mirror of
https://github.com/element-hq/synapse.git
synced 2025-12-09 01:30:18 +00:00
Compare commits
1 Commits
v1.124.0
...
dependabot
| Author | SHA1 | Date | |
|---|---|---|---|
|
|
26562f1ebb |
52
CHANGES.md
52
CHANGES.md
@@ -1,55 +1,3 @@
|
||||
# Synapse 1.124.0 (2025-02-11)
|
||||
|
||||
No significant changes since 1.124.0rc3.
|
||||
|
||||
|
||||
|
||||
|
||||
# Synapse 1.124.0rc3 (2025-02-07)
|
||||
|
||||
### Bugfixes
|
||||
|
||||
- Fix regression in performance of sending events due to superfluous reads and locks. Introduced in v1.124.0rc1. ([\#18141](https://github.com/element-hq/synapse/issues/18141))
|
||||
|
||||
|
||||
|
||||
|
||||
# Synapse 1.124.0rc2 (2025-02-05)
|
||||
|
||||
### Bugfixes
|
||||
|
||||
- Fix regression where persisting events in some rooms could fail after a previous unclean shutdown. Introduced in v1.124.0rc1. ([\#18137](https://github.com/element-hq/synapse/issues/18137))
|
||||
|
||||
|
||||
|
||||
|
||||
# Synapse 1.124.0rc1 (2025-02-04)
|
||||
|
||||
### Bugfixes
|
||||
|
||||
- Add rate limit `rc_presence.per_user`. This prevents load from excessive presence updates sent by clients via sync api. Also rate limit `/_matrix/client/v3/presence` as per the spec. Contributed by @rda0. ([\#18000](https://github.com/element-hq/synapse/issues/18000))
|
||||
- Deactivated users will no longer automatically accept an invite when `auto_accept_invites` is enabled. ([\#18073](https://github.com/element-hq/synapse/issues/18073))
|
||||
- Fix join being denied after being invited over federation. Also fixes other out-of-band membership transitions. ([\#18075](https://github.com/element-hq/synapse/issues/18075))
|
||||
- Updates contributed `docker-compose.yml` file to PostgreSQL v15, as v12 is no longer supported by Synapse.
|
||||
Contributed by @maxkratz. ([\#18089](https://github.com/element-hq/synapse/issues/18089))
|
||||
- Fix rare edge case where state groups could be deleted while we are persisting new events that reference them. ([\#18107](https://github.com/element-hq/synapse/issues/18107), [\#18130](https://github.com/element-hq/synapse/issues/18130), [\#18131](https://github.com/element-hq/synapse/issues/18131))
|
||||
- Raise an error if someone is using an incorrect suffix in a config duration string. ([\#18112](https://github.com/element-hq/synapse/issues/18112))
|
||||
- Fix a bug where the [Delete Room Admin API](https://element-hq.github.io/synapse/latest/admin_api/rooms.html#version-2-new-version) would fail if the `block` parameter was set to `true` and a worker other than the main process was configured to handle background tasks. ([\#18119](https://github.com/element-hq/synapse/issues/18119))
|
||||
|
||||
### Internal Changes
|
||||
|
||||
- Increase the length of the generated `nonce` parameter when perfoming OIDC logins to comply with the TI-Messenger spec. ([\#18109](https://github.com/element-hq/synapse/issues/18109))
|
||||
|
||||
|
||||
|
||||
### Updates to locked dependencies
|
||||
|
||||
* Bump dawidd6/action-download-artifact from 7 to 8. ([\#18108](https://github.com/element-hq/synapse/issues/18108))
|
||||
* Bump log from 0.4.22 to 0.4.25. ([\#18098](https://github.com/element-hq/synapse/issues/18098))
|
||||
* Bump python-multipart from 0.0.18 to 0.0.20. ([\#18096](https://github.com/element-hq/synapse/issues/18096))
|
||||
* Bump serde_json from 1.0.135 to 1.0.137. ([\#18099](https://github.com/element-hq/synapse/issues/18099))
|
||||
* Bump types-bleach from 6.1.0.20240331 to 6.2.0.20241123. ([\#18082](https://github.com/element-hq/synapse/issues/18082))
|
||||
|
||||
# Synapse 1.123.0 (2025-01-28)
|
||||
|
||||
No significant changes since 1.123.0rc1.
|
||||
|
||||
1
changelog.d/18000.bugfix
Normal file
1
changelog.d/18000.bugfix
Normal file
@@ -0,0 +1 @@
|
||||
Add rate limit `rc_presence.per_user`. This prevents load from excessive presence updates sent by clients via sync api. Also rate limit `/_matrix/client/v3/presence` as per the spec. Contributed by @rda0.
|
||||
1
changelog.d/18073.bugfix
Normal file
1
changelog.d/18073.bugfix
Normal file
@@ -0,0 +1 @@
|
||||
Deactivated users will no longer automatically accept an invite when `auto_accept_invites` is enabled.
|
||||
1
changelog.d/18075.bugfix
Normal file
1
changelog.d/18075.bugfix
Normal file
@@ -0,0 +1 @@
|
||||
Fix join being denied after being invited over federation. Also fixes other out-of-band membership transitions.
|
||||
2
changelog.d/18089.bugfix
Normal file
2
changelog.d/18089.bugfix
Normal file
@@ -0,0 +1,2 @@
|
||||
Updates contributed `docker-compose.yml` file to PostgreSQL v15, as v12 is no longer supported by Synapse.
|
||||
Contributed by @maxkratz.
|
||||
1
changelog.d/18109.misc
Normal file
1
changelog.d/18109.misc
Normal file
@@ -0,0 +1 @@
|
||||
Increase the length of the generated `nonce` parameter when perfoming OIDC logins to comply with the TI-Messenger spec.
|
||||
1
changelog.d/18112.bugfix
Normal file
1
changelog.d/18112.bugfix
Normal file
@@ -0,0 +1 @@
|
||||
Raise an error if someone is using an incorrect suffix in a config duration string.
|
||||
1
changelog.d/18119.bugfix
Normal file
1
changelog.d/18119.bugfix
Normal file
@@ -0,0 +1 @@
|
||||
Fix a bug where the [Delete Room Admin API](https://element-hq.github.io/synapse/latest/admin_api/rooms.html#version-2-new-version) would fail if the `block` parameter was set to `true` and a worker other than the main process was configured to handle background tasks.
|
||||
24
debian/changelog
vendored
24
debian/changelog
vendored
@@ -1,27 +1,3 @@
|
||||
matrix-synapse-py3 (1.124.0) stable; urgency=medium
|
||||
|
||||
* New Synapse release 1.124.0.
|
||||
|
||||
-- Synapse Packaging team <packages@matrix.org> Tue, 11 Feb 2025 11:55:22 +0100
|
||||
|
||||
matrix-synapse-py3 (1.124.0~rc3) stable; urgency=medium
|
||||
|
||||
* New Synapse release 1.124.0rc3.
|
||||
|
||||
-- Synapse Packaging team <packages@matrix.org> Fri, 07 Feb 2025 13:42:55 +0000
|
||||
|
||||
matrix-synapse-py3 (1.124.0~rc2) stable; urgency=medium
|
||||
|
||||
* New Synapse release 1.124.0rc2.
|
||||
|
||||
-- Synapse Packaging team <packages@matrix.org> Wed, 05 Feb 2025 16:35:53 +0000
|
||||
|
||||
matrix-synapse-py3 (1.124.0~rc1) stable; urgency=medium
|
||||
|
||||
* New Synapse release 1.124.0rc1.
|
||||
|
||||
-- Synapse Packaging team <packages@matrix.org> Tue, 04 Feb 2025 11:53:05 +0000
|
||||
|
||||
matrix-synapse-py3 (1.123.0) stable; urgency=medium
|
||||
|
||||
* New Synapse release 1.123.0.
|
||||
|
||||
8
poetry.lock
generated
8
poetry.lock
generated
@@ -472,20 +472,20 @@ smmap = ">=3.0.1,<6"
|
||||
|
||||
[[package]]
|
||||
name = "gitpython"
|
||||
version = "3.1.43"
|
||||
version = "3.1.44"
|
||||
description = "GitPython is a Python library used to interact with Git repositories"
|
||||
optional = false
|
||||
python-versions = ">=3.7"
|
||||
files = [
|
||||
{file = "GitPython-3.1.43-py3-none-any.whl", hash = "sha256:eec7ec56b92aad751f9912a73404bc02ba212a23adb2c7098ee668417051a1ff"},
|
||||
{file = "GitPython-3.1.43.tar.gz", hash = "sha256:35f314a9f878467f5453cc1fee295c3e18e52f1b99f10f6cf5b1682e968a9e7c"},
|
||||
{file = "GitPython-3.1.44-py3-none-any.whl", hash = "sha256:9e0e10cda9bed1ee64bc9a6de50e7e38a9c9943241cd7f585f6df3ed28011110"},
|
||||
{file = "gitpython-3.1.44.tar.gz", hash = "sha256:c87e30b26253bf5418b01b0660f818967f3c503193838337fe5e573331249269"},
|
||||
]
|
||||
|
||||
[package.dependencies]
|
||||
gitdb = ">=4.0.1,<5"
|
||||
|
||||
[package.extras]
|
||||
doc = ["sphinx (==4.3.2)", "sphinx-autodoc-typehints", "sphinx-rtd-theme", "sphinxcontrib-applehelp (>=1.0.2,<=1.0.4)", "sphinxcontrib-devhelp (==1.0.2)", "sphinxcontrib-htmlhelp (>=2.0.0,<=2.0.1)", "sphinxcontrib-qthelp (==1.0.3)", "sphinxcontrib-serializinghtml (==1.1.5)"]
|
||||
doc = ["sphinx (>=7.1.2,<7.2)", "sphinx-autodoc-typehints", "sphinx_rtd_theme"]
|
||||
test = ["coverage[toml]", "ddt (>=1.1.1,!=1.4.3)", "mock", "mypy", "pre-commit", "pytest (>=7.3.1)", "pytest-cov", "pytest-instafail", "pytest-mock", "pytest-sugar", "typing-extensions"]
|
||||
|
||||
[[package]]
|
||||
|
||||
@@ -97,7 +97,7 @@ module-name = "synapse.synapse_rust"
|
||||
|
||||
[tool.poetry]
|
||||
name = "matrix-synapse"
|
||||
version = "1.124.0"
|
||||
version = "1.123.0"
|
||||
description = "Homeserver for the Matrix decentralised comms protocol"
|
||||
authors = ["Matrix.org Team and Contributors <packages@matrix.org>"]
|
||||
license = "AGPL-3.0-or-later"
|
||||
|
||||
@@ -151,8 +151,6 @@ class FederationEventHandler:
|
||||
def __init__(self, hs: "HomeServer"):
|
||||
self._clock = hs.get_clock()
|
||||
self._store = hs.get_datastores().main
|
||||
self._state_store = hs.get_datastores().state
|
||||
self._state_deletion_store = hs.get_datastores().state_deletion
|
||||
self._storage_controllers = hs.get_storage_controllers()
|
||||
self._state_storage_controller = self._storage_controllers.state
|
||||
|
||||
@@ -582,9 +580,7 @@ class FederationEventHandler:
|
||||
room_version.identifier,
|
||||
state_maps_to_resolve,
|
||||
event_map=None,
|
||||
state_res_store=StateResolutionStore(
|
||||
self._store, self._state_deletion_store
|
||||
),
|
||||
state_res_store=StateResolutionStore(self._store),
|
||||
)
|
||||
)
|
||||
else:
|
||||
@@ -1183,9 +1179,7 @@ class FederationEventHandler:
|
||||
room_version,
|
||||
state_maps,
|
||||
event_map={event_id: event},
|
||||
state_res_store=StateResolutionStore(
|
||||
self._store, self._state_deletion_store
|
||||
),
|
||||
state_res_store=StateResolutionStore(self._store),
|
||||
)
|
||||
|
||||
except Exception as e:
|
||||
@@ -1880,9 +1874,7 @@ class FederationEventHandler:
|
||||
room_version,
|
||||
[local_state_id_map, claimed_auth_events_id_map],
|
||||
event_map=None,
|
||||
state_res_store=StateResolutionStore(
|
||||
self._store, self._state_deletion_store
|
||||
),
|
||||
state_res_store=StateResolutionStore(self._store),
|
||||
)
|
||||
)
|
||||
else:
|
||||
@@ -2022,9 +2014,7 @@ class FederationEventHandler:
|
||||
room_version,
|
||||
state_sets,
|
||||
event_map=None,
|
||||
state_res_store=StateResolutionStore(
|
||||
self._store, self._state_deletion_store
|
||||
),
|
||||
state_res_store=StateResolutionStore(self._store),
|
||||
)
|
||||
)
|
||||
else:
|
||||
|
||||
@@ -59,13 +59,11 @@ from synapse.types.state import StateFilter
|
||||
from synapse.util.async_helpers import Linearizer
|
||||
from synapse.util.caches.expiringcache import ExpiringCache
|
||||
from synapse.util.metrics import Measure, measure_func
|
||||
from synapse.util.stringutils import shortstr
|
||||
|
||||
if TYPE_CHECKING:
|
||||
from synapse.server import HomeServer
|
||||
from synapse.storage.controllers import StateStorageController
|
||||
from synapse.storage.databases.main import DataStore
|
||||
from synapse.storage.databases.state.deletion import StateDeletionDataStore
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
metrics_logger = logging.getLogger("synapse.state.metrics")
|
||||
@@ -196,8 +194,6 @@ class StateHandler:
|
||||
self._storage_controllers = hs.get_storage_controllers()
|
||||
self._events_shard_config = hs.config.worker.events_shard_config
|
||||
self._instance_name = hs.get_instance_name()
|
||||
self._state_store = hs.get_datastores().state
|
||||
self._state_deletion_store = hs.get_datastores().state_deletion
|
||||
|
||||
self._update_current_state_client = (
|
||||
ReplicationUpdateCurrentStateRestServlet.make_client(hs)
|
||||
@@ -359,28 +355,6 @@ class StateHandler:
|
||||
await_full_state=False,
|
||||
)
