AND stream_id <= ?
         """
         return self._execute(
-            "get_users_and_hosts_device_list", None,
-            sql, from_key,
+            "get_all_device_list_changes_for_remotes", None,
+            sql, from_key, to_key
         )
 
     @defer.inlineCallbacks
@@ -546,6 +573,16 @@ class DeviceStore(SQLBaseStore):
                 host, stream_id,
             )
 
+        # Delete older entries in the table, as we really only care about
+        # when the latest change happened.
+        txn.executemany(
+            """
+            DELETE FROM device_lists_stream
+            WHERE user_id = ? AND device_id = ? AND stream_id < ?
+            """,
+            [(user_id, device_id, stream_id) for device_id in device_ids]
+        )
+
         self._simple_insert_many_txn(
             txn,
             table="device_lists_stream",

synapse/storage/end_to_end_keys.py

@@ -14,6 +14,8 @@
 # limitations under the License.
 from twisted.internet import defer
 
+from synapse.api.errors import SynapseError
+
 from canonicaljson import encode_canonical_json
 import ujson as json
 
@@ -120,24 +122,63 @@ class EndToEndKeyStore(SQLBaseStore):
 
         return result
 
+    @defer.inlineCallbacks
     def add_e2e_one_time_keys(self, user_id, device_id, time_now, key_list):
+        """Insert some new one time keys for a device.
+
+        Checks if any of the keys are already inserted, if they are then check
+        if they match. If they don't then we raise an error.
+        """
+
+        # First we check if we have already persisted any of the keys.
+        rows = yield self._simple_select_many_batch(
+            table="e2e_one_time_keys_json",
+            column="key_id",
+            iterable=[key_id for _, key_id, _ in key_list],
+            retcols=("algorithm", "key_id", "key_json",),
+            keyvalues={
+                "user_id": user_id,
+                "device_id": device_id,
+            },
+            desc="add_e2e_one_time_keys_check",
+        )
+
+        existing_key_map = {
+            (row["algorithm"], row["key_id"]): row["key_json"] for row in rows
+        }
+
+        new_keys = []  # Keys that we need to insert
+        for algorithm, key_id, json_bytes in key_list:
+            ex_bytes = existing_key_map.get((algorithm, key_id), None)
+            if ex_bytes:
+                if json_bytes != ex_bytes:
+                    raise SynapseError(
+                        400, "One time key with key_id %r already exists" % (key_id,)
+                    )
+            else:
+                new_keys.append((algorithm, key_id, json_bytes))
+
         def _add_e2e_one_time_keys(txn):
-            for (algorithm, key_id, json_bytes) in key_list:
-                self._simple_upsert_txn(
-                    txn, table="e2e_one_time_keys_json",
-                    keyvalues={
+            # We are protected from race between lookup and insertion due to
+            # a unique constraint. If there is a race of two calls to
+            # `add_e2e_one_time_keys` then they'll conflict and we will only
+            # insert one set.
+            self._simple_insert_many_txn(
+                txn, table="e2e_one_time_keys_json",
+                values=[
+                    {
                         "user_id": user_id,
                         "device_id": device_id,
                         "algorithm": algorithm,
                         "key_id": key_id,
-                    },
-                    values={
                         "ts_added_ms": time_now,
                         "key_json": json_bytes,
                     }
-                )
-        return self.runInteraction(
-            "add_e2e_one_time_keys", _add_e2e_one_time_keys
+                    for algorithm, key_id, json_bytes in new_keys
+                ],
+            )
+        yield self.runInteraction(
+            "add_e2e_one_time_keys_insert", _add_e2e_one_time_keys
         )
 
     def count_e2e_one_time_keys(self, user_id, device_id):
@@ -153,7 +194,7 @@ class EndToEndKeyStore(SQLBaseStore):
             )
             txn.execute(sql, (user_id, device_id))
             result = {}
-            for algorithm, key_count in txn.fetchall():
+            for algorithm, key_count in txn:
                 result[algorithm] = key_count
             return result
         return self.runInteraction(
@@ -174,7 +215,7 @@ class EndToEndKeyStore(SQLBaseStore):
                 user_result = result.setdefault(user_id, {})
                 device_result = user_result.setdefault(device_id, {})
                 txn.execute(sql, (user_id, device_id, algorithm))
-                for key_id, key_json in txn.fetchall():
+                for key_id, key_json in txn:
                     device_result[algorithm + ":" + key_id] = key_json
                     delete.append((user_id, device_id, algorithm, key_id))
             sql = (

synapse/storage/event_federation.py

@@ -74,7 +74,7 @@ class EventFederationStore(SQLBaseStore):
                     base_sql % (",".join(["?"] * len(chunk)),),
                     chunk
                 )
-                new_front.update([r[0] for r in txn.fetchall()])
+                new_front.update([r[0] for r in txn])
 
             new_front -= results
 
@@ -110,7 +110,7 @@ class EventFederationStore(SQLBaseStore):
 
         txn.execute(sql, (room_id, False,))
 
-        return dict(txn.fetchall())
+        return dict(txn)
 
     def _get_oldest_events_in_room_txn(self, txn, room_id):
         return self._simple_select_onecol_txn(
@@ -201,19 +201,19 @@ class EventFederationStore(SQLBaseStore):
     def _update_min_depth_for_room_txn(self, txn, room_id, depth):
         min_depth = self._get_min_depth_interaction(txn, room_id)
 
-        do_insert = depth < min_depth if min_depth else True
+        if min_depth and depth >= min_depth:
+            return
 
-        if do_insert:
-            self._simple_upsert_txn(
-                txn,
-                table="room_depth",
-                keyvalues={
-                    "room_id": room_id,
-                },
-                values={
-                    "min_depth": depth,
-                },
-            )
+        self._simple_upsert_txn(
+            txn,
+            table="room_depth",
+            keyvalues={
+                "room_id": room_id,
+            },
+            values={
+                "min_depth": depth,
+            },
+        )
 
     def _handle_mult_prev_events(self, txn, events):
         """
@@ -334,8 +334,7 @@ class EventFederationStore(SQLBaseStore):
 
         def get_forward_extremeties_for_room_txn(txn):
             txn.execute(sql, (stream_ordering, room_id))
-            rows = txn.fetchall()
-            return [event_id for event_id, in rows]
+            return [event_id for event_id, in txn]
 
         return self.runInteraction(
             "get_forward_extremeties_for_room",
@@ -436,7 +435,7 @@ class EventFederationStore(SQLBaseStore):
                 (room_id, event_id, False, limit - len(event_results))
             )
 
-            for row in txn.fetchall():
+            for row in txn:
                 if row[1] not in event_results:
                     queue.put((-row[0], row[1]))
 
@@ -482,7 +481,7 @@ class EventFederationStore(SQLBaseStore):
                     (room_id, event_id, False, limit - len(event_results))
                 )
 
