Try and drop stale extremities. #8929

erikjohnston · 2020-12-11T13:34:03Z

If we see stale extremities while persisting events, and notice that they don't change the result of state resolution, we drop them.

~~I need to add some tests~~ A couple of tests have been added.

synapse/storage/persist_events.py

clokep

Seems OK, it would be nice to have tests for the other causes (dummy events / events from other servers / events that are too old/deep).

clokep · 2020-12-11T16:04:09Z

changelog.d/8929.misc

@@ -0,0 +1 @@
+Automatically drop stale extremities under some specific conditions.


Is this really a misc change?

We tend to put perf related stuff in misc 🤷

synapse/storage/persist_events.py

tests/storage/test_events.py

changelog.d/8929.misc

synapse/storage/persist_events.py

richvdh · 2020-12-11T17:50:25Z

synapse/storage/persist_events.py

+        one_day_ago = self._clock.time_msec() - 20 * 60 * 60 * 1000
+        current_depth = max(e.depth for e, _ in events_context)
+        for event in dropped_events.values():
+            if self.is_mine_id(event.sender) and event.type != "org.matrix.dummy_event":


should use a constant from EventTypes for this, to make for easier searching. (I'm disappointed to find we don't already have one)

This heuristic feels magical. What makes it safe to ignore dummy events?

This heuristic feels magical. What makes it safe to ignore dummy events?

My, perhaps flawed, thinking is that since we generally only send dummy events in rooms that the server isn't sending "proper" events into, old dummy events are likely to only point to old remote events, which we are OK with dropping.

The reason to do this is that dummy events are going to be a source of stuck extremities, so it would be good to be able to prune them.

old dummy events are likely to only point to old remote events, which we are OK with dropping.

the implication here is that it is not ok to drop forward extremities that are local events. You haven't spelt out the reason for this, but presumably it's to avoid us leaving dangling extremities across the whole federation?

the thing is, it's absolutely not a given that all a dummy event's predecessors are remote events, particularly in a situation where federation starts dropping out due to load on one side or the other (or there simply aren't enough "real" events moving around to de-extremify any local events). So dropping such a dummy event would be equivalent to dropping those local predecessors as extremities.

I wonder if we could/should strengthen the test to: "if it's a dummy event, then it's ok to drop it if either (1) all the dummy's predecessors are remote or (2) any local predecessors have other successors (that we aren't dropping)."

the implication here is that it is not ok to drop forward extremities that are local events. You haven't spelt out the reason for this, but presumably it's to avoid us leaving dangling extremities across the whole federation?

So my reasoning to not drop local extremities is to give the server a chance to include them as prev events when sending new events, so that other servers will notice if they didn't get the previous extremities. I don't think dropping local, stale extremities will cause them to dangle elsewhere? It's just that either a) we missed a remote event that referenced it or b) we failed to send it out to the server that is generating events (if not us).

the thing is, it's absolutely not a given that all a dummy event's predecessors are remote events, particularly in a situation where federation starts dropping out due to load on one side or the other (or there simply aren't enough "real" events moving around to de-extremify any local events). So dropping such a dummy event would be equivalent to dropping those local predecessors as extremities.

Yes, though to be a candidate for pruning the dummy event needs to either be old, or for the server to be generating events. I think the danger here is that the server ends up with many forward extremities, so new events don't reference all of the existing extremities, and so we prune extremities even though they weren't "stale" as such.

wonder if we could/should strengthen the test to: "if it's a dummy event, then it's ok to drop it if either (1) all the dummy's predecessors are remote or (2) any local predecessors have other successors (that we aren't dropping)."

This is actually fine to do, for some reason I didn't think that was a thing we could look at. Will add

Actually, as a first cut I might just only check that dummy events point to remote events. That should catch the majority of stuck dummy extremities.

I've implemented dropping of dummy events if and only if they only reference remote events or other dummy events that only reference remote events (etc).

synapse/storage/persist_events.py

erikjohnston · 2020-12-14T13:09:10Z

synapse/storage/persist_events.py

+                and event.depth < current_depth - 100
+            ):
+                continue
+            if get_domain_from_id(event.sender) in new_senders:


Thinking about this, we should probably still check that depth has increased a bit, otherwise we'll drop extremities immediately.

I've added a check on depth as well

richvdh · 2020-12-14T14:01:32Z

synapse/storage/persist_events.py

+            # we'll immediately remove all extremities from a given server).
+            if (
+                get_domain_from_id(event.sender) in new_senders
+                and event.depth < current_depth - 20


this now seems to be in conflict with 'the extremity must give the same state as the "current state"'. In the rooms where this is a problem, it's relatively uncommon to make it 20 events without a state change. My concern is therefore that this whole algorithm will never actually be useful.

Why 20 here?

this now seems to be in conflict with 'the extremity must give the same state as the "current state"'. In the rooms where this is a problem, it's relatively uncommon to make it 20 events without a state change. My concern is therefore that this whole algorithm will never actually be useful.

State changes are fine, what we're looking for is that the resolved state with all extremities is the same as the resolved state without the old extremities. This should be true for simple gaps (rather than forks in the graph for example).

Why 20 here?