|
||||
|
||||
# Ensure we still have the state groups we're relying on, and bump
|
||||
# their usage time to avoid them being deleted from under us.
|
||||
if entry.state_group:
|
||||
missing_state_group = await self._state_deletion_store.check_state_groups_and_bump_deletion(
|
||||
{entry.state_group}
|
||||
)
|
||||
if missing_state_group:
|
||||
raise Exception(f"Missing state group: {entry.state_group}")
|
||||
elif entry.prev_group:
|
||||
# We only rely on the prev group when persisting the event if we
|
||||
# don't have an `entry.state_group`.
|
||||
missing_state_group = await self._state_deletion_store.check_state_groups_and_bump_deletion(
|
||||
{entry.prev_group}
|
||||
)
|
||||
|
||||
if missing_state_group:
|
||||
# If we're missing the prev group then we can just clear the
|
||||
# entries, and rely on `entry._state` (which must exist if
|
||||
# `entry.state_group` is None)
|
||||
entry.prev_group = None
|
||||
entry.delta_ids = None
|
||||
|
||||
state_group_before_event_prev_group = entry.prev_group
|
||||
deltas_to_state_group_before_event = entry.delta_ids
|
||||
state_ids_before_event = None
|
||||
@@ -501,10 +475,7 @@ class StateHandler:
|
||||
@trace
|
||||
@measure_func()
|
||||
async def resolve_state_groups_for_events(
|
||||
self,
|
||||
room_id: str,
|
||||
event_ids: StrCollection,
|
||||
await_full_state: bool = True,
|
||||
self, room_id: str, event_ids: StrCollection, await_full_state: bool = True
|
||||
) -> _StateCacheEntry:
|
||||
"""Given a list of event_ids this method fetches the state at each
|
||||
event, resolves conflicts between them and returns them.
|
||||
@@ -540,7 +511,6 @@ class StateHandler:
|
||||
) = await self._state_storage_controller.get_state_group_delta(
|
||||
state_group_id
|
||||
)
|
||||
|
||||
return _StateCacheEntry(
|
||||
state=None,
|
||||
state_group=state_group_id,
|
||||
@@ -561,9 +531,7 @@ class StateHandler:
|
||||
room_version,
|
||||
state_to_resolve,
|
||||
None,
|
||||
state_res_store=StateResolutionStore(
|
||||
self.store, self._state_deletion_store
|
||||
),
|
||||
state_res_store=StateResolutionStore(self.store),
|
||||
)
|
||||
return result
|
||||
|
||||
@@ -695,25 +663,7 @@ class StateResolutionHandler:
|
||||
async with self.resolve_linearizer.queue(group_names):
|
||||
cache = self._state_cache.get(group_names, None)
|
||||
if cache:
|
||||
# Check that the returned cache entry doesn't point to deleted
|
||||
# state groups.
|
||||
state_groups_to_check = set()
|
||||
if cache.state_group is not None:
|
||||
state_groups_to_check.add(cache.state_group)
|
||||
|
||||
if cache.prev_group is not None:
|
||||
state_groups_to_check.add(cache.prev_group)
|
||||
|
||||
missing_state_groups = await state_res_store.state_deletion_store.check_state_groups_and_bump_deletion(
|
||||
state_groups_to_check
|
||||
)
|
||||
|
||||
if not missing_state_groups:
|
||||
return cache
|
||||
else:
|
||||
# There are missing state groups, so let's remove the stale
|
||||
# entry and continue as if it was a cache miss.
|
||||
self._state_cache.pop(group_names, None)
|
||||
return cache
|
||||
|
||||
logger.info(
|
||||
"Resolving state for %s with groups %s",
|
||||
@@ -721,16 +671,6 @@ class StateResolutionHandler:
|
||||
list(group_names),
|
||||
)
|
||||
|
||||
# We double check that none of the state groups have been deleted.
|
||||
# They shouldn't be as all these state groups should be referenced.
|
||||
missing_state_groups = await state_res_store.state_deletion_store.check_state_groups_and_bump_deletion(
|
||||
group_names
|
||||
)
|
||||
if missing_state_groups:
|
||||
raise Exception(
|
||||
f"State groups have been deleted: {shortstr(missing_state_groups)}"
|
||||
)
|
||||
|
||||
state_groups_histogram.observe(len(state_groups_ids))
|
||||
|
||||
new_state = await self.resolve_events_with_store(
|
||||
@@ -944,8 +884,7 @@ class StateResolutionStore:
|
||||
in well defined way.
|
||||
"""
|
||||
|
||||
main_store: "DataStore"
|
||||
state_deletion_store: "StateDeletionDataStore"
|
||||
store: "DataStore"
|
||||
|
||||
def get_events(
|
||||
self, event_ids: StrCollection, allow_rejected: bool = False
|
||||
@@ -960,7 +899,7 @@ class StateResolutionStore:
|
||||
An awaitable which resolves to a dict from event_id to event.
|
||||
"""
|
||||
|
||||
return self.main_store.get_events(
|
||||
return self.store.get_events(
|
||||
event_ids,
|
||||
redact_behaviour=EventRedactBehaviour.as_is,
|
||||
get_prev_content=False,
|
||||
@@ -981,4 +920,4 @@ class StateResolutionStore:
|
||||
An awaitable that resolves to a set of event IDs.
|
||||
"""
|
||||
|
||||
return self.main_store.get_auth_chain_difference(room_id, state_sets)
|
||||
return self.store.get_auth_chain_difference(room_id, state_sets)
|
||||
|
||||
@@ -332,7 +332,6 @@ class EventsPersistenceStorageController:
|
||||
# store for now.
|
||||
self.main_store = stores.main
|
||||
self.state_store = stores.state
|
||||
self._state_deletion_store = stores.state_deletion
|
||||
|
||||
assert stores.persist_events
|
||||
self.persist_events_store = stores.persist_events
|
||||
@@ -550,9 +549,7 @@ class EventsPersistenceStorageController:
|
||||
room_version,
|
||||
state_maps_by_state_group,
|
||||
event_map=None,
|
||||
state_res_store=StateResolutionStore(
|
||||
self.main_store, self._state_deletion_store
|
||||
),
|
||||
state_res_store=StateResolutionStore(self.main_store),
|
||||
)
|
||||
|
||||
return await res.get_state(self._state_controller, StateFilter.all())
|
||||
@@ -638,20 +635,15 @@ class EventsPersistenceStorageController:
|
||||
room_id, [e for e, _ in chunk]
|
||||
)
|
||||
|
||||
# Stop the state groups from being deleted while we're persisting
|
||||
# them.
|
||||
async with self._state_deletion_store.persisting_state_group_references(
|
||||
events_and_contexts
|
||||
):
|
||||
await self.persist_events_store._persist_events_and_state_updates(
|
||||
room_id,
|
||||
chunk,
|
||||
state_delta_for_room=state_delta_for_room,
|
||||
new_forward_extremities=new_forward_extremities,
|
||||
use_negative_stream_ordering=backfilled,
|
||||
inhibit_local_membership_updates=backfilled,
|
||||
new_event_links=new_event_links,
|
||||
)
|
||||
await self.persist_events_store._persist_events_and_state_updates(
|
||||
room_id,
|
||||
chunk,
|
||||
state_delta_for_room=state_delta_for_room,
|
||||
new_forward_extremities=new_forward_extremities,
|
||||
use_negative_stream_ordering=backfilled,
|
||||
inhibit_local_membership_updates=backfilled,
|
||||
new_event_links=new_event_links,
|
||||
)
|
||||
|
||||
return replaced_events
|
||||
|
||||
@@ -973,9 +965,7 @@ class EventsPersistenceStorageController:
|
||||
room_version,
|
||||
state_groups,
|
||||
events_map,
|
||||
state_res_store=StateResolutionStore(
|
||||
self.main_store, self._state_deletion_store
|
||||
),
|
||||
state_res_store=StateResolutionStore(self.main_store),
|
||||
)
|
||||
|
||||
state_resolutions_during_persistence.inc()
|
||||
|
||||
@@ -21,10 +21,9 @@
|
||||
|
||||
import itertools
|
||||
import logging
|
||||
from typing import TYPE_CHECKING, Collection, Mapping, Set
|
||||
from typing import TYPE_CHECKING, Set
|
||||
|
||||
from synapse.logging.context import nested_logging_context
|
||||
from synapse.metrics.background_process_metrics import wrap_as_background_process
|
||||
from synapse.storage.databases import Databases
|
||||
|
||||
if TYPE_CHECKING:
|
||||
@@ -39,11 +38,6 @@ class PurgeEventsStorageController:
|
||||
def __init__(self, hs: "HomeServer", stores: Databases):
|
||||
self.stores = stores
|
||||
|
||||
if hs.config.worker.run_background_tasks:
|
||||
self._delete_state_loop_call = hs.get_clock().looping_call(
|
||||
self._delete_state_groups_loop, 60 * 1000
|
||||
)
|
||||
|
||||
async def purge_room(self, room_id: str) -> None:
|
||||
"""Deletes all record of a room"""
|
||||
|
||||
@@ -74,15 +68,11 @@ class PurgeEventsStorageController:
|
||||
logger.info("[purge] finding state groups that can be deleted")
|
||||
sg_to_delete = await self._find_unreferenced_groups(state_groups)
|
||||
|
||||
# Mark these state groups as pending deletion, they will actually
|
||||
# get deleted automatically later.
|
||||
await self.stores.state_deletion.mark_state_groups_as_pending_deletion(
|
||||
sg_to_delete
|
||||
await self.stores.state.purge_unreferenced_state_groups(
|
||||
room_id, sg_to_delete
|
||||
)
|
||||
|
||||
async def _find_unreferenced_groups(
|
||||
self, state_groups: Collection[int]
|
||||
) -> Set[int]:
|
||||
async def _find_unreferenced_groups(self, state_groups: Set[int]) -> Set[int]:
|
||||
"""Used when purging history to figure out which state groups can be
|
||||
deleted.