-                for e_id, in txn.fetchall():
+                for e_id, in txn:
                     new_front.add(e_id)
 
             new_front -= earliest_events

synapse/storage/event_push_actions.py

@@ -208,7 +208,7 @@ class EventPushActionsStore(SQLBaseStore):
                 " stream_ordering >= ? AND stream_ordering <= ?"
             )
             txn.execute(sql, (min_stream_ordering, max_stream_ordering))
-            return [r[0] for r in txn.fetchall()]
+            return [r[0] for r in txn]
         ret = yield self.runInteraction("get_push_action_users_in_range", f)
         defer.returnValue(ret)
 

synapse/storage/events.py

@@ -34,14 +34,16 @@ from canonicaljson import encode_canonical_json
 from collections import deque, namedtuple, OrderedDict
 from functools import wraps
 
-import synapse
 import synapse.metrics
 
-
 import logging
 import math
 import ujson as json
 
+# these are only included to make the type annotations work
+from synapse.events import EventBase    # noqa: F401
+from synapse.events.snapshot import EventContext   # noqa: F401
+
 logger = logging.getLogger(__name__)
 
 
@@ -82,6 +84,11 @@ class _EventPeristenceQueue(object):
 
     def add_to_queue(self, room_id, events_and_contexts, backfilled):
         """Add events to the queue, with the given persist_event options.
+
+        Args:
+            room_id (str):
+            events_and_contexts (list[(EventBase, EventContext)]):
+            backfilled (bool):
         """
         queue = self._event_persist_queues.setdefault(room_id, deque())
         if queue:
@@ -210,14 +217,14 @@ class EventsStore(SQLBaseStore):
             partitioned.setdefault(event.room_id, []).append((event, ctx))
 
         deferreds = []
-        for room_id, evs_ctxs in partitioned.items():
+        for room_id, evs_ctxs in partitioned.iteritems():
             d = preserve_fn(self._event_persist_queue.add_to_queue)(
                 room_id, evs_ctxs,
                 backfilled=backfilled,
             )
             deferreds.append(d)
 
-        for room_id in partitioned.keys():
+        for room_id in partitioned:
             self._maybe_start_persisting(room_id)
 
         return preserve_context_over_deferred(
@@ -227,6 +234,17 @@ class EventsStore(SQLBaseStore):
     @defer.inlineCallbacks
     @log_function
     def persist_event(self, event, context, backfilled=False):
+        """
+
+        Args:
+            event (EventBase):
+            context (EventContext):
+            backfilled (bool):
+
+        Returns:
+            Deferred: resolves to (int, int): the stream ordering of ``event``,
+            and the stream ordering of the latest persisted event
+        """
         deferred = self._event_persist_queue.add_to_queue(
             event.room_id, [(event, context)],
             backfilled=backfilled,
@@ -253,6 +271,16 @@ class EventsStore(SQLBaseStore):
     @defer.inlineCallbacks
     def _persist_events(self, events_and_contexts, backfilled=False,
                         delete_existing=False):
+        """Persist events to db
+
+        Args:
+            events_and_contexts (list[(EventBase, EventContext)]):
+            backfilled (bool):
+            delete_existing (bool):
+
+        Returns:
+            Deferred: resolves when the events have been persisted
+        """
         if not events_and_contexts:
             return
 
@@ -295,7 +323,7 @@ class EventsStore(SQLBaseStore):
                                 (event, context)
                             )
 
-                        for room_id, ev_ctx_rm in events_by_room.items():
+                        for room_id, ev_ctx_rm in events_by_room.iteritems():
                             # Work out new extremities by recursively adding and removing
                             # the new events.
                             latest_event_ids = yield self.get_latest_event_ids_in_room(
@@ -311,6 +339,21 @@ class EventsStore(SQLBaseStore):
 
                             new_forward_extremeties[room_id] = new_latest_event_ids
 
+                            len_1 = (
+                                len(latest_event_ids) == 1
+                                and len(new_latest_event_ids) == 1
+                            )
+                            if len_1:
+                                all_single_prev_not_state = all(
+                                    len(event.prev_events) == 1
+                                    and not event.is_state()
+                                    for event, ctx in ev_ctx_rm
+                                )
+                                # Don't bother calculating state if they're just
+                                # a long chain of single ancestor non-state events.
+                                if all_single_prev_not_state:
+                                    continue
+
                             state = yield self._calculate_state_delta(
                                 room_id, ev_ctx_rm, new_latest_event_ids
                             )
@@ -385,6 +428,7 @@ class EventsStore(SQLBaseStore):
         # Now we need to work out the different state sets for
         # each state extremities
         state_sets = []
+        state_groups = set()
         missing_event_ids = []
         was_updated = False
         for event_id in new_latest_event_ids:
@@ -394,9 +438,17 @@ class EventsStore(SQLBaseStore):
                 if event_id == ev.event_id:
                     if ctx.current_state_ids is None:
                         raise Exception("Unknown current state")
-                    state_sets.append(ctx.current_state_ids)
-                    if ctx.delta_ids or hasattr(ev, "state_key"):
-                        was_updated = True
+
+                    # If we've already seen the state group don't bother adding
+                    # it to the state sets again
+                    if ctx.state_group not in state_groups:
+                        state_sets.append(ctx.current_state_ids)
+                        if ctx.delta_ids or hasattr(ev, "state_key"):
+                            was_updated = True
+                        if ctx.state_group:
+                            # Add this as a seen state group (if it has a state
+                            # group)
+                            state_groups.add(ctx.state_group)
                     break
             else:
                 # If we couldn't find it, then we'll need to pull
@@ -410,31 +462,57 @@ class EventsStore(SQLBaseStore):
                 missing_event_ids,
             )
 
-            groups = set(event_to_groups.values())
-            group_to_state = yield self._get_state_for_groups(groups)
+            groups = set(event_to_groups.itervalues()) - state_groups
 
-            state_sets.extend(group_to_state.values())
+            if groups:
+                group_to_state = yield self._get_state_for_groups(groups)
+                state_sets.extend(group_to_state.itervalues())
 
         if not new_latest_event_ids:
             current_state = {}
         elif was_updated:
-            current_state = yield resolve_events(
-                state_sets,
-                state_map_factory=lambda ev_ids: self.get_events(
-                    ev_ids, get_prev_content=False, check_redacted=False,
-                ),
-            )
+            if len(state_sets) == 1:
+                # If there is only one state set, then we know what the current
+                # state is.
+                current_state = state_sets[0]
+            else:
+                # We work out the current state by passing the state sets to the
+                # state resolution algorithm. It may ask for some events, including
+                # the events we have yet to persist, so we need a slightly more
+                # complicated event lookup function than simply looking the events
+                # up in the db.
+                events_map = {ev.event_id: ev for ev, _ in events_context}
+
+                @defer.inlineCallbacks
+                def get_events(ev_ids):
+                    # We get the events by first looking at the list of events we
+                    # are trying to persist, and then fetching the rest from the DB.
+                    db = []
+                    to_return = {}
+                    for ev_id in ev_ids:
+                        ev = events_map.get(ev_id, None)
+                        if ev:
+                            to_return[ev_id] = ev
+                        else:
+                            db.append(ev_id)
+
+                    if db:
+                        evs = yield self.get_events(
+                            ev_ids, get_prev_content=False, check_redacted=False,
+                        )
+                        to_return.update(evs)
+                    defer.returnValue(to_return)
+
+                current_state = yield resolve_events(
+                    state_sets,
+                    state_map_factory=get_events,
+                )
         else:
             return
 
-        existing_state_rows = yield self._simple_select_list(
-            table="current_state_events",
-            keyvalues={"room_id": room_id},
-            retcols=["event_id", "type", "state_key"],
-            desc="_calculate_state_delta",
-        )
+        existing_state = yield self.get_current_state_ids(room_id)
 
-        existing_events = set(row["event_id"] for row in existing_state_rows)
+        existing_events = set(existing_state.itervalues())
         new_events = set(ev_id for ev_id in current_state.itervalues())
         changed_events = existing_events ^ new_events
 
@@ -442,9 +520,8 @@ class EventsStore(SQLBaseStore):
             return
 
         to_delete = {
-            (row["type"], row["state_key"]): row["event_id"]
-            for row in existing_state_rows
-            if row["event_id"] in changed_events
+            key: ev_id for key, ev_id in existing_state.iteritems()
+            if ev_id in changed_events
         }
         events_to_insert = (new_events - existing_events)
         to_insert = {
@@ -520,11 +597,91 @@ class EventsStore(SQLBaseStore):
         and the rejections table. Things reading from those table will need to check
         whether the event was rejected.
 
-        If delete_existing is True then existing events will be purged from the
-        database before insertion. This is useful when retrying due to IntegrityError.
+        Args:
+            txn (twisted.enterprise.adbapi.Connection): db connection
+            events_and_contexts (list[(EventBase, EventContext)]):
+                events to persist
+            backfilled (bool): True if the events were backfilled
+            delete_existing (bool): True to purge existing table rows for the
+                events from the database. This is useful when retrying due to
+                IntegrityError.
+            current_state_for_room (dict[str, (list[str], list[str])]):
+                The current-state delta for each room. For each room, a tuple
+                (to_delete, to_insert), being a list of event ids to be removed
+                from the current state, and a list of event ids to be added to
+                the current state.
+            new_forward_extremeties (dict[str, list[str]]):
+                The new forward extremities for each room. For each room, a
+                list of the event ids which are the forward extremities.
+
         """
+        self._update_current_state_txn(txn, current_state_for_room)
+
         max_stream_order = events_and_contexts[-1][0].internal_metadata.stream_ordering
-        for room_id, current_state_tuple in current_state_for_room.iteritems():
+        self._update_forward_extremities_txn(
+            txn,
+            new_forward_extremities=new_forward_extremeties,
+            max_stream_order=max_stream_order,
+        )
+
+        # Ensure that we don't have the same event twice.
+        events_and_contexts = self._filter_events_and_contexts_for_duplicates(
+            events_and_contexts,
+        )
+
+        self._update_room_depths_txn(
+            txn,
+            events_and_contexts=events_and_contexts,
+            backfilled=backfilled,
+        )
+
+        # _update_outliers_txn filters out any events which have already been
+        # persisted, and returns the filtered list.
+        events_and_contexts = self._update_outliers_txn(
+            txn,
+            events_and_contexts=events_and_contexts,
+        )
+
+        # From this point onwards the events are only events that we haven't
+        # seen before.
+
+        if delete_existing:
+            # For paranoia reasons, we go and delete all the existing entries
+            # for these events so we can reinsert them.
+            # This gets around any problems with some tables already having
+            # entries.
+            self._delete_existing_rows_txn(
+                txn,
+                events_and_contexts=events_and_contexts,
+            )
+
+        self._store_event_txn(
+            txn,
+            events_and_contexts=events_and_contexts,
+        )
+
+        # Insert into the state_groups, state_groups_state, and
+        # event_to_state_groups tables.
+        self._store_mult_state_groups_txn(txn, events_and_contexts)
+
+        # _store_rejected_events_txn filters out any events which were
+        # rejected, and returns the filtered list.
+        events_and_contexts = self._store_rejected_events_txn(
+            txn,
+            events_and_contexts=events_and_contexts,
+        )
+
+        # From this point onwards the events are only ones that weren't
+        # rejected.
+
+        self._update_metadata_tables_txn(
+            txn,
+            events_and_contexts=events_and_contexts,
+            backfilled=backfilled,
+        )
+
+    def _update_current_state_txn(self, txn, state_delta_by_room):
+        for room_id, current_state_tuple in state_delta_by_room.iteritems():
                 to_delete, to_insert = current_state_tuple
                 txn.executemany(
                     "DELETE FROM current_state_events WHERE event_id = ?",
@@ -570,7 +727,13 @@ class EventsStore(SQLBaseStore):
                     txn, self.get_users_in_room, (room_id,)
                 )
 
-        for room_id, new_extrem in new_forward_extremeties.items():
+                self._invalidate_cache_and_stream(
+                    txn, self.get_current_state_ids, (room_id,)
+                )
+
+    def _update_forward_extremities_txn(self, txn, new_forward_extremities,
+                                        max_stream_order):
+        for room_id, new_extrem in new_forward_extremities.iteritems():
             self._simple_delete_txn(
                 txn,
                 table="event_forward_extremities",
@@ -588,7 +751,7 @@ class EventsStore(SQLBaseStore):
                     "event_id": ev_id,
                     "room_id": room_id,
                 }
-                for room_id, new_extrem in new_forward_extremeties.items()
+                for room_id, new_extrem in new_forward_extremities.iteritems()
                 for ev_id in new_extrem
             ],
         )
@@ -605,13 +768,22 @@ class EventsStore(SQLBaseStore):
                     "event_id": event_id,
                     "stream_ordering": max_stream_order,
                 }
-                for room_id, new_extrem in new_forward_extremeties.items()
+                for room_id, new_extrem in new_forward_extremities.iteritems()
                 for event_id in new_extrem
             ]
         )
 
-        # Ensure that we don't have the same event twice.
-        # Pick the earliest non-outlier if there is one, else the earliest one.
+    @classmethod
+    def _filter_events_and_contexts_for_duplicates(cls, events_and_contexts):
+        """Ensure that we don't have the same event twice.
+
+        Pick the earliest non-outlier if there is one, else the earliest one.
+
+        Args:
+            events_and_contexts (list[(EventBase, EventContext)]):
+        Returns:
+            list[(EventBase, EventContext)]: filtered list
+        """
         new_events_and_contexts = OrderedDict()
         for event, context in events_and_contexts:
             prev_event_context = new_events_and_contexts.get(event.event_id)
@@ -624,9 +796,17 @@ class EventsStore(SQLBaseStore):
                         new_events_and_contexts[event.event_id] = (event, context)
             else:
                 new_events_and_contexts[event.event_id] = (event, context)
+        return new_events_and_contexts.values()
 
-        events_and_contexts = new_events_and_contexts.values()
+    def _update_room_depths_txn(self, txn, events_and_contexts, backfilled):
+        """Update min_depth for each room
 
+        Args:
+            txn (twisted.enterprise.adbapi.Connection): db connection
+            events_and_contexts (list[(EventBase, EventContext)]): events
+                we are persisting
+            backfilled (bool): True if the events were backfilled
+        """
         depth_updates = {}
         for event, context in events_and_contexts:
             # Remove the any existing cache entries for the event_ids
@@ -642,9 +822,24 @@ class EventsStore(SQLBaseStore):
                     event.depth, depth_updates.get(event.room_id, event.depth)
                 )
 