|
||||
|
||||
@@ -128,78 +118,6 @@ class PurgeEventsStorageController:
|
||||
next_to_search |= prevs
|
||||
state_groups_seen |= prevs
|
||||
|
||||
# We also check to see if anything referencing the state groups are
|
||||
# also unreferenced. This helps ensure that we delete unreferenced
|
||||
# state groups, if we don't then we will de-delta them when we
|
||||
# delete the other state groups leading to increased DB usage.
|
||||
next_edges = await self.stores.state.get_next_state_groups(current_search)
|
||||
nexts = set(next_edges.keys())
|
||||
nexts -= state_groups_seen
|
||||
next_to_search |= nexts
|
||||
state_groups_seen |= nexts
|
||||
|
||||
to_delete = state_groups_seen - referenced_groups
|
||||
|
||||
return to_delete
|
||||
|
||||
@wrap_as_background_process("_delete_state_groups_loop")
|
||||
async def _delete_state_groups_loop(self) -> None:
|
||||
"""Background task that deletes any state groups that may be pending
|
||||
deletion."""
|
||||
|
||||
while True:
|
||||
next_to_delete = await self.stores.state_deletion.get_next_state_group_collection_to_delete()
|
||||
if next_to_delete is None:
|
||||
break
|
||||
|
||||
(room_id, groups_to_sequences) = next_to_delete
|
||||
made_progress = await self._delete_state_groups(
|
||||
room_id, groups_to_sequences
|
||||
)
|
||||
|
||||
# If no progress was made in deleting the state groups, then we
|
||||
# break to allow a pause before trying again next time we get
|
||||
# called.
|
||||
if not made_progress:
|
||||
break
|
||||
|
||||
async def _delete_state_groups(
|
||||
self, room_id: str, groups_to_sequences: Mapping[int, int]
|
||||
) -> bool:
|
||||
"""Tries to delete the given state groups.
|
||||
|
||||
Returns:
|
||||
Whether we made progress in deleting the state groups (or marking
|
||||
them as referenced).
|
||||
"""
|
||||
|
||||
# We double check if any of the state groups have become referenced.
|
||||
# This shouldn't happen, as any usages should cause the state group to
|
||||
# be removed as pending deletion.
|
||||
referenced_state_groups = await self.stores.main.get_referenced_state_groups(
|
||||
groups_to_sequences
|
||||
)
|
||||
|
||||
if referenced_state_groups:
|
||||
# We mark any state groups that have become referenced as being
|
||||
# used.
|
||||
await self.stores.state_deletion.mark_state_groups_as_used(
|
||||
referenced_state_groups
|
||||
)
|
||||
|
||||
# Update list of state groups to remove referenced ones
|
||||
groups_to_sequences = {
|
||||
state_group: sequence_number
|
||||
for state_group, sequence_number in groups_to_sequences.items()
|
||||
if state_group not in referenced_state_groups
|
||||
}
|
||||
|
||||
if not groups_to_sequences:
|
||||
# We made progress here as long as we marked some state groups as
|
||||
# now referenced.
|
||||
return len(referenced_state_groups) > 0
|
||||
|
||||
return await self.stores.state.purge_unreferenced_state_groups(
|
||||
room_id,
|
||||
groups_to_sequences,
|
||||
)
|
||||
|
||||
@@ -26,7 +26,6 @@ from synapse.storage._base import SQLBaseStore
|
||||
from synapse.storage.database import DatabasePool, make_conn
|
||||
from synapse.storage.databases.main.events import PersistEventsStore
|
||||
from synapse.storage.databases.state import StateGroupDataStore
|
||||
from synapse.storage.databases.state.deletion import StateDeletionDataStore
|
||||
from synapse.storage.engines import create_engine
|
||||
from synapse.storage.prepare_database import prepare_database
|
||||
|
||||
@@ -50,14 +49,12 @@ class Databases(Generic[DataStoreT]):
|
||||
main
|
||||
state
|
||||
persist_events
|
||||
state_deletion
|
||||
"""
|
||||
|
||||
databases: List[DatabasePool]
|
||||
main: "DataStore" # FIXME: https://github.com/matrix-org/synapse/issues/11165: actually an instance of `main_store_class`
|
||||
state: StateGroupDataStore
|
||||
persist_events: Optional[PersistEventsStore]
|
||||
state_deletion: StateDeletionDataStore
|
||||
|
||||
def __init__(self, main_store_class: Type[DataStoreT], hs: "HomeServer"):
|
||||
# Note we pass in the main store class here as workers use a different main
|
||||
@@ -66,7 +63,6 @@ class Databases(Generic[DataStoreT]):
|
||||
self.databases = []
|
||||
main: Optional[DataStoreT] = None
|
||||
state: Optional[StateGroupDataStore] = None
|
||||
state_deletion: Optional[StateDeletionDataStore] = None
|
||||
persist_events: Optional[PersistEventsStore] = None
|
||||
|
||||
for database_config in hs.config.database.databases:
|
||||
@@ -118,8 +114,7 @@ class Databases(Generic[DataStoreT]):
|
||||
if state:
|
||||
raise Exception("'state' data store already configured")
|
||||
|
||||
state_deletion = StateDeletionDataStore(database, db_conn, hs)
|
||||
state = StateGroupDataStore(database, db_conn, hs, state_deletion)
|
||||
state = StateGroupDataStore(database, db_conn, hs)
|
||||
|
||||
db_conn.commit()
|
||||
|
||||
@@ -140,7 +135,7 @@ class Databases(Generic[DataStoreT]):
|
||||
if not main:
|
||||
raise Exception("No 'main' database configured")
|
||||
|
||||
if not state or not state_deletion:
|
||||
if not state:
|
||||
raise Exception("No 'state' database configured")
|
||||
|
||||
# We use local variables here to ensure that the databases do not have
|
||||
@@ -148,4 +143,3 @@ class Databases(Generic[DataStoreT]):
|
||||
self.main = main # type: ignore[assignment]
|
||||
self.state = state
|
||||
self.persist_events = persist_events
|
||||
self.state_deletion = state_deletion
|
||||
|
||||
@@ -1,537 +0,0 @@
|
||||
#
|
||||
# This file is licensed under the Affero General Public License (AGPL) version 3.
|
||||
#
|
||||
# Copyright (C) 2025 New Vector, Ltd
|
||||
#
|
||||
# This program is free software: you can redistribute it and/or modify
|
||||
# it under the terms of the GNU Affero General Public License as
|
||||
# published by the Free Software Foundation, either version 3 of the
|
||||
# License, or (at your option) any later version.
|
||||
#
|
||||
# See the GNU Affero General Public License for more details:
|
||||
# <https://www.gnu.org/licenses/agpl-3.0.html>.
|
||||
#
|
||||
|
||||
|
||||
import contextlib
|
||||
from typing import (
|
||||
TYPE_CHECKING,
|
||||
AbstractSet,
|
||||
AsyncIterator,
|
||||
Collection,
|
||||
Mapping,
|
||||
Optional,
|
||||
Set,
|
||||
Tuple,
|
||||
)
|
||||
|
||||
from synapse.events import EventBase
|
||||
from synapse.events.snapshot import EventContext
|
||||
from synapse.storage.database import (
|
||||
DatabasePool,
|
||||
LoggingDatabaseConnection,
|
||||
LoggingTransaction,
|
||||
make_in_list_sql_clause,
|
||||
)
|
||||
from synapse.storage.engines import PostgresEngine
|
||||
from synapse.util.stringutils import shortstr
|
||||
|
||||
if TYPE_CHECKING:
|
||||
from synapse.server import HomeServer
|
||||
|
||||
|
||||
class StateDeletionDataStore:
|
||||
"""Manages deletion of state groups in a safe manner.
|
||||
|
||||
Deleting state groups is challenging as before we actually delete them we
|
||||
need to ensure that there are no in-flight events that refer to the state
|
||||
groups that we want to delete.
|
||||
|
||||
To handle this, we take two approaches. First, before we persist any event
|
||||
we ensure that the state group still exists and mark in the
|
||||
`state_groups_persisting` table that the state group is about to be used.
|
||||
(Note that we have to have the extra table here as state groups and events
|
||||
can be in different databases, and thus we can't check for the existence of
|
||||
state groups in the persist event transaction). Once the event has been
|
||||
persisted, we can remove the row from `state_groups_persisting`. So long as
|
||||
we check that table before deleting state groups, we can ensure that we
|
||||
never persist events that reference deleted state groups, maintaining
|
||||
database integrity.
|
||||
|
||||
However, we want to avoid throwing exceptions so deep in the process of
|
||||
persisting events. So instead of deleting state groups immediately, we mark
|
||||
them as pending/proposed for deletion and wait for a certain amount of time
|
||||
before performing the deletion. When we come to handle new events that
|
||||
reference state groups, we check if they are pending deletion and bump the
|
||||
time for when they'll be deleted (to give a chance for the event to be
|
||||
persisted, or not).
|
||||
|
||||
When deleting, we need to check that state groups remain unreferenced. There
|
||||
is a race here where we a) fetch state groups that are ready for deletion,
|
||||
b) check they're unreferenced, c) the state group becomes referenced but
|
||||
then gets marked as pending deletion again, d) during the deletion
|
||||
transaction we recheck `state_groups_pending_deletion` table again and see
|
||||
that it exists and so continue with the deletion. To prevent this from
|
||||
happening we add a `sequence_number` column to
|
||||
`state_groups_pending_deletion`, and during deletion we ensure that for a
|
||||
state group we're about to delete that the sequence number doesn't change
|
||||
between steps (a) and (d). So long as we always bump the sequence number
|
||||
whenever an event may become used the race can never happen.
|
||||
"""
|
||||
|
||||
# How long to wait before we delete state groups. This should be long enough
|
||||
# for any in-flight events to be persisted. If events take longer to persist
|
||||
# and any of the state groups they reference have been deleted, then the
|
||||
# event will fail to persist (as well as any event in the same batch).
|
||||
DELAY_BEFORE_DELETION_MS = 10 * 60 * 1000
|
||||
|
||||
def __init__(
|
||||
self,
|
||||
database: DatabasePool,
|
||||
db_conn: LoggingDatabaseConnection,
|
||||
hs: "HomeServer",
|
||||
):
|
||||
self._clock = hs.get_clock()
|
||||
self.db_pool = database
|
||||
self._instance_name = hs.get_instance_name()
|
||||
|
||||
with db_conn.cursor(txn_name="_clear_existing_persising") as txn:
|
||||
self._clear_existing_persising(txn)
|
||||
|
||||
def _clear_existing_persising(self, txn: LoggingTransaction) -> None:
|
||||
"""On startup we clear any entries in `state_groups_persisting` that
|
||||
match our instance name, in case of a previous unclean shutdown"""
|
||||
|
||||
self.db_pool.simple_delete_txn(
|
||||
txn,
|
||||
table="state_groups_persisting",
|
||||
keyvalues={"instance_name": self._instance_name},
|
||||
)
|
||||
|
||||
async def check_state_groups_and_bump_deletion(
|
||||
self, state_groups: AbstractSet[int]
|
||||
) -> Collection[int]:
|
||||
"""Checks to make sure that the state groups haven't been deleted, and
|
||||
if they're pending deletion we delay it (allowing time for any event
|
||||
that will use them to finish persisting).
|
||||
|
||||
Returns:
|
||||
The state groups that are missing, if any.
|
||||
"""
|
||||
|
||||
return await self.db_pool.runInteraction(
|
||||
"check_state_groups_and_bump_deletion",
|
||||
self._check_state_groups_and_bump_deletion_txn,
|
||||
state_groups,
|
||||
# We don't need to lock if we're just doing a quick check, as the
|
||||
# lock doesn't prevent any races here.
|
||||
lock=False,
|
||||
)
|
||||
|
||||
def _check_state_groups_and_bump_deletion_txn(
|
||||
self, txn: LoggingTransaction, state_groups: AbstractSet[int], lock: bool = True
|
||||
) -> Collection[int]:
|
||||
"""Checks to make sure that the state groups haven't been deleted, and
|
||||
if they're pending deletion we delay it (allowing time for any event
|
||||
that will use them to finish persisting).
|
||||
|
||||
The `lock` flag sets if we should lock the `state_group` rows we're
|
||||
checking, which we should do when storing new groups.
|
||||
|
||||
Returns:
|
||||
The state groups that are missing, if any.
|
||||
"""
|
||||
|
||||
existing_state_groups = self._get_existing_groups_with_lock(
|
||||
txn, state_groups, lock=lock
|
||||
)
|
||||
|
||||
self._bump_deletion_txn(txn, existing_state_groups)
|
||||
|
||||
missing_state_groups = state_groups - existing_state_groups
|
||||
if missing_state_groups:
|
||||
return missing_state_groups
|
||||
|
||||
return ()
|
||||
|
||||
def _bump_deletion_txn(
|
||||
self, txn: LoggingTransaction, state_groups: Collection[int]
|
||||
) -> None:
|
||||
"""Update any pending deletions of the state group that they may now be
|
||||
referenced."""
|
||||
|
||||
if not state_groups:
|
||||
return
|
||||
|
||||
now = self._clock.time_msec()
|
||||
if isinstance(self.db_pool.engine, PostgresEngine):
|
||||
clause, args = make_in_list_sql_clause(
|
||||
self.db_pool.engine, "state_group", state_groups
|
||||
)
|
||||
sql = f"""
|
||||
UPDATE state_groups_pending_deletion
|
||||
SET sequence_number = DEFAULT, insertion_ts = ?
|
||||
WHERE {clause}
|
||||
"""
|
||||
args.insert(0, now)
|
||||
txn.execute(sql, args)
|
||||
else:
|
||||
rows = self.db_pool.simple_select_many_txn(
|
||||
txn,
|
||||
table="state_groups_pending_deletion",
|
||||
column="state_group",
|
||||
iterable=state_groups,
|
||||
keyvalues={},
|
||||
retcols=("state_group",),
|
||||
)
|
||||
if not rows:
|
||||
return
|
||||
|
||||
state_groups_to_update = [state_group for (state_group,) in rows]
|
||||
|
||||
self.db_pool.simple_delete_many_txn(
|
||||
txn,
|
||||
table="state_groups_pending_deletion",
|
||||
column="state_group",
|
||||
values=state_groups_to_update,
|
||||
keyvalues={},
|
||||
)
|
||||
self.db_pool.simple_insert_many_txn(
|
||||
txn,
|
||||
table="state_groups_pending_deletion",
|
||||
keys=("state_group", "insertion_ts"),
|
||||
values=[(state_group, now) for state_group in state_groups_to_update],
|
||||
)
|
||||
|
||||
def _get_existing_groups_with_lock(
|
||||
self, txn: LoggingTransaction, state_groups: Collection[int], lock: bool = True
|
||||
) -> AbstractSet[int]:
|
||||
"""Return which of the given state groups are in the database, and locks
|
||||
those rows with `KEY SHARE` to ensure they don't get concurrently
|
||||
deleted (if `lock` is true)."""
|
||||
clause, args = make_in_list_sql_clause(self.db_pool.engine, "id", state_groups)
|
||||
|
||||
sql = f"""
|
||||
SELECT id FROM state_groups
|
||||
WHERE {clause}
|
||||
"""
|
||||
if lock and isinstance(self.db_pool.engine, PostgresEngine):
|
||||
# On postgres we add a row level lock to the rows to ensure that we
|
||||
# conflict with any concurrent DELETEs. `FOR KEY SHARE` lock will
|
||||
# not conflict with other read
|
||||
sql += """
|
||||
FOR KEY SHARE
|
||||
"""
|
||||
|
||||
txn.execute(sql, args)
|
||||
return {state_group for (state_group,) in txn}
|
||||
|
||||
@contextlib.asynccontextmanager
|
||||
async def persisting_state_group_references(
|
||||
self, event_and_contexts: Collection[Tuple[EventBase, EventContext]]
|
||||
) -> AsyncIterator[None]:
|
||||
"""Wraps the persistence of the given events and contexts, ensuring that
|
||||
any state groups referenced still exist and that they don't get deleted
|
||||
during this."""
|
||||
|
||||
referenced_state_groups: Set[int] = set()
|
||||
for event, ctx in event_and_contexts:
|
||||
if ctx.rejected or event.internal_metadata.is_outlier():
|
||||
continue
|
||||
|
||||
assert ctx.state_group is not None
|
||||
|
||||
referenced_state_groups.add(ctx.state_group)
|
||||
|
||||
if ctx.state_group_before_event:
|
||||
referenced_state_groups.add(ctx.state_group_before_event)
|
||||
|
||||
if not referenced_state_groups:
|
||||
# We don't reference any state groups, so nothing to do
|
||||
yield
|
||||
return
|
||||
|
||||
await self.db_pool.runInteraction(
|
||||
"mark_state_groups_as_persisting",
|
||||
self._mark_state_groups_as_persisting_txn,
|
||||
referenced_state_groups,
|
||||
)
|
||||
|
||||
error = True
|
||||
try:
|
||||
yield None
|
||||
error = False
|
||||
finally:
|
||||
await self.db_pool.runInteraction(
|
||||
"finish_persisting",
|
||||
self._finish_persisting_txn,
|
||||
referenced_state_groups,
|
||||
error=error,
|
||||
)
|
||||
|
||||
def _mark_state_groups_as_persisting_txn(
|
||||
self, txn: LoggingTransaction, state_groups: Set[int]
|
||||
) -> None:
|
||||
"""Marks the given state groups as being persisted."""
|
||||
|
||||
existing_state_groups = self._get_existing_groups_with_lock(txn, state_groups)
|
||||
missing_state_groups = state_groups - existing_state_groups
|
||||
if missing_state_groups:
|
||||
raise Exception(
|
||||
f"state groups have been deleted: {shortstr(missing_state_groups)}"
|
||||
)
|
||||
|
||||
self.db_pool.simple_insert_many_txn(
|
||||
txn,
|
||||
table="state_groups_persisting",
|
||||
keys=("state_group", "instance_name"),
|
||||
values=[(state_group, self._instance_name) for state_group in state_groups],
|
||||
)
|
||||
|
||||
def _finish_persisting_txn(
|
||||
self, txn: LoggingTransaction, state_groups: Collection[int], error: bool
|
||||
) -> None:
|
||||
"""Mark the state groups as having finished persistence.
|
||||
|
||||
If `error` is true then we assume the state groups were not persisted,
|
||||
and so we do not clear them from the pending deletion table.
|
||||
"""
|
||||
self.db_pool.simple_delete_many_txn(
|
||||
txn,
|
||||
table="state_groups_persisting",
|
||||
column="state_group",
|
||||
values=state_groups,
|
||||
keyvalues={"instance_name": self._instance_name},
|
||||
)
|
||||
|
||||
if error:
|
||||
# The state groups may or may not have been persisted, so we need to
|
||||
# bump the deletion to ensure we recheck if they have become
|
||||
# referenced.
|
||||
self._bump_deletion_txn(txn, state_groups)
|
||||
return
|
||||
|
||||
self.db_pool.simple_delete_many_batch_txn(
|
||||
txn,
|
||||
table="state_groups_pending_deletion",
|
||||
keys=("state_group",),
|
||||
values=[(state_group,) for state_group in state_groups],
|
||||
)
|
||||
|
||||
async def mark_state_groups_as_pending_deletion(
|
||||
self, state_groups: Collection[int]
|
||||
) -> None:
|
||||
"""Mark the given state groups as pending deletion"""
|
||||
|
||||
now = self._clock.time_msec()
|
||||
|
||||
await self.db_pool.simple_upsert_many(
|
||||
table="state_groups_pending_deletion",
|
||||
key_names=("state_group",),
|
||||
key_values=[(state_group,) for state_group in state_groups],
|
||||
value_names=("insertion_ts",),
|
||||
value_values=[(now,) for _ in state_groups],
|
||||
desc="mark_state_groups_as_pending_deletion",
|
||||
)
|
||||
|
||||
async def mark_state_groups_as_used(self, state_groups: Collection[int]) -> None:
|
||||
"""Mark the given state groups as now being referenced"""
|
||||
|
||||
await self.db_pool.simple_delete_many(
|
||||
table="state_groups_pending_deletion",
|
||||
column="state_group",
|
||||
iterable=state_groups,
|
||||
keyvalues={},
|
||||
desc="mark_state_groups_as_used",
|
||||
)
|
||||
|
||||
async def get_pending_deletions(
|
||||
self, state_groups: Collection[int]
|
||||
) -> Mapping[int, int]:
|
||||
"""Get which state groups are pending deletion.
|
||||
|
||||
Returns:
|
||||
a mapping from state groups that are pending deletion to their
|
||||
sequence number
|
||||
"""
|
||||
|
||||
rows = await self.db_pool.simple_select_many_batch(
|
||||
table="state_groups_pending_deletion",
|
||||
column="state_group",
|
||||
iterable=state_groups,
|
||||
retcols=("state_group", "sequence_number"),
|
||||
keyvalues={},
|
||||
desc="get_pending_deletions",
|
||||
)
|
||||
|
||||
return dict(rows)
|
||||
|
||||
def get_state_groups_ready_for_potential_deletion_txn(
|
||||
self,
|
||||
txn: LoggingTransaction,
|
||||
state_groups_to_sequence_numbers: Mapping[int, int],
|
||||
) -> Collection[int]:
|
||||
"""Given a set of state groups, return which state groups can
|
||||
potentially be deleted.
|
||||
|
||||
The state groups must have been checked to see if they remain
|
||||
unreferenced before calling this function.
|
||||
|
||||
Note: This must be called within the same transaction that the state
|
||||
groups are deleted.
|
||||
|
||||
Args:
|
||||
state_groups_to_sequence_numbers: The state groups, and the sequence
|
||||
numbers from before the state groups were checked to see if they
|
||||
were unreferenced.