-        for room_id, depth in depth_updates.items():
+        for room_id, depth in depth_updates.iteritems():
             self._update_min_depth_for_room_txn(txn, room_id, depth)
 
+    def _update_outliers_txn(self, txn, events_and_contexts):
+        """Update any outliers with new event info.
+
+        This turns outliers into ex-outliers (unless the new event was
+        rejected).
+
+        Args:
+            txn (twisted.enterprise.adbapi.Connection): db connection
+            events_and_contexts (list[(EventBase, EventContext)]): events
+                we are persisting
+
+        Returns:
+            list[(EventBase, EventContext)] new list, without events which
+            are already in the events table.
+        """
         txn.execute(
             "SELECT event_id, outlier FROM events WHERE event_id in (%s)" % (
                 ",".join(["?"] * len(events_and_contexts)),
@@ -654,24 +849,21 @@ class EventsStore(SQLBaseStore):
 
         have_persisted = {
             event_id: outlier
-            for event_id, outlier in txn.fetchall()
+            for event_id, outlier in txn
         }
 
         to_remove = set()
         for event, context in events_and_contexts:
-            if context.rejected:
-                # If the event is rejected then we don't care if the event
-                # was an outlier or not.
-                if event.event_id in have_persisted:
-                    # If we have already seen the event then ignore it.
-                    to_remove.add(event)
-                continue
-
             if event.event_id not in have_persisted:
                 continue
 
             to_remove.add(event)
 
+            if context.rejected:
+                # If the event is rejected then we don't care if the event
+                # was an outlier or not.
+                continue
+
             outlier_persisted = have_persisted[event.event_id]
             if not event.internal_metadata.is_outlier() and outlier_persisted:
                 # We received a copy of an event that we had already stored as
@@ -726,37 +918,19 @@ class EventsStore(SQLBaseStore):
                 # event isn't an outlier any more.
                 self._update_backward_extremeties(txn, [event])
 
-        events_and_contexts = [
+        return [
             ec for ec in events_and_contexts if ec[0] not in to_remove
         ]
 
+    @classmethod
+    def _delete_existing_rows_txn(cls, txn, events_and_contexts):
         if not events_and_contexts:
-            # Make sure we don't pass an empty list to functions that expect to
-            # be storing at least one element.
+            # nothing to do here
             return
 
-        # From this point onwards the events are only events that we haven't
-        # seen before.
+        logger.info("Deleting existing")
 
-        def event_dict(event):
-            return {
-                k: v
-                for k, v in event.get_dict().items()
-                if k not in [
-                    "redacted",
-                    "redacted_because",
-                ]
-            }
-
-        if delete_existing:
-            # For paranoia reasons, we go and delete all the existing entries
-            # for these events so we can reinsert them.
-            # This gets around any problems with some tables already having
-            # entries.
-
-            logger.info("Deleting existing")
-
-            for table in (
+        for table in (
                 "events",
                 "event_auth",
                 "event_json",
@@ -778,18 +952,30 @@ class EventsStore(SQLBaseStore):
                 "redactions",
                 "room_memberships",
                 "topics"
-            ):
-                txn.executemany(
-                    "DELETE FROM %s WHERE event_id = ?" % (table,),
-                    [(ev.event_id,) for ev, _ in events_and_contexts]
-                )
-            for table in (
-                "event_push_actions",
-            ):
-                txn.executemany(
-                    "DELETE FROM %s WHERE room_id = ? AND event_id = ?" % (table,),
-                    [(ev.event_id,) for ev, _ in events_and_contexts]
-                )
+        ):
+            txn.executemany(
+                "DELETE FROM %s WHERE event_id = ?" % (table,),
+                [(ev.event_id,) for ev, _ in events_and_contexts]
+            )
+
+    def _store_event_txn(self, txn, events_and_contexts):
+        """Insert new events into the event and event_json tables
+
+        Args:
+            txn (twisted.enterprise.adbapi.Connection): db connection
+            events_and_contexts (list[(EventBase, EventContext)]): events
+                we are persisting
+        """
+
+        if not events_and_contexts:
+            # nothing to do here
+            return
+
+        def event_dict(event):
+            d = event.get_dict()
+            d.pop("redacted", None)
+            d.pop("redacted_because", None)
+            return d
 
         self._simple_insert_many_txn(
             txn,
@@ -833,6 +1019,19 @@ class EventsStore(SQLBaseStore):
             ],
         )
 
+    def _store_rejected_events_txn(self, txn, events_and_contexts):
+        """Add rows to the 'rejections' table for received events which were
+        rejected
+
+        Args:
+            txn (twisted.enterprise.adbapi.Connection): db connection
+            events_and_contexts (list[(EventBase, EventContext)]): events
+                we are persisting
+
+        Returns:
+            list[(EventBase, EventContext)] new list, without the rejected
+                events.
+        """
         # Remove the rejected events from the list now that we've added them
         # to the events table and the events_json table.
         to_remove = set()
@@ -844,16 +1043,23 @@ class EventsStore(SQLBaseStore):
                 )
                 to_remove.add(event)
 
-        events_and_contexts = [
+        return [
             ec for ec in events_and_contexts if ec[0] not in to_remove
         ]
 
-        if not events_and_contexts:
-            # Make sure we don't pass an empty list to functions that expect to
-            # be storing at least one element.
-            return
+    def _update_metadata_tables_txn(self, txn, events_and_contexts, backfilled):
+        """Update all the miscellaneous tables for new events
 
-        # From this point onwards the events are only ones that weren't rejected.
+        Args:
+            txn (twisted.enterprise.adbapi.Connection): db connection
+            events_and_contexts (list[(EventBase, EventContext)]): events
+                we are persisting
+            backfilled (bool): True if the events were backfilled
+        """
+
+        if not events_and_contexts:
+            # nothing to do here
+            return
 
         for event, context in events_and_contexts:
             # Insert all the push actions into the event_push_actions table.
@@ -883,10 +1089,6 @@ class EventsStore(SQLBaseStore):
             ],
         )
 
-        # Insert into the state_groups, state_groups_state, and
-        # event_to_state_groups tables.
-        self._store_mult_state_groups_txn(txn, events_and_contexts)
-
         # Update the event_forward_extremities, event_backward_extremities and
         # event_edges tables.
         self._handle_mult_prev_events(
@@ -973,13 +1175,6 @@ class EventsStore(SQLBaseStore):
         # Prefill the event cache
         self._add_to_cache(txn, events_and_contexts)
 
-        if backfilled:
-            # Backfilled events come before the current state so we don't need
-            # to update the current state table
-            return
-
-        return
-
     def _add_to_cache(self, txn, events_and_contexts):
         to_prefill = []
 
@@ -1575,6 +1770,94 @@ class EventsStore(SQLBaseStore):
         """The current minimum token that backfilled events have reached"""
         return -self._backfill_id_gen.get_current_token()
 