|
||||
|
||||
Returns:
|
||||
The subset of state groups that can safely be deleted
|
||||
|
||||
"""
|
||||
|
||||
if not state_groups_to_sequence_numbers:
|
||||
return state_groups_to_sequence_numbers
|
||||
|
||||
if isinstance(self.db_pool.engine, PostgresEngine):
|
||||
# On postgres we want to lock the rows FOR UPDATE as early as
|
||||
# possible to help conflicts.
|
||||
clause, args = make_in_list_sql_clause(
|
||||
self.db_pool.engine, "id", state_groups_to_sequence_numbers
|
||||
)
|
||||
sql = f"""
|
||||
SELECT id FROM state_groups
|
||||
WHERE {clause}
|
||||
FOR UPDATE
|
||||
"""
|
||||
txn.execute(sql, args)
|
||||
|
||||
# Check the deletion status in the DB of the given state groups
|
||||
clause, args = make_in_list_sql_clause(
|
||||
self.db_pool.engine,
|
||||
column="state_group",
|
||||
iterable=state_groups_to_sequence_numbers,
|
||||
)
|
||||
|
||||
sql = f"""
|
||||
SELECT state_group, insertion_ts, sequence_number FROM (
|
||||
SELECT state_group, insertion_ts, sequence_number FROM state_groups_pending_deletion
|
||||
UNION
|
||||
SELECT state_group, null, null FROM state_groups_persisting
|
||||
) AS s
|
||||
WHERE {clause}
|
||||
"""
|
||||
|
||||
txn.execute(sql, args)
|
||||
|
||||
# The above query will return potentially two rows per state group (one
|
||||
# for each table), so we track which state groups have enough time
|
||||
# elapsed and which are not ready to be persisted.
|
||||
ready_to_be_deleted = set()
|
||||
not_ready_to_be_deleted = set()
|
||||
|
||||
now = self._clock.time_msec()
|
||||
for state_group, insertion_ts, sequence_number in txn:
|
||||
if insertion_ts is None:
|
||||
# A null insertion_ts means that we are currently persisting
|
||||
# events that reference the state group, so we don't delete
|
||||
# them.
|
||||
not_ready_to_be_deleted.add(state_group)
|
||||
continue
|
||||
|
||||
# We know this can't be None if insertion_ts is not None
|
||||
assert sequence_number is not None
|
||||
|
||||
# Check if the sequence number has changed, if it has then it
|
||||
# indicates that the state group may have become referenced since we
|
||||
# checked.
|
||||
if state_groups_to_sequence_numbers[state_group] != sequence_number:
|
||||
not_ready_to_be_deleted.add(state_group)
|
||||
continue
|
||||
|
||||
if now - insertion_ts < self.DELAY_BEFORE_DELETION_MS:
|
||||
# Not enough time has elapsed to allow us to delete.
|
||||
not_ready_to_be_deleted.add(state_group)
|
||||
continue
|
||||
|
||||
ready_to_be_deleted.add(state_group)
|
||||
|
||||
can_be_deleted = ready_to_be_deleted - not_ready_to_be_deleted
|
||||
if not_ready_to_be_deleted:
|
||||
# If there are any state groups that aren't ready to be deleted,
|
||||
# then we also need to remove any state groups that are referenced
|
||||
# by them.
|
||||
clause, args = make_in_list_sql_clause(
|
||||
self.db_pool.engine,
|
||||
column="state_group",
|
||||
iterable=state_groups_to_sequence_numbers,
|
||||
)
|
||||
sql = f"""
|
||||
WITH RECURSIVE ancestors(state_group) AS (
|
||||
SELECT DISTINCT prev_state_group
|
||||
FROM state_group_edges WHERE {clause}
|
||||
UNION
|
||||
SELECT prev_state_group
|
||||
FROM state_group_edges
|
||||
INNER JOIN ancestors USING (state_group)
|
||||
)
|
||||
SELECT state_group FROM ancestors
|
||||
"""
|
||||
txn.execute(sql, args)
|
||||
|
||||
can_be_deleted.difference_update(state_group for (state_group,) in txn)
|
||||
|
||||
return can_be_deleted
|
||||
|
||||
async def get_next_state_group_collection_to_delete(
|
||||
self,
|
||||
) -> Optional[Tuple[str, Mapping[int, int]]]:
|
||||
"""Get the next set of state groups to try and delete
|
||||
|
||||
Returns:
|
||||
2-tuple of room_id and mapping of state groups to sequence number.
|
||||
"""
|
||||
return await self.db_pool.runInteraction(
|
||||
"get_next_state_group_collection_to_delete",
|
||||
self._get_next_state_group_collection_to_delete_txn,
|
||||
)
|
||||
|
||||
def _get_next_state_group_collection_to_delete_txn(
|
||||
self,
|
||||
txn: LoggingTransaction,
|
||||
) -> Optional[Tuple[str, Mapping[int, int]]]:
|
||||
"""Implementation of `get_next_state_group_collection_to_delete`"""
|
||||
|
||||
# We want to return chunks of state groups that were marked for deletion
|
||||
# at the same time (this isn't necessary, just more efficient). We do
|
||||
# this by looking for the oldest insertion_ts, and then pulling out all
|
||||
# rows that have the same insertion_ts (and room ID).
|
||||
now = self._clock.time_msec()
|
||||
|
||||
sql = """
|
||||
SELECT room_id, insertion_ts
|
||||
FROM state_groups_pending_deletion AS sd
|
||||
INNER JOIN state_groups AS sg ON (id = sd.state_group)
|
||||
LEFT JOIN state_groups_persisting AS sp USING (state_group)
|
||||
WHERE insertion_ts < ? AND sp.state_group IS NULL
|
||||
ORDER BY insertion_ts
|
||||
LIMIT 1
|
||||
"""
|
||||
txn.execute(sql, (now - self.DELAY_BEFORE_DELETION_MS,))
|
||||
row = txn.fetchone()
|
||||
if not row:
|
||||
return None
|
||||
|
||||
(room_id, insertion_ts) = row
|
||||
|
||||
sql = """
|
||||
SELECT state_group, sequence_number
|
||||
FROM state_groups_pending_deletion AS sd
|
||||
INNER JOIN state_groups AS sg ON (id = sd.state_group)
|
||||
LEFT JOIN state_groups_persisting AS sp USING (state_group)
|
||||
WHERE room_id = ? AND insertion_ts = ? AND sp.state_group IS NULL
|
||||
ORDER BY insertion_ts
|
||||
"""
|
||||
txn.execute(sql, (room_id, insertion_ts))
|
||||
|
||||
return room_id, dict(txn)
|
||||
@@ -22,10 +22,10 @@
|
||||
import logging
|
||||
from typing import (
|
||||
TYPE_CHECKING,
|
||||
Collection,
|
||||
Dict,
|
||||
Iterable,
|
||||
List,
|
||||
Mapping,
|
||||
Optional,
|
||||
Set,
|
||||
Tuple,
|
||||
@@ -36,10 +36,7 @@ import attr
|
||||
|
||||
from synapse.api.constants import EventTypes
|
||||
from synapse.events import EventBase
|
||||
from synapse.events.snapshot import (
|
||||
UnpersistedEventContext,
|
||||
UnpersistedEventContextBase,
|
||||
)
|
||||
from synapse.events.snapshot import UnpersistedEventContext, UnpersistedEventContextBase
|
||||
from synapse.logging.opentracing import tag_args, trace
|
||||
from synapse.storage._base import SQLBaseStore
|
||||
from synapse.storage.database import (
|
||||
@@ -58,7 +55,6 @@ from synapse.util.cancellation import cancellable
|
||||
|
||||
if TYPE_CHECKING:
|
||||
from synapse.server import HomeServer
|
||||
from synapse.storage.databases.state.deletion import StateDeletionDataStore
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
@@ -87,10 +83,8 @@ class StateGroupDataStore(StateBackgroundUpdateStore, SQLBaseStore):
|
||||
database: DatabasePool,
|
||||
db_conn: LoggingDatabaseConnection,
|
||||
hs: "HomeServer",
|
||||
state_deletion_store: "StateDeletionDataStore",
|
||||
):
|
||||
super().__init__(database, db_conn, hs)
|
||||
self._state_deletion_store = state_deletion_store
|
||||
|
||||
# Originally the state store used a single DictionaryCache to cache the
|
||||
# event IDs for the state types in a given state group to avoid hammering
|
||||
@@ -473,15 +467,14 @@ class StateGroupDataStore(StateBackgroundUpdateStore, SQLBaseStore):
|
||||
Returns:
|
||||
A list of state groups
|
||||
"""
|
||||
|
||||
# We need to check that the prev group isn't about to be deleted
|
||||
is_missing = (
|
||||
self._state_deletion_store._check_state_groups_and_bump_deletion_txn(
|
||||
txn,
|
||||
{prev_group},
|
||||
)
|
||||
is_in_db = self.db_pool.simple_select_one_onecol_txn(
|
||||
txn,
|
||||
table="state_groups",
|
||||
keyvalues={"id": prev_group},
|
||||
retcol="id",
|
||||
allow_none=True,
|
||||
)
|
||||
if is_missing:
|
||||
if not is_in_db:
|
||||
raise Exception(
|
||||
"Trying to persist state with unpersisted prev_group: %r"
|
||||
% (prev_group,)
|
||||
@@ -553,7 +546,6 @@ class StateGroupDataStore(StateBackgroundUpdateStore, SQLBaseStore):
|
||||
for key, state_id in context.state_delta_due_to_event.items()
|
||||
],
|
||||
)
|
||||
|
||||
return events_and_context
|
||||
|
||||
return await self.db_pool.runInteraction(
|
||||
@@ -609,15 +601,14 @@ class StateGroupDataStore(StateBackgroundUpdateStore, SQLBaseStore):
|
||||
The state group if successfully created, or None if the state
|
||||
needs to be persisted as a full state.
|
||||
"""
|
||||
|
||||
# We need to check that the prev group isn't about to be deleted
|
||||
is_missing = (
|
||||
self._state_deletion_store._check_state_groups_and_bump_deletion_txn(
|
||||
txn,
|
||||
{prev_group},
|
||||
)
|
||||
is_in_db = self.db_pool.simple_select_one_onecol_txn(
|
||||
txn,
|
||||
table="state_groups",
|
||||
keyvalues={"id": prev_group},
|
||||
retcol="id",
|
||||
allow_none=True,
|
||||
)
|
||||
if is_missing:
|
||||
if not is_in_db:
|
||||
raise Exception(
|
||||
"Trying to persist state with unpersisted prev_group: %r"
|
||||
% (prev_group,)
|
||||
@@ -735,10 +726,8 @@ class StateGroupDataStore(StateBackgroundUpdateStore, SQLBaseStore):
|
||||
)
|
||||
|
||||
async def purge_unreferenced_state_groups(
|
||||
self,
|
||||
room_id: str,
|
||||
state_groups_to_sequence_numbers: Mapping[int, int],
|
||||
) -> bool:
|
||||
self, room_id: str, state_groups_to_delete: Collection[int]
|
||||
) -> None:
|
||||
"""Deletes no longer referenced state groups and de-deltas any state
|
||||
groups that reference them.
|
||||
|
||||
@@ -746,31 +735,21 @@ class StateGroupDataStore(StateBackgroundUpdateStore, SQLBaseStore):
|
||||
room_id: The room the state groups belong to (must all be in the
|
||||
same room).
|
||||
state_groups_to_delete: Set of all state groups to delete.
|
||||
|
||||
Returns:
|
||||
Whether any state groups were actually deleted.
|
||||
"""
|
||||
|
||||
return await self.db_pool.runInteraction(
|
||||
await self.db_pool.runInteraction(
|
||||
"purge_unreferenced_state_groups",
|
||||
self._purge_unreferenced_state_groups,
|
||||
room_id,
|
||||
state_groups_to_sequence_numbers,
|
||||
state_groups_to_delete,
|
||||
)
|
||||
|
||||
def _purge_unreferenced_state_groups(
|
||||
self,
|
||||
txn: LoggingTransaction,
|
||||
room_id: str,
|
||||
state_groups_to_sequence_numbers: Mapping[int, int],
|
||||
) -> bool:
|
||||
state_groups_to_delete = self._state_deletion_store.get_state_groups_ready_for_potential_deletion_txn(
|
||||
txn, state_groups_to_sequence_numbers
|
||||
)
|
||||
|
||||
if not state_groups_to_delete:
|
||||
return False
|
||||
|
||||
state_groups_to_delete: Collection[int],
|
||||
) -> None:
|
||||
logger.info(
|
||||
"[purge] found %i state groups to delete", len(state_groups_to_delete)
|
||||
)
|
||||
@@ -833,8 +812,6 @@ class StateGroupDataStore(StateBackgroundUpdateStore, SQLBaseStore):
|
||||
[(sg,) for sg in state_groups_to_delete],
|
||||
)
|
||||
|
||||
return True
|
||||
|
||||
@trace
|
||||
@tag_args
|
||||
async def get_previous_state_groups(
|
||||
@@ -849,35 +826,6 @@ class StateGroupDataStore(StateBackgroundUpdateStore, SQLBaseStore):
|
||||
A mapping from state group to previous state group.