+    def get_current_events_token(self):
+        """The current maximum token that events have reached"""
+        return self._stream_id_gen.get_current_token()
+
+    def get_all_new_forward_event_rows(self, last_id, current_id, limit):
+        if last_id == current_id:
+            return defer.succeed([])
+
+        def get_all_new_forward_event_rows(txn):
+            sql = (
+                "SELECT e.stream_ordering, e.event_id, e.room_id, e.type,"
+                " state_key, redacts"
+                " FROM events AS e"
+                " LEFT JOIN redactions USING (event_id)"
+                " LEFT JOIN state_events USING (event_id)"
+                " WHERE ? < stream_ordering AND stream_ordering <= ?"
+                " ORDER BY stream_ordering ASC"
+                " LIMIT ?"
+            )
+            txn.execute(sql, (last_id, current_id, limit))
+            new_event_updates = txn.fetchall()
+
+            if len(new_event_updates) == limit:
+                upper_bound = new_event_updates[-1][0]
+            else:
+                upper_bound = current_id
+
+            sql = (
+                "SELECT event_stream_ordering, e.event_id, e.room_id, e.type,"
+                " state_key, redacts"
+                " FROM events AS e"
+                " INNER JOIN ex_outlier_stream USING (event_id)"
+                " LEFT JOIN redactions USING (event_id)"
+                " LEFT JOIN state_events USING (event_id)"
+                " WHERE ? < event_stream_ordering"
+                " AND event_stream_ordering <= ?"
+                " ORDER BY event_stream_ordering DESC"
+            )
+            txn.execute(sql, (last_id, upper_bound))
+            new_event_updates.extend(txn)
+
+            return new_event_updates
+        return self.runInteraction(
+            "get_all_new_forward_event_rows", get_all_new_forward_event_rows
+        )
+
+    def get_all_new_backfill_event_rows(self, last_id, current_id, limit):
+        if last_id == current_id:
+            return defer.succeed([])
+
+        def get_all_new_backfill_event_rows(txn):
+            sql = (
+                "SELECT -e.stream_ordering, e.event_id, e.room_id, e.type,"
+                " state_key, redacts"
+                " FROM events AS e"
+                " LEFT JOIN redactions USING (event_id)"
+                " LEFT JOIN state_events USING (event_id)"
+                " WHERE ? > stream_ordering AND stream_ordering >= ?"
+                " ORDER BY stream_ordering ASC"
+                " LIMIT ?"
+            )
+            txn.execute(sql, (-last_id, -current_id, limit))
+            new_event_updates = txn.fetchall()
+
+            if len(new_event_updates) == limit:
+                upper_bound = new_event_updates[-1][0]
+            else:
+                upper_bound = current_id
+
+            sql = (
+                "SELECT -event_stream_ordering, e.event_id, e.room_id, e.type,"
+                " state_key, redacts"
+                " FROM events AS e"
+                " INNER JOIN ex_outlier_stream USING (event_id)"
+                " LEFT JOIN redactions USING (event_id)"
+                " LEFT JOIN state_events USING (event_id)"
+                " WHERE ? > event_stream_ordering"
+                " AND event_stream_ordering >= ?"
+                " ORDER BY event_stream_ordering DESC"
+            )
+            txn.execute(sql, (-last_id, -upper_bound))
+            new_event_updates.extend(txn.fetchall())
+
+            return new_event_updates
+        return self.runInteraction(
+            "get_all_new_backfill_event_rows", get_all_new_backfill_event_rows
+        )
+
     @cached(num_args=5, max_entries=10)
     def get_all_new_events(self, last_backfill_id, last_forward_id,
                            current_backfill_id, current_forward_id, limit):
@@ -1588,14 +1871,13 @@ class EventsStore(SQLBaseStore):
 
         def get_all_new_events_txn(txn):
             sql = (
-                "SELECT e.stream_ordering, ej.internal_metadata, ej.json, eg.state_group"
-                " FROM events as e"
-                " JOIN event_json as ej"
-                " ON e.event_id = ej.event_id AND e.room_id = ej.room_id"
-                " LEFT JOIN event_to_state_groups as eg"
-                " ON e.event_id = eg.event_id"
-                " WHERE ? < e.stream_ordering AND e.stream_ordering <= ?"
-                " ORDER BY e.stream_ordering ASC"
+                "SELECT e.stream_ordering, e.event_id, e.room_id, e.type,"
+                " state_key, redacts"
+                " FROM events AS e"
+                " LEFT JOIN redactions USING (event_id)"
+                " LEFT JOIN state_events USING (event_id)"
+                " WHERE ? < stream_ordering AND stream_ordering <= ?"
+                " ORDER BY stream_ordering ASC"
                 " LIMIT ?"
             )
             if have_forward_events:
@@ -1621,15 +1903,13 @@ class EventsStore(SQLBaseStore):
                 forward_ex_outliers = []
 
             sql = (
-                "SELECT -e.stream_ordering, ej.internal_metadata, ej.json,"
-                " eg.state_group"
-                " FROM events as e"
-                " JOIN event_json as ej"
-                " ON e.event_id = ej.event_id AND e.room_id = ej.room_id"
-                " LEFT JOIN event_to_state_groups as eg"
-                " ON e.event_id = eg.event_id"
-                " WHERE ? > e.stream_ordering AND e.stream_ordering >= ?"
-                " ORDER BY e.stream_ordering DESC"
+                "SELECT -e.stream_ordering, e.event_id, e.room_id, e.type,"
+                " state_key, redacts"
+                " FROM events AS e"
+                " LEFT JOIN redactions USING (event_id)"
+                " LEFT JOIN state_events USING (event_id)"
+                " WHERE ? > stream_ordering AND stream_ordering >= ?"
+                " ORDER BY stream_ordering DESC"
                 " LIMIT ?"
             )
             if have_backfill_events:
@@ -1816,7 +2096,7 @@ class EventsStore(SQLBaseStore):
                         "state_key": key[1],
                         "event_id": state_id,
                     }
-                    for key, state_id in curr_state.items()
+                    for key, state_id in curr_state.iteritems()
                 ],
             )
 

synapse/storage/keys.py

@@ -101,9 +101,10 @@ class KeyStore(SQLBaseStore):
         key_ids
         Args:
             server_name (str): The name of the server.
-            key_ids (list of str): List of key_ids to try and look up.
+            key_ids (iterable[str]): key_ids to try and look up.
         Returns:
-            (list of VerifyKey): The verification keys.
+            Deferred: resolves to dict[str, VerifyKey]: map from
+               key_id to verification key.
         """
         keys = {}
         for key_id in key_ids:

synapse/storage/prepare_database.py

@@ -25,7 +25,7 @@ logger = logging.getLogger(__name__)
 
 # Remember to update this number every time a change is made to database
 # schema files, so the users will be informed on server restarts.
-SCHEMA_VERSION = 40
+SCHEMA_VERSION = 41
 
 dir_path = os.path.abspath(os.path.dirname(__file__))
 
@@ -356,7 +356,7 @@ def _get_or_create_schema_state(txn, database_engine):
             ),
             (current_version,)
         )
-        applied_deltas = [d for d, in txn.fetchall()]
+        applied_deltas = [d for d, in txn]
         return current_version, applied_deltas, upgraded
 
     return None

synapse/storage/presence.py

@@ -85,8 +85,8 @@ class PresenceStore(SQLBaseStore):
                 self.presence_stream_cache.entity_has_changed,
                 state.user_id, stream_id,
             )
-            self._invalidate_cache_and_stream(
-                txn, self._get_presence_for_user, (state.user_id,)
+            txn.call_after(
+                self._get_presence_for_user.invalidate, (state.user_id,)
             )
 
         # Actually insert new rows

synapse/storage/pusher.py

@@ -135,6 +135,48 @@ class PusherStore(SQLBaseStore):
             "get_all_updated_pushers", get_all_updated_pushers_txn
         )
 
+    def get_all_updated_pushers_rows(self, last_id, current_id, limit):
+        """Get all the pushers that have changed between the given tokens.
+
+        Returns:
+            list(tuple): each tuple consists of:
+                stream_id (str)
+                user_id (str)
+                app_id (str)
+                pushkey (str)
+                was_deleted (bool): whether the pusher was added/updated (False)
+                    or deleted (True)
+        """
+
+        if last_id == current_id:
+            return defer.succeed([])
+
+        def get_all_updated_pushers_rows_txn(txn):
+            sql = (
+                "SELECT id, user_name, app_id, pushkey"
+                " FROM pushers"
+                " WHERE ? < id AND id <= ?"
+                " ORDER BY id ASC LIMIT ?"
+            )
+            txn.execute(sql, (last_id, current_id, limit))
+            results = [list(row) + [False] for row in txn]
+
+            sql = (
+                "SELECT stream_id, user_id, app_id, pushkey"
+                " FROM deleted_pushers"
+                " WHERE ? < stream_id AND stream_id <= ?"
+                " ORDER BY stream_id ASC LIMIT ?"
+            )
+            txn.execute(sql, (last_id, current_id, limit))
+
+            results.extend(list(row) + [True] for row in txn)
+            results.sort()  # Sort so that they're ordered by stream id
+
+            return results
+        return self.runInteraction(
+            "get_all_updated_pushers_rows", get_all_updated_pushers_rows_txn
+        )
+
     @cachedInlineCallbacks(num_args=1, max_entries=15000)
     def get_if_user_has_pusher(self, user_id):
         # This only exists for the cachedList decorator

synapse/storage/receipts.py

@@ -313,10 +313,9 @@ class ReceiptsStore(SQLBaseStore):
         )
 
         txn.execute(sql, (room_id, receipt_type, user_id))
-        results = txn.fetchall()
 
-        if results and topological_ordering:
-            for to, so, _ in results:
+        if topological_ordering:
+            for to, so, _ in txn:
                 if int(to) > topological_ordering:
                     return False
                 elif int(to) == topological_ordering and int(so) >= stream_ordering:

synapse/storage/registration.py

@@ -209,7 +209,7 @@ class RegistrationStore(background_updates.BackgroundUpdateStore):
                 " WHERE lower(name) = lower(?)"
             )
             txn.execute(sql, (user_id,))
-            return dict(txn.fetchall())
+            return dict(txn)
 
         return self.runInteraction("get_users_by_id_case_insensitive", f)
 

synapse/storage/room.py

@@ -396,7 +396,7 @@ class RoomStore(SQLBaseStore):
                     sql % ("AND appservice_id IS NULL",),
                     (stream_id,)
                 )
-            return dict(txn.fetchall())
+            return dict(txn)
         else:
             # We want to get from all lists, so we need to aggregate the results
 
@@ -422,7 +422,7 @@ class RoomStore(SQLBaseStore):
 
             results = {}
             # A room is visible if its visible on any list.
-            for room_id, visibility in txn.fetchall():
+            for room_id, visibility in txn:
                 results[room_id] = bool(visibility) or results.get(room_id, False)
 
             return results

synapse/storage/roommember.py

@@ -129,17 +129,30 @@ class RoomMemberStore(SQLBaseStore):
         with self._stream_id_gen.get_next() as stream_ordering:
             yield self.runInteraction("locally_reject_invite", f, stream_ordering)
 
+    @cachedInlineCallbacks(max_entries=100000, iterable=True, cache_context=True)
+    def get_hosts_in_room(self, room_id, cache_context):
+        """Returns the set of all hosts currently in the room
+        """
+        user_ids = yield self.get_users_in_room(
+            room_id, on_invalidate=cache_context.invalidate,
+        )
+        hosts = frozenset(get_domain_from_id(user_id) for user_id in user_ids)
+        defer.returnValue(hosts)
+
     @cached(max_entries=500000, iterable=True)
     def get_users_in_room(self, room_id):
         def f(txn):
-
-            rows = self._get_members_rows_txn(
-                txn,
-                room_id=room_id,
-                membership=Membership.JOIN,
+            sql = (
+                "SELECT m.user_id FROM room_memberships as m"
+                " INNER JOIN current_state_events as c"
+                " ON m.event_id = c.event_id "
+                " AND m.room_id = c.room_id "
+                " AND m.user_id = c.state_key"
+                " WHERE c.type = 'm.room.member' AND c.room_id = ? AND m.membership = ?"
             )
 
-            return [r["user_id"] for r in rows]
+            txn.execute(sql, (room_id, Membership.JOIN,))
+            return [r[0] for r in txn]
         return self.runInteraction("get_users_in_room", f)
 
     @cached()
@@ -246,52 +259,27 @@ class RoomMemberStore(SQLBaseStore):
 
         return results
 
-    def _get_members_rows_txn(self, txn, room_id, membership=None, user_id=None):
-        where_clause = "c.room_id = ?"
-        where_values = [room_id]
-
-        if membership:
-            where_clause += " AND m.membership = ?"
-            where_values.append(membership)
-
-        if user_id:
-            where_clause += " AND m.user_id = ?"
-            where_values.append(user_id)
-
-        sql = (
-            "SELECT m.* FROM room_memberships as m"
-            " INNER JOIN current_state_events as c"
-            " ON m.event_id = c.event_id "
-            " AND m.room_id = c.room_id "
-            " AND m.user_id = c.state_key"
-            " WHERE c.type = 'm.room.member' AND %(where)s"
-        ) % {
-            "where": where_clause,
-        }
-
-        txn.execute(sql, where_values)
-        rows = self.cursor_to_dict(txn)
-
-        return rows
-
-    @cached(max_entries=500000, iterable=True)
+    @cachedInlineCallbacks(max_entries=500000, iterable=True)
     def get_rooms_for_user(self, user_id):
-        return self.get_rooms_for_user_where_membership_is(
+        """Returns a set of room_ids the user is currently joined to
+        """
+        rooms = yield self.get_rooms_for_user_where_membership_is(
             user_id, membership_list=[Membership.JOIN],
         )
+        defer.returnValue(frozenset(r.room_id for r in rooms))
 