|
||||
"""
|
||||
|
||||
rows = cast(
|
||||
List[Tuple[int, int]],
|
||||
await self.db_pool.simple_select_many_batch(
|
||||
table="state_group_edges",
|
||||
column="state_group",
|
||||
iterable=state_groups,
|
||||
keyvalues={},
|
||||
retcols=("state_group", "prev_state_group"),
|
||||
desc="get_previous_state_groups",
|
||||
),
|
||||
)
|
||||
|
||||
return dict(rows)
|
||||
|
||||
@trace
|
||||
@tag_args
|
||||
async def get_next_state_groups(
|
||||
self, state_groups: Iterable[int]
|
||||
) -> Dict[int, int]:
|
||||
"""Fetch the groups that have the given state groups as their previous
|
||||
state groups.
|
||||
|
||||
Args:
|
||||
state_groups
|
||||
|
||||
Returns:
|
||||
A mapping from state group to previous state group.
|
||||
"""
|
||||
|
||||
rows = cast(
|
||||
List[Tuple[int, int]],
|
||||
await self.db_pool.simple_select_many_batch(
|
||||
@@ -886,7 +834,7 @@ class StateGroupDataStore(StateBackgroundUpdateStore, SQLBaseStore):
|
||||
iterable=state_groups,
|
||||
keyvalues={},
|
||||
retcols=("state_group", "prev_state_group"),
|
||||
desc="get_next_state_groups",
|
||||
desc="get_previous_state_groups",
|
||||
),
|
||||
)
|
||||
|
||||
|
||||
@@ -19,7 +19,7 @@
|
||||
#
|
||||
#
|
||||
|
||||
SCHEMA_VERSION = 89 # remember to update the list below when updating
|
||||
SCHEMA_VERSION = 88 # remember to update the list below when updating
|
||||
"""Represents the expectations made by the codebase about the database schema
|
||||
|
||||
This should be incremented whenever the codebase changes its requirements on the
|
||||
@@ -155,9 +155,6 @@ Changes in SCHEMA_VERSION = 88
|
||||
be posted in response to a resettable timeout or an on-demand action.
|
||||
- Add background update to fix data integrity issue in the
|
||||
`sliding_sync_membership_snapshots` -> `forgotten` column
|
||||
|
||||
Changes in SCHEMA_VERSION = 89
|
||||
- Add `state_groups_pending_deletion` and `state_groups_persisting` tables.
|
||||
"""
|
||||
|
||||
|
||||
|
||||
@@ -1,39 +0,0 @@
|
||||
--
|
||||
-- This file is licensed under the Affero General Public License (AGPL) version 3.
|
||||
--
|
||||
-- Copyright (C) 2025 New Vector, Ltd
|
||||
--
|
||||
-- This program is free software: you can redistribute it and/or modify
|
||||
-- it under the terms of the GNU Affero General Public License as
|
||||
-- published by the Free Software Foundation, either version 3 of the
|
||||
-- License, or (at your option) any later version.
|
||||
--
|
||||
-- See the GNU Affero General Public License for more details:
|
||||
-- <https://www.gnu.org/licenses/agpl-3.0.html>.
|
||||
|
||||
-- See the `StateDeletionDataStore` for details of these tables.
|
||||
|
||||
-- We add state groups to this table when we want to later delete them. The
|
||||
-- `insertion_ts` column indicates when the state group was proposed for
|
||||
-- deletion (rather than when it should be deleted).
|
||||
CREATE TABLE IF NOT EXISTS state_groups_pending_deletion (
|
||||
sequence_number $%AUTO_INCREMENT_PRIMARY_KEY%$,
|
||||
state_group BIGINT NOT NULL,
|
||||
insertion_ts BIGINT NOT NULL
|
||||
);
|
||||
|
||||
CREATE UNIQUE INDEX state_groups_pending_deletion_state_group ON state_groups_pending_deletion(state_group);
|
||||
CREATE INDEX state_groups_pending_deletion_insertion_ts ON state_groups_pending_deletion(insertion_ts);
|
||||
|
||||
|
||||
-- Holds the state groups the worker is currently persisting.
|
||||
--
|
||||
-- The `sequence_number` column of the `state_groups_pending_deletion` table
|
||||
-- *must* be updated whenever a state group may have become referenced.
|
||||
CREATE TABLE IF NOT EXISTS state_groups_persisting (
|
||||
state_group BIGINT NOT NULL,
|
||||
instance_name TEXT NOT NULL,
|
||||
PRIMARY KEY (state_group, instance_name)
|
||||
);
|
||||
|
||||
CREATE INDEX state_groups_persisting_instance_name ON state_groups_persisting(instance_name);
|
||||
@@ -807,7 +807,6 @@ class FederationEventHandlerTests(unittest.FederatingHomeserverTestCase):
|
||||
|
||||
OTHER_USER = f"@user:{self.OTHER_SERVER_NAME}"
|
||||
main_store = self.hs.get_datastores().main
|
||||
state_deletion_store = self.hs.get_datastores().state_deletion
|
||||
|
||||
# Create the room.
|
||||
kermit_user_id = self.register_user("kermit", "test")
|
||||
@@ -959,9 +958,7 @@ class FederationEventHandlerTests(unittest.FederatingHomeserverTestCase):
|
||||
bert_member_event.event_id: bert_member_event,
|
||||
rejected_kick_event.event_id: rejected_kick_event,
|
||||
},
|
||||
state_res_store=StateResolutionStore(
|
||||
main_store, state_deletion_store
|
||||
),
|
||||
state_res_store=StateResolutionStore(main_store),
|
||||
)
|
||||
),
|
||||
[bert_member_event.event_id, rejected_kick_event.event_id],
|
||||
@@ -1006,9 +1003,7 @@ class FederationEventHandlerTests(unittest.FederatingHomeserverTestCase):
|
||||
rejected_power_levels_event.event_id,
|
||||
],
|
||||
event_map={},
|
||||
state_res_store=StateResolutionStore(
|
||||
main_store, state_deletion_store
|
||||
),
|
||||
state_res_store=StateResolutionStore(main_store),
|
||||
full_conflicted_set=set(),
|
||||
)
|
||||
),
|
||||
|
||||
@@ -742,7 +742,7 @@ class RoomsCreateTestCase(RoomBase):
|
||||
self.assertEqual(HTTPStatus.OK, channel.code, channel.result)
|
||||
self.assertTrue("room_id" in channel.json_body)
|
||||
assert channel.resource_usage is not None
|
||||
self.assertEqual(35, channel.resource_usage.db_txn_count)
|
||||
self.assertEqual(34, channel.resource_usage.db_txn_count)
|
||||
|
||||
def test_post_room_initial_state(self) -> None:
|
||||
# POST with initial_state config key, expect new room id
|
||||
@@ -755,7 +755,7 @@ class RoomsCreateTestCase(RoomBase):
|
||||
self.assertEqual(HTTPStatus.OK, channel.code, channel.result)
|
||||
self.assertTrue("room_id" in channel.json_body)
|
||||
assert channel.resource_usage is not None
|
||||
self.assertEqual(37, channel.resource_usage.db_txn_count)
|
||||
self.assertEqual(36, channel.resource_usage.db_txn_count)
|
||||
|
||||
def test_post_room_visibility_key(self) -> None:
|
||||
# POST with visibility config key, expect new room id
|
||||
|
||||
@@ -548,7 +548,7 @@ class RestHelper:
|
||||
room_id: str,
|
||||
event_type: str,
|
||||
body: Dict[str, Any],
|
||||
tok: Optional[str] = None,
|
||||
tok: Optional[str],
|
||||
expect_code: int = HTTPStatus.OK,
|
||||
state_key: str = "",
|
||||
) -> JsonDict:
|
||||
|
||||
@@ -23,7 +23,6 @@ from twisted.test.proto_helpers import MemoryReactor
|
||||
from synapse.api.errors import NotFoundError, SynapseError
|
||||
from synapse.rest.client import room
|
||||
from synapse.server import HomeServer
|
||||
from synapse.types.state import StateFilter
|
||||
from synapse.util import Clock
|
||||
|
||||
from tests.unittest import HomeserverTestCase
|
||||
@@ -41,8 +40,6 @@ class PurgeTests(HomeserverTestCase):
|
||||
self.room_id = self.helper.create_room_as(self.user_id)
|
||||
|
||||
self.store = hs.get_datastores().main
|
||||
self.state_store = hs.get_datastores().state
|
||||
self.state_deletion_store = hs.get_datastores().state_deletion
|
||||
self._storage_controllers = self.hs.get_storage_controllers()
|
||||
|
||||
def test_purge_history(self) -> None:
|
||||
@@ -131,142 +128,3 @@ class PurgeTests(HomeserverTestCase):
|
||||
self.store._invalidate_local_get_event_cache(create_event.event_id)
|
||||
self.get_failure(self.store.get_event(create_event.event_id), NotFoundError)
|
||||
self.get_failure(self.store.get_event(first["event_id"]), NotFoundError)
|
||||
|
||||
def test_purge_history_deletes_state_groups(self) -> None:
|
||||
"""Test that unreferenced state groups get cleaned up after purge"""
|
||||
|
||||
# Send four state changes to the room.
|
||||
first = self.helper.send_state(
|
||||
self.room_id, event_type="m.foo", body={"test": 1}
|
||||
)
|
||||
second = self.helper.send_state(
|
||||
self.room_id, event_type="m.foo", body={"test": 2}
|
||||
)
|
||||
third = self.helper.send_state(
|
||||
self.room_id, event_type="m.foo", body={"test": 3}
|
||||
)
|
||||
last = self.helper.send_state(
|
||||
self.room_id, event_type="m.foo", body={"test": 4}
|
||||
)
|
||||
|
||||
# Get references to the state groups
|
||||
event_to_groups = self.get_success(
|
||||
self.store._get_state_group_for_events(
|
||||
[
|
||||
first["event_id"],
|
||||
second["event_id"],
|
||||
third["event_id"],
|
||||
last["event_id"],
|
||||
]
|
||||
)
|
||||
)
|
||||
|
||||
# Get the topological token
|
||||
token = self.get_success(
|
||||
self.store.get_topological_token_for_event(last["event_id"])
|
||||
)
|
||||
token_str = self.get_success(token.to_string(self.hs.get_datastores().main))
|
||||
|
||||
# Purge everything before this topological token
|
||||
self.get_success(
|
||||
self._storage_controllers.purge_events.purge_history(
|
||||
self.room_id, token_str, True
|
||||
)
|
||||
)
|
||||
|
||||
# Advance so that the background jobs to delete the state groups runs
|
||||
self.reactor.advance(
|
||||
1 + self.state_deletion_store.DELAY_BEFORE_DELETION_MS / 1000
|
||||
)
|
||||
|
||||
# We expect all the state groups associated with events above, except
|
||||
# the last one, should return no state.
|
||||
state_groups = self.get_success(
|
||||
self.state_store._get_state_groups_from_groups(
|
||||
list(event_to_groups.values()), StateFilter.all()
|
||||
)
|
||||
)
|
||||
first_state = state_groups[event_to_groups[first["event_id"]]]
|
||||
second_state = state_groups[event_to_groups[second["event_id"]]]
|
||||
third_state = state_groups[event_to_groups[third["event_id"]]]
|
||||
last_state = state_groups[event_to_groups[last["event_id"]]]
|
||||
|
||||
self.assertEqual(first_state, {})
|
||||
self.assertEqual(second_state, {})
|
||||
self.assertEqual(third_state, {})
|
||||
self.assertNotEqual(last_state, {})
|
||||
|
||||
def test_purge_unreferenced_state_group(self) -> None:
|
||||
"""Test that purging a room also gets rid of unreferenced state groups
|
||||
it encounters during the purge.
|
||||
|
||||
This is important, as otherwise these unreferenced state groups get
|
||||
"de-deltaed" during the purge process, consuming lots of disk space.