     @cachedInlineCallbacks(max_entries=500000, cache_context=True, iterable=True)
     def get_users_who_share_room_with_user(self, user_id, cache_context):
         """Returns the set of users who share a room with `user_id`
         """
-        rooms = yield self.get_rooms_for_user(
+        room_ids = yield self.get_rooms_for_user(
             user_id, on_invalidate=cache_context.invalidate,
         )
 
         user_who_share_room = set()
-        for room in rooms:
+        for room_id in room_ids:
             user_ids = yield self.get_users_in_room(
-                room.room_id, on_invalidate=cache_context.invalidate,
+                room_id, on_invalidate=cache_context.invalidate,
             )
             user_who_share_room.update(user_ids)
 

synapse/storage/schema/delta/41/device_list_stream_idx.sql

@@ -0,0 +1,17 @@
+/* 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.
+ */
+
+INSERT into background_updates (update_name, progress_json)
+    VALUES ('device_lists_stream_idx', '{}');

synapse/storage/schema/delta/41/device_outbound_index.sql

@@ -0,0 +1,16 @@
+/* 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.
+ */
+
+CREATE INDEX device_lists_outbound_pokes_stream ON device_lists_outbound_pokes(stream_id);

synapse/storage/signatures.py

@@ -72,7 +72,7 @@ class SignatureStore(SQLBaseStore):
             " WHERE event_id = ?"
         )
         txn.execute(query, (event_id, ))
-        return {k: v for k, v in txn.fetchall()}
+        return {k: v for k, v in txn}
 
     def _store_event_reference_hashes_txn(self, txn, events):
         """Store a hash for a PDU

synapse/storage/state.py

@@ -14,7 +14,7 @@
 # limitations under the License.
 
 from ._base import SQLBaseStore
-from synapse.util.caches.descriptors import cached, cachedList
+from synapse.util.caches.descriptors import cached, cachedList, cachedInlineCallbacks
 from synapse.util.caches import intern_string
 from synapse.storage.engines import PostgresEngine
 
@@ -69,6 +69,18 @@ class StateStore(SQLBaseStore):
             where_clause="type='m.room.member'",
         )
 
+    @cachedInlineCallbacks(max_entries=100000, iterable=True)
+    def get_current_state_ids(self, room_id):
+        rows = yield self._simple_select_list(
+            table="current_state_events",
+            keyvalues={"room_id": room_id},
+            retcols=["event_id", "type", "state_key"],
+            desc="_calculate_state_delta",
+        )
+        defer.returnValue({
+            (r["type"], r["state_key"]): r["event_id"] for r in rows
+        })
+
     @defer.inlineCallbacks
     def get_state_groups_ids(self, room_id, event_ids):
         if not event_ids:
@@ -78,7 +90,7 @@ class StateStore(SQLBaseStore):
             event_ids,
         )
 
-        groups = set(event_to_groups.values())
+        groups = set(event_to_groups.itervalues())
         group_to_state = yield self._get_state_for_groups(groups)
 
         defer.returnValue(group_to_state)
@@ -96,17 +108,18 @@ class StateStore(SQLBaseStore):
 
         state_event_map = yield self.get_events(
             [
-                ev_id for group_ids in group_to_ids.values()
-                for ev_id in group_ids.values()
+                ev_id for group_ids in group_to_ids.itervalues()
+                for ev_id in group_ids.itervalues()
             ],
             get_prev_content=False
         )
 
         defer.returnValue({
             group: [
-                state_event_map[v] for v in event_id_map.values() if v in state_event_map
+                state_event_map[v] for v in event_id_map.itervalues()
+                if v in state_event_map
             ]
-            for group, event_id_map in group_to_ids.items()
+            for group, event_id_map in group_to_ids.iteritems()
         })
 
     def _have_persisted_state_group_txn(self, txn, state_group):
@@ -124,6 +137,16 @@ class StateStore(SQLBaseStore):
                 continue
 
             if context.current_state_ids is None:
+                # AFAIK, this can never happen
+                logger.error(
+                    "Non-outlier event %s had current_state_ids==None",
+                    event.event_id)
+                continue
+
+            # if the event was rejected, just give it the same state as its
+            # predecessor.
+            if context.rejected:
+                state_groups[event.event_id] = context.prev_group
                 continue
 
             state_groups[event.event_id] = context.state_group
@@ -168,7 +191,7 @@ class StateStore(SQLBaseStore):
                             "state_key": key[1],
                             "event_id": state_id,
                         }
-                        for key, state_id in context.delta_ids.items()
+                        for key, state_id in context.delta_ids.iteritems()
                     ],
                 )
             else:
@@ -183,7 +206,7 @@ class StateStore(SQLBaseStore):
                             "state_key": key[1],
                             "event_id": state_id,
                         }
-                        for key, state_id in context.current_state_ids.items()
+                        for key, state_id in context.current_state_ids.iteritems()
                     ],
                 )
 
@@ -195,7 +218,7 @@ class StateStore(SQLBaseStore):
                     "state_group": state_group_id,
                     "event_id": event_id,
                 }
-                for event_id, state_group_id in state_groups.items()
+                for event_id, state_group_id in state_groups.iteritems()
             ],
         )
 
@@ -319,10 +342,10 @@ class StateStore(SQLBaseStore):
                     args.extend(where_args)
 
                     txn.execute(sql % (where_clause,), args)
-                    rows = self.cursor_to_dict(txn)
-                    for row in rows:
-                        key = (row["type"], row["state_key"])
-                        results[group][key] = row["event_id"]
+                    for row in txn:
+                        typ, state_key, event_id = row
+                        key = (typ, state_key)
+                        results[group][key] = event_id
         else:
             if types is not None:
                 where_clause = "AND (%s)" % (
@@ -351,12 +374,11 @@ class StateStore(SQLBaseStore):
                         " WHERE state_group = ? %s" % (where_clause,),
                         args
                     )
-                    rows = txn.fetchall()
-                    results[group].update({
-                        (typ, state_key): event_id
-                        for typ, state_key, event_id in rows
+                    results[group].update(
+                        ((typ, state_key), event_id)
+                        for typ, state_key, event_id in txn
                         if (typ, state_key) not in results[group]
-                    })
+                    )
 
                     # If the lengths match then we must have all the types,
                     # so no need to go walk further down the tree.
@@ -393,21 +415,21 @@ class StateStore(SQLBaseStore):
             event_ids,
         )
 
-        groups = set(event_to_groups.values())
+        groups = set(event_to_groups.itervalues())
         group_to_state = yield self._get_state_for_groups(groups, types)
 
         state_event_map = yield self.get_events(
-            [ev_id for sd in group_to_state.values() for ev_id in sd.values()],
+            [ev_id for sd in group_to_state.itervalues() for ev_id in sd.itervalues()],
             get_prev_content=False
         )
 
         event_to_state = {
             event_id: {
                 k: state_event_map[v]
-                for k, v in group_to_state[group].items()
+                for k, v in group_to_state[group].iteritems()
                 if v in state_event_map
             }
-            for event_id, group in event_to_groups.items()
+            for event_id, group in event_to_groups.iteritems()
         }
 
         defer.returnValue({event: event_to_state[event] for event in event_ids})
@@ -430,12 +452,12 @@ class StateStore(SQLBaseStore):
             event_ids,
         )
 