|
||||
"""
|
||||
|
||||
self.helper.send(self.room_id, body="test1")
|
||||
state1 = self.helper.send_state(
|
||||
self.room_id, "org.matrix.test", body={"number": 2}
|
||||
)
|
||||
state2 = self.helper.send_state(
|
||||
self.room_id, "org.matrix.test", body={"number": 3}
|
||||
)
|
||||
self.helper.send(self.room_id, body="test4")
|
||||
last = self.helper.send(self.room_id, body="test5")
|
||||
|
||||
# Create an unreferenced state group that has a prev group of one of the
|
||||
# to-be-purged events.
|
||||
prev_group = self.get_success(
|
||||
self.store._get_state_group_for_event(state1["event_id"])
|
||||
)
|
||||
unreferenced_state_group = self.get_success(
|
||||
self.state_store.store_state_group(
|
||||
event_id=last["event_id"],
|
||||
room_id=self.room_id,
|
||||
prev_group=prev_group,
|
||||
delta_ids={("org.matrix.test", ""): state2["event_id"]},
|
||||
current_state_ids=None,
|
||||
)
|
||||
)
|
||||
|
||||
# Get the topological token
|
||||
token = self.get_success(
|
||||
self.store.get_topological_token_for_event(last["event_id"])
|
||||
)
|
||||
token_str = self.get_success(token.to_string(self.hs.get_datastores().main))
|
||||
|
||||
# Purge everything before this topological token
|
||||
self.get_success(
|
||||
self._storage_controllers.purge_events.purge_history(
|
||||
self.room_id, token_str, True
|
||||
)
|
||||
)
|
||||
|
||||
# Advance so that the background jobs to delete the state groups runs
|
||||
self.reactor.advance(
|
||||
1 + self.state_deletion_store.DELAY_BEFORE_DELETION_MS / 1000
|
||||
)
|
||||
|
||||
# We expect that the unreferenced state group has been deleted.
|
||||
row = self.get_success(
|
||||
self.state_store.db_pool.simple_select_one_onecol(
|
||||
table="state_groups",
|
||||
keyvalues={"id": unreferenced_state_group},
|
||||
retcol="id",
|
||||
allow_none=True,
|
||||
desc="test_purge_unreferenced_state_group",
|
||||
)
|
||||
)
|
||||
self.assertIsNone(row)
|
||||
|
||||
# We expect there to now only be one state group for the room, which is
|
||||
# the state group of the last event (as the only outlier).
|
||||
state_groups = self.get_success(
|
||||
self.state_store.db_pool.simple_select_onecol(
|
||||
table="state_groups",
|
||||
keyvalues={"room_id": self.room_id},
|
||||
retcol="id",
|
||||
desc="test_purge_unreferenced_state_group",
|
||||
)
|
||||
)
|
||||
self.assertEqual(len(state_groups), 1)
|
||||
|
||||
@@ -1,475 +0,0 @@
|
||||
#
|
||||
# This file is licensed under the Affero General Public License (AGPL) version 3.
|
||||
#
|
||||
# Copyright (C) 2025 New Vector, Ltd
|
||||
#
|
||||
# This program is free software: you can redistribute it and/or modify
|
||||
# it under the terms of the GNU Affero General Public License as
|
||||
# published by the Free Software Foundation, either version 3 of the
|
||||
# License, or (at your option) any later version.
|
||||
#
|
||||
# See the GNU Affero General Public License for more details:
|
||||
# <https://www.gnu.org/licenses/agpl-3.0.html>.
|
||||
#
|
||||
|
||||
|
||||
import logging
|
||||
|
||||
from twisted.test.proto_helpers import MemoryReactor
|
||||
|
||||
from synapse.rest import admin
|
||||
from synapse.rest.client import login, room
|
||||
from synapse.server import HomeServer
|
||||
from synapse.util import Clock
|
||||
|
||||
from tests.test_utils.event_injection import create_event
|
||||
from tests.unittest import HomeserverTestCase
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
|
||||
class StateDeletionStoreTestCase(HomeserverTestCase):
|
||||
"""Tests for the StateDeletionStore."""
|
||||
|
||||
servlets = [
|
||||
admin.register_servlets,
|
||||
room.register_servlets,
|
||||
login.register_servlets,
|
||||
]
|
||||
|
||||
def prepare(self, reactor: MemoryReactor, clock: Clock, hs: HomeServer) -> None:
|
||||
self.store = hs.get_datastores().main
|
||||
self.state_store = hs.get_datastores().state
|
||||
self.state_deletion_store = hs.get_datastores().state_deletion
|
||||
self.purge_events = hs.get_storage_controllers().purge_events
|
||||
|
||||
# We want to disable the automatic deletion of state groups in the
|
||||
# background, so we can do controlled tests.
|
||||
self.purge_events._delete_state_loop_call.stop()
|
||||
|
||||
self.user_id = self.register_user("test", "password")
|
||||
tok = self.login("test", "password")
|
||||
self.room_id = self.helper.create_room_as(self.user_id, tok=tok)
|
||||
|
||||
def check_if_can_be_deleted(self, state_group: int) -> bool:
|
||||
"""Check if the state group is pending deletion."""
|
||||
|
||||
state_group_to_sequence_number = self.get_success(
|
||||
self.state_deletion_store.get_pending_deletions([state_group])
|
||||
)
|
||||
|
||||
can_be_deleted = self.get_success(
|
||||
self.state_deletion_store.db_pool.runInteraction(
|
||||
"test_existing_pending_deletion_is_cleared",
|
||||
self.state_deletion_store.get_state_groups_ready_for_potential_deletion_txn,
|
||||
state_group_to_sequence_number,
|
||||
)
|
||||
)
|
||||
|
||||
return state_group in can_be_deleted
|
||||
|
||||
def test_no_deletion(self) -> None:
|
||||
"""Test that calling persisting_state_group_references is fine if
|
||||
nothing is pending deletion"""
|
||||
event, context = self.get_success(
|
||||
create_event(
|
||||
self.hs,
|
||||
room_id=self.room_id,
|
||||
type="m.test",
|
||||
sender=self.user_id,
|
||||
)
|
||||
)
|
||||
|
||||
ctx_mgr = self.state_deletion_store.persisting_state_group_references(
|
||||
[(event, context)]
|
||||
)
|
||||
|
||||
self.get_success(ctx_mgr.__aenter__())
|
||||
self.get_success(ctx_mgr.__aexit__(None, None, None))
|
||||
|
||||
def test_no_deletion_error(self) -> None:
|
||||
"""Test that calling persisting_state_group_references is fine if
|
||||
nothing is pending deletion, but an error occurs."""
|
||||
|
||||
event, context = self.get_success(
|
||||
create_event(
|
||||
self.hs,
|
||||
room_id=self.room_id,
|
||||
type="m.test",
|
||||
sender=self.user_id,
|
||||
)
|
||||
)
|
||||
|
||||
ctx_mgr = self.state_deletion_store.persisting_state_group_references(
|
||||
[(event, context)]
|
||||
)
|
||||
|
||||
self.get_success(ctx_mgr.__aenter__())
|
||||
self.get_success(ctx_mgr.__aexit__(Exception, Exception("test"), None))
|
||||
|
||||
def test_existing_pending_deletion_is_cleared(self) -> None:
|
||||
"""Test that the pending deletion flag gets cleared when the state group
|
||||
gets persisted."""
|
||||
|
||||
event, context = self.get_success(
|
||||
create_event(
|
||||
self.hs,
|
||||
room_id=self.room_id,
|
||||
type="m.test",
|
||||
state_key="",
|
||||
sender=self.user_id,
|
||||
)
|
||||
)
|
||||
assert context.state_group is not None
|
||||
|
||||
# Mark a state group that we're referencing as pending deletion.
|
||||
self.get_success(
|
||||
self.state_deletion_store.mark_state_groups_as_pending_deletion(
|
||||
[context.state_group]
|
||||
)
|
||||
)
|
||||
|
||||
ctx_mgr = self.state_deletion_store.persisting_state_group_references(
|
||||
[(event, context)]
|
||||
)
|
||||
|
||||
self.get_success(ctx_mgr.__aenter__())
|
||||
self.get_success(ctx_mgr.__aexit__(None, None, None))
|
||||
|
||||
# The pending deletion flag should be cleared
|
||||
pending_deletion = self.get_success(
|
||||
self.state_deletion_store.db_pool.simple_select_one_onecol(
|
||||
table="state_groups_pending_deletion",
|
||||
keyvalues={"state_group": context.state_group},
|
||||
retcol="1",
|
||||
allow_none=True,
|
||||
desc="test_existing_pending_deletion_is_cleared",
|
||||
)
|
||||
)
|
||||
self.assertIsNone(pending_deletion)
|
||||
|
||||
def test_pending_deletion_is_cleared_during_persist(self) -> None:
|
||||
"""Test that the pending deletion flag is cleared when a state group
|
||||
gets marked for deletion during persistence"""
|
||||
|
||||
event, context = self.get_success(
|
||||
create_event(
|
||||
self.hs,
|
||||
room_id=self.room_id,
|
||||
type="m.test",
|
||||
state_key="",
|
||||
sender=self.user_id,
|
||||
)
|
||||
)
|
||||
assert context.state_group is not None
|
||||
|
||||
ctx_mgr = self.state_deletion_store.persisting_state_group_references(
|
||||
[(event, context)]
|
||||
)
|
||||
self.get_success(ctx_mgr.__aenter__())
|
||||
|
||||
# Mark the state group that we're referencing as pending deletion,
|
||||
# *after* we have started persisting.
|
||||
self.get_success(
|
||||
self.state_deletion_store.mark_state_groups_as_pending_deletion(
|
||||
[context.state_group]
|
||||
)
|
||||
)
|
||||
|
||||
self.get_success(ctx_mgr.__aexit__(None, None, None))
|
||||
|
||||
# The pending deletion flag should be cleared
|
||||
pending_deletion = self.get_success(
|
||||
self.state_deletion_store.db_pool.simple_select_one_onecol(
|
||||
table="state_groups_pending_deletion",
|
||||
keyvalues={"state_group": context.state_group},
|
||||
retcol="1",
|
||||
allow_none=True,
|
||||
desc="test_existing_pending_deletion_is_cleared",
|
||||
)
|
||||
)
|
||||
self.assertIsNone(pending_deletion)
|
||||
|
||||
def test_deletion_check(self) -> None:
|
||||
"""Test that the `get_state_groups_that_can_be_purged_txn` check is
|
||||
correct during different points of the lifecycle of persisting an
|
||||
event."""
|
||||
event, context = self.get_success(
|
||||
create_event(
|
||||
self.hs,
|
||||
room_id=self.room_id,
|
||||
type="m.test",
|
||||
state_key="",
|
||||
sender=self.user_id,
|
||||
)
|
||||
)
|
||||
assert context.state_group is not None
|
||||
|
||||
self.get_success(
|
||||
self.state_deletion_store.mark_state_groups_as_pending_deletion(
|
||||
[context.state_group]
|
||||
)
|
||||
)
|
||||
|
||||
# We shouldn't be able to delete the state group as not enough time as passed
|
||||
can_be_deleted = self.check_if_can_be_deleted(context.state_group)
|
||||
self.assertFalse(can_be_deleted)
|
||||
|
||||
# After enough time we can delete the state group
|
||||
self.reactor.advance(
|
||||
1 + self.state_deletion_store.DELAY_BEFORE_DELETION_MS / 1000
|
||||
)
|
||||
can_be_deleted = self.check_if_can_be_deleted(context.state_group)
|
||||
self.assertTrue(can_be_deleted)
|
||||
|
||||
ctx_mgr = self.state_deletion_store.persisting_state_group_references(
|
||||
[(event, context)]
|
||||
)
|
||||
self.get_success(ctx_mgr.__aenter__())
|
||||
|
||||
# But once we start persisting we can't delete the state group
|
||||
can_be_deleted = self.check_if_can_be_deleted(context.state_group)
|
||||
self.assertFalse(can_be_deleted)
|
||||
|
||||
self.get_success(ctx_mgr.__aexit__(None, None, None))
|
||||
|
||||
# The pending deletion flag should remain cleared after persistence has
|
||||
# finished.
|
||||
can_be_deleted = self.check_if_can_be_deleted(context.state_group)
|
||||
self.assertFalse(can_be_deleted)
|
||||
|
||||
def test_deletion_error_during_persistence(self) -> None:
|
||||
"""Test that state groups remain marked as pending deletion if persisting
|
||||
the event fails."""