-        groups = set(event_to_groups.values())
+        groups = set(event_to_groups.itervalues())
         group_to_state = yield self._get_state_for_groups(groups, types)
 
         event_to_state = {
             event_id: group_to_state[group]
-            for event_id, group in event_to_groups.items()
+            for event_id, group in event_to_groups.iteritems()
         }
 
         defer.returnValue({event: event_to_state[event] for event in event_ids})
@@ -474,7 +496,7 @@ class StateStore(SQLBaseStore):
         state_map = yield self.get_state_ids_for_events([event_id], types)
         defer.returnValue(state_map[event_id])
 
-    @cached(num_args=2, max_entries=10000)
+    @cached(num_args=2, max_entries=100000)
     def _get_state_group_for_event(self, room_id, event_id):
         return self._simple_select_one_onecol(
             table="event_to_state_groups",
@@ -547,7 +569,7 @@ class StateStore(SQLBaseStore):
         got_all = not (missing_types or types is None)
 
         return {
-            k: v for k, v in state_dict_ids.items()
+            k: v for k, v in state_dict_ids.iteritems()
             if include(k[0], k[1])
         }, missing_types, got_all
 
@@ -606,7 +628,7 @@ class StateStore(SQLBaseStore):
 
             # Now we want to update the cache with all the things we fetched
             # from the database.
-            for group, group_state_dict in group_to_state_dict.items():
+            for group, group_state_dict in group_to_state_dict.iteritems():
                 if types:
                     # We delibrately put key -> None mappings into the cache to
                     # cache absence of the key, on the assumption that if we've
@@ -621,10 +643,10 @@ class StateStore(SQLBaseStore):
                 else:
                     state_dict = results[group]
 
-                state_dict.update({
-                    (intern_string(k[0]), intern_string(k[1])): v
-                    for k, v in group_state_dict.items()
-                })
+                state_dict.update(
+                    ((intern_string(k[0]), intern_string(k[1])), v)
+                    for k, v in group_state_dict.iteritems()
+                )
 
                 self._state_group_cache.update(
                     cache_seq_num,
@@ -635,10 +657,10 @@ class StateStore(SQLBaseStore):
 
         # Remove all the entries with None values. The None values were just
         # used for bookkeeping in the cache.
-        for group, state_dict in results.items():
+        for group, state_dict in results.iteritems():
             results[group] = {
                 key: event_id
-                for key, event_id in state_dict.items()
+                for key, event_id in state_dict.iteritems()
                 if event_id
             }
 
@@ -727,7 +749,7 @@ class StateStore(SQLBaseStore):
                         # of keys
 
                         delta_state = {
-                            key: value for key, value in curr_state.items()
+                            key: value for key, value in curr_state.iteritems()
                             if prev_state.get(key, None) != value
                         }
 
@@ -767,7 +789,7 @@ class StateStore(SQLBaseStore):
                                     "state_key": key[1],
                                     "event_id": state_id,
                                 }
-                                for key, state_id in delta_state.items()
+                                for key, state_id in delta_state.iteritems()
                             ],
                         )
 

synapse/storage/stream.py

@@ -829,3 +829,6 @@ class StreamStore(SQLBaseStore):
             updatevalues={"stream_id": stream_id},
             desc="update_federation_out_pos",
         )
+
+    def has_room_changed_since(self, room_id, stream_id):
+        return self._events_stream_cache.has_entity_changed(room_id, stream_id)

synapse/storage/tags.py

@@ -95,7 +95,7 @@ class TagsStore(SQLBaseStore):
             for stream_id, user_id, room_id in tag_ids:
                 txn.execute(sql, (user_id, room_id))
                 tags = []
-                for tag, content in txn.fetchall():
+                for tag, content in txn:
                     tags.append(json.dumps(tag) + ":" + content)
                 tag_json = "{" + ",".join(tags) + "}"
                 results.append((stream_id, user_id, room_id, tag_json))
@@ -132,7 +132,7 @@ class TagsStore(SQLBaseStore):
                 " WHERE user_id = ? AND stream_id > ?"
             )
             txn.execute(sql, (user_id, stream_id))
-            room_ids = [row[0] for row in txn.fetchall()]
+            room_ids = [row[0] for row in txn]
             return room_ids
 
         changed = self._account_data_stream_cache.has_entity_changed(

synapse/storage/util/id_generators.py

@@ -30,6 +30,17 @@ class IdGenerator(object):
 
 
 def _load_current_id(db_conn, table, column, step=1):
+    """
+
+    Args:
+        db_conn (object):
+        table (str):
+        column (str):
+        step (int):
+
+    Returns:
+        int
+    """
     cur = db_conn.cursor()
     if step == 1:
         cur.execute("SELECT MAX(%s) FROM %s" % (column, table,))
@@ -131,6 +142,9 @@ class StreamIdGenerator(object):
     def get_current_token(self):
         """Returns the maximum stream id such that all stream ids less than or
         equal to it have been successfully persisted.
+
+        Returns:
+            int
         """
         with self._lock:
             if self._unfinished_ids:

synapse/types.py

@@ -216,9 +216,7 @@ class StreamToken(
             return self
 
     def copy_and_replace(self, key, new_value):
-        d = self._asdict()
-        d[key] = new_value
-        return StreamToken(**d)
+        return self._replace(**{key: new_value})
 
 
 StreamToken.START = StreamToken(

synapse/util/__init__.py

@@ -26,7 +26,7 @@ logger = logging.getLogger(__name__)
 
 class DeferredTimedOutError(SynapseError):
     def __init__(self):
-        super(SynapseError).__init__(504, "Timed out")
+        super(SynapseError, self).__init__(504, "Timed out")
 
 
 def unwrapFirstError(failure):
@@ -93,8 +93,10 @@ class Clock(object):
         ret_deferred = defer.Deferred()
 
         def timed_out_fn():
+            e = DeferredTimedOutError()
+
             try:
-                ret_deferred.errback(DeferredTimedOutError())
+                ret_deferred.errback(e)
             except:
                 pass
 
@@ -114,7 +116,7 @@ class Clock(object):
 
         ret_deferred.addBoth(cancel)
 
-        def sucess(res):
+        def success(res):
             try:
                 ret_deferred.callback(res)
             except:
@@ -128,7 +130,7 @@ class Clock(object):
             except:
                 pass
 
-        given_deferred.addCallbacks(callback=sucess, errback=err)
+        given_deferred.addCallbacks(callback=success, errback=err)
 
         timer = self.call_later(time_out, timed_out_fn)
 

synapse/util/async.py

@@ -89,6 +89,11 @@ class ObservableDeferred(object):
         deferred.addCallbacks(callback, errback)
 
     def observe(self):
+        """Observe the underlying deferred.
+
+        Can return either a deferred if the underlying deferred is still pending
+        (or has failed), or the actual value. Callers may need to use maybeDeferred.
+        """
         if not self._result:
             d = defer.Deferred()
 
@@ -101,7 +106,7 @@ class ObservableDeferred(object):
             return d
         else:
             success, res = self._result
-            return defer.succeed(res) if success else defer.fail(res)
+            return res if success else defer.fail(res)
 
     def observers(self):
         return self._observers