|
||||
|
||||
event, context = self.get_success(
|
||||
create_event(
|
||||
self.hs,
|
||||
room_id=self.room_id,
|
||||
type="m.test",
|
||||
state_key="",
|
||||
sender=self.user_id,
|
||||
)
|
||||
)
|
||||
assert context.state_group is not None
|
||||
|
||||
# Mark a state group that we're referencing as pending deletion.
|
||||
self.get_success(
|
||||
self.state_deletion_store.mark_state_groups_as_pending_deletion(
|
||||
[context.state_group]
|
||||
)
|
||||
)
|
||||
|
||||
ctx_mgr = self.state_deletion_store.persisting_state_group_references(
|
||||
[(event, context)]
|
||||
)
|
||||
|
||||
self.get_success(ctx_mgr.__aenter__())
|
||||
self.get_success(ctx_mgr.__aexit__(Exception, Exception("test"), None))
|
||||
|
||||
# We should be able to delete the state group after a certain amount of
|
||||
# time
|
||||
self.reactor.advance(
|
||||
1 + self.state_deletion_store.DELAY_BEFORE_DELETION_MS / 1000
|
||||
)
|
||||
can_be_deleted = self.check_if_can_be_deleted(context.state_group)
|
||||
self.assertTrue(can_be_deleted)
|
||||
|
||||
def test_race_between_check_and_insert(self) -> None:
|
||||
"""Check that we correctly handle the race where we go to delete a
|
||||
state group, check that it is unreferenced, and then it becomes
|
||||
referenced just before we delete it."""
|
||||
|
||||
event, context = self.get_success(
|
||||
create_event(
|
||||
self.hs,
|
||||
room_id=self.room_id,
|
||||
type="m.test",
|
||||
state_key="",
|
||||
sender=self.user_id,
|
||||
)
|
||||
)
|
||||
assert context.state_group is not None
|
||||
|
||||
# Mark a state group that we're referencing as pending deletion.
|
||||
self.get_success(
|
||||
self.state_deletion_store.mark_state_groups_as_pending_deletion(
|
||||
[context.state_group]
|
||||
)
|
||||
)
|
||||
|
||||
# Advance time enough so we can delete the state group
|
||||
self.reactor.advance(
|
||||
1 + self.state_deletion_store.DELAY_BEFORE_DELETION_MS / 1000
|
||||
)
|
||||
|
||||
# Check that we'd be able to delete this state group.
|
||||
state_group_to_sequence_number = self.get_success(
|
||||
self.state_deletion_store.get_pending_deletions([context.state_group])
|
||||
)
|
||||
|
||||
can_be_deleted = self.get_success(
|
||||
self.state_deletion_store.db_pool.runInteraction(
|
||||
"test_existing_pending_deletion_is_cleared",
|
||||
self.state_deletion_store.get_state_groups_ready_for_potential_deletion_txn,
|
||||
state_group_to_sequence_number,
|
||||
)
|
||||
)
|
||||
self.assertCountEqual(can_be_deleted, [context.state_group])
|
||||
|
||||
# ... in the real world we'd check that the state group isn't referenced here ...
|
||||
|
||||
# Now we persist the event to reference the state group, *after* we
|
||||
# check that the state group wasn't referenced
|
||||
ctx_mgr = self.state_deletion_store.persisting_state_group_references(
|
||||
[(event, context)]
|
||||
)
|
||||
|
||||
self.get_success(ctx_mgr.__aenter__())
|
||||
self.get_success(ctx_mgr.__aexit__(Exception, Exception("test"), None))
|
||||
|
||||
# We simulate a pause (required to hit the race)
|
||||
self.reactor.advance(
|
||||
1 + self.state_deletion_store.DELAY_BEFORE_DELETION_MS / 1000
|
||||
)
|
||||
|
||||
# We should no longer be able to delete the state group, without having
|
||||
# to recheck if its referenced.
|
||||
can_be_deleted = self.get_success(
|
||||
self.state_deletion_store.db_pool.runInteraction(
|
||||
"test_existing_pending_deletion_is_cleared",
|
||||
self.state_deletion_store.get_state_groups_ready_for_potential_deletion_txn,
|
||||
state_group_to_sequence_number,
|
||||
)
|
||||
)
|
||||
self.assertCountEqual(can_be_deleted, [])
|
||||
|
||||
def test_remove_ancestors_from_can_delete(self) -> None:
|
||||
"""Test that if a state group is not ready to be deleted, we also don't
|
||||
delete anything that is referenced by it"""
|
||||
|
||||
event, context = self.get_success(
|
||||
create_event(
|
||||
self.hs,
|
||||
room_id=self.room_id,
|
||||
type="m.test",
|
||||
state_key="",
|
||||
sender=self.user_id,
|
||||
)
|
||||
)
|
||||
assert context.state_group is not None
|
||||
|
||||
# Create a new state group that references the one from the event
|
||||
new_state_group = self.get_success(
|
||||
self.state_store.store_state_group(
|
||||
event.event_id,
|
||||
event.room_id,
|
||||
prev_group=context.state_group,
|
||||
delta_ids={},
|
||||
current_state_ids=None,
|
||||
)
|
||||
)
|
||||
|
||||
# Mark them both as pending deletion
|
||||
self.get_success(
|
||||
self.state_deletion_store.mark_state_groups_as_pending_deletion(
|
||||
[context.state_group, new_state_group]
|
||||
)
|
||||
)
|
||||
|
||||
# Advance time enough so we can delete the state group so they're both
|
||||
# ready for deletion.
|
||||
self.reactor.advance(
|
||||
1 + self.state_deletion_store.DELAY_BEFORE_DELETION_MS / 1000
|
||||
)
|
||||
|
||||
# We can now delete both state groups
|
||||
self.assertTrue(self.check_if_can_be_deleted(context.state_group))
|
||||
self.assertTrue(self.check_if_can_be_deleted(new_state_group))
|
||||
|
||||
# Use the new_state_group to bump its deletion time
|
||||
self.get_success(
|
||||
self.state_store.store_state_group(
|
||||
event.event_id,
|
||||
event.room_id,
|
||||
prev_group=new_state_group,
|
||||
delta_ids={},
|
||||
current_state_ids=None,
|
||||
)
|
||||
)
|
||||
|
||||
# We should now not be able to delete either of the state groups.
|
||||
state_group_to_sequence_number = self.get_success(
|
||||
self.state_deletion_store.get_pending_deletions(
|
||||
[context.state_group, new_state_group]
|
||||
)
|
||||
)
|
||||
|
||||
# We shouldn't be able to delete the state group as not enough time has passed
|
||||
can_be_deleted = self.get_success(
|
||||
self.state_deletion_store.db_pool.runInteraction(
|
||||
"test_existing_pending_deletion_is_cleared",
|
||||
self.state_deletion_store.get_state_groups_ready_for_potential_deletion_txn,
|
||||
state_group_to_sequence_number,
|
||||
)
|
||||
)
|
||||
self.assertCountEqual(can_be_deleted, [])
|
||||
|
||||
def test_newly_referenced_state_group_gets_removed_from_pending(self) -> None:
|
||||
"""Check that if a state group marked for deletion becomes referenced
|
||||
(without being removed from pending deletion table), it gets removed
|
||||
from pending deletion table."""
|
||||
|
||||
event, context = self.get_success(
|
||||
create_event(
|
||||
self.hs,
|
||||
room_id=self.room_id,
|
||||
type="m.test",
|
||||
state_key="",
|
||||
sender=self.user_id,
|
||||
)
|
||||
)
|
||||
assert context.state_group is not None
|
||||
|
||||
# Mark a state group that we're referencing as pending deletion.
|
||||
self.get_success(
|
||||
self.state_deletion_store.mark_state_groups_as_pending_deletion(
|
||||
[context.state_group]
|
||||
)
|
||||
)
|
||||
|
||||
# Advance time enough so we can delete the state group so they're both
|
||||
# ready for deletion.
|
||||
self.reactor.advance(
|
||||
1 + self.state_deletion_store.DELAY_BEFORE_DELETION_MS / 1000
|
||||
)
|
||||
|
||||
# Manually insert into the table to mimic the state group getting used.
|
||||
self.get_success(
|
||||
self.store.db_pool.simple_insert(
|
||||
table="event_to_state_groups",
|
||||
values={"state_group": context.state_group, "event_id": event.event_id},
|
||||
desc="test_newly_referenced_state_group_gets_removed_from_pending",
|
||||
)
|
||||
)
|
||||
|
||||
# Manually run the background task to delete pending state groups.
|
||||
self.get_success(self.purge_events._delete_state_groups_loop())
|
||||
|
||||
# The pending deletion flag should be cleared...
|
||||
pending_deletion = self.get_success(
|
||||
self.state_deletion_store.db_pool.simple_select_one_onecol(
|
||||
table="state_groups_pending_deletion",
|
||||
keyvalues={"state_group": context.state_group},
|
||||
retcol="1",
|
||||
allow_none=True,
|
||||
desc="test_newly_referenced_state_group_gets_removed_from_pending",
|
||||
)
|
||||
)
|
||||
self.assertIsNone(pending_deletion)
|
||||
|
||||
# .. but the state should not have been deleted.
|
||||
state = self.get_success(
|
||||
self.state_store._get_state_for_groups([context.state_group])
|
||||
)
|
||||
self.assertGreater(len(state[context.state_group]), 0)
|
||||
@@ -31,7 +31,7 @@ from typing import (
|
||||
Tuple,
|
||||
cast,
|
||||
)
|
||||
from unittest.mock import AsyncMock, Mock
|
||||
from unittest.mock import Mock
|
||||
|
||||
from twisted.internet import defer
|
||||
|
||||
@@ -221,16 +221,7 @@ class Graph:
|
||||
class StateTestCase(unittest.TestCase):
|
||||
def setUp(self) -> None:
|
||||
self.dummy_store = _DummyStore()
|
||||
|
||||
# Add a dummy epoch store that always retruns that we have all the
|
||||
# necessary state groups.
|
||||
dummy_deletion_store = AsyncMock()
|
||||
dummy_deletion_store.check_state_groups_and_bump_deletion.return_value = []
|
||||
|
||||
storage_controllers = Mock(
|
||||
main=self.dummy_store,
|
||||
state=self.dummy_store,
|
||||
)
|
||||
storage_controllers = Mock(main=self.dummy_store, state=self.dummy_store)
|
||||
hs = Mock(
|
||||
spec_set=[
|
||||
"config",
|
||||
@@ -250,10 +241,7 @@ class StateTestCase(unittest.TestCase):
|
||||
)
|
||||
clock = cast(Clock, MockClock())
|
||||
hs.config = default_config("tesths", True)
|
||||
hs.get_datastores.return_value = Mock(
|
||||
main=self.dummy_store,
|
||||
state_deletion=dummy_deletion_store,
|
||||
)
|
||||
hs.get_datastores.return_value = Mock(main=self.dummy_store)
|
||||
hs.get_state_handler.return_value = None
|
||||
hs.get_clock.return_value = clock
|
||||
hs.get_macaroon_generator.return_value = MacaroonGenerator(
|
||||
|
||||
Reference in New Issue
Block a user