It's a fairly arbitrary low but non-zero number. What I saw on jki.re over the weekend is that we'd get two extremities from matrix.org and then immediately prune one, even though a new event will come along and point to both, which feels a bit wrong. Maybe using depth here is sort of the wrong thing.

I wonder if we should also add a check for it being an hour old, or something? I guess we're saying dropping events shouldn't be problematic for remote events, but I could imagine a case where we have a lot of events sent all at once and so we end up with lots of extremities that will take some time to be referenced

at this point we should probably just deploy something and maybe iterate on it?

Yup, agreed

If we see stale extremities while persisting events, and notice that they don't change the result of state resolution, we drop them.

Co-authored-by: Richard van der Hoff <1389908+richvdh@users.noreply.github.com>

erikjohnston · 2020-12-14T16:04:38Z

I've rebased off of v1.24 as discussed elsewhere

If we see stale extremities while persisting events, and notice that they don't change the result of state resolution, we drop them.

@dklimpel

Synapse 1.25.0rc1 (2021-01-06) ============================== Removal warning --------------- The old [Purge Room API](https://github.com/matrix-org/synapse/tree/master/docs/admin_api/purge_room.md) and [Shutdown Room API](https://github.com/matrix-org/synapse/tree/master/docs/admin_api/shutdown_room.md) are deprecated and will be removed in a future release. They will be replaced by the [Delete Room API](https://github.com/matrix-org/synapse/tree/master/docs/admin_api/rooms.md#delete-room-api). `POST /_synapse/admin/v1/rooms/<room_id>/delete` replaces `POST /_synapse/admin/v1/purge_room` and `POST /_synapse/admin/v1/shutdown_room/<room_id>`. Features -------- - Add an admin API that lets server admins get power in rooms in which local users have power. ([\#8756](#8756)) - Add optional HTTP authentication to replication endpoints. ([\#8853](#8853)) - Improve the error messages printed as a result of configuration problems for extension modules. ([\#8874](#8874)) - Add the number of local devices to Room Details Admin API. Contributed by @dklimpel. ([\#8886](#8886)) - Add `X-Robots-Tag` header to stop web crawlers from indexing media. Contributed by Aaron Raimist. ([\#8887](#8887)) - Spam-checkers may now define their methods as `async`. ([\#8890](#8890)) - Add support for allowing users to pick their own user ID during a single-sign-on login. ([\#8897](#8897), [\#8900](#8900), [\#8911](#8911), [\#8938](#8938), [\#8941](#8941), [\#8942](#8942), [\#8951](#8951)) - Add an `email.invite_client_location` configuration option to send a web client location to the invite endpoint on the identity server which allows customisation of the email template. ([\#8930](#8930)) - The search term in the list room and list user Admin APIs is now treated as case-insensitive. ([\#8931](#8931)) - Apply an IP range blacklist to push and key revocation requests. ([\#8821](#8821), [\#8870](#8870), [\#8954](#8954)) - Add an option to allow re-use of user-interactive authentication sessions for a period of time. ([\#8970](#8970)) - Allow running the redact endpoint on workers. ([\#8994](#8994)) Bugfixes -------- - Fix bug where we might not correctly calculate the current state for rooms with multiple extremities. ([\#8827](#8827)) - Fix a long-standing bug in the register admin endpoint (`/_synapse/admin/v1/register`) when the `mac` field was not provided. The endpoint now properly returns a 400 error. Contributed by @edwargix. ([\#8837](#8837)) - Fix a long-standing bug on Synapse instances supporting Single-Sign-On, where users would be prompted to enter their password to confirm certain actions, even though they have not set a password. ([\#8858](#8858)) - Fix a longstanding bug where a 500 error would be returned if the `Content-Length` header was not provided to the upload media resource. ([\#8862](#8862)) - Add additional validation to pusher URLs to be compliant with the specification. ([\#8865](#8865)) - Fix the error code that is returned when a user tries to register on a homeserver on which new-user registration has been disabled. ([\#8867](#8867)) - Fix a bug where `PUT /_synapse/admin/v2/users/<user_id>` failed to create a new user when `avatar_url` is specified. Bug introduced in Synapse v1.9.0. ([\#8872](#8872)) - Fix a 500 error when attempting to preview an empty HTML file. ([\#8883](#8883)) - Fix occasional deadlock when handling SIGHUP. ([\#8918](#8918)) - Fix login API to not ratelimit application services that have ratelimiting disabled. ([\#8920](#8920)) - Fix bug where we ratelimited auto joining of rooms on registration (using `auto_join_rooms` config). ([\#8921](#8921)) - Fix a bug where deactivated users appeared in the user directory when their profile information was updated. ([\#8933](#8933), [\#8964](#8964)) - Fix bug introduced in Synapse v1.24.0 which would cause an exception on startup if both `enabled` and `localdb_enabled` were set to `False` in the `password_config` setting of the configuration file. ([\#8937](#8937)) - Fix a bug where 500 errors would be returned if the `m.room_history_visibility` event had invalid content. ([\#8945](#8945)) - Fix a bug causing common English words to not be considered for a user directory search. ([\#8959](#8959)) - Fix bug where application services couldn't register new ghost users if the server had reached its MAU limit. ([\#8962](#8962)) - Fix a long-standing bug where a `m.image` event without a `url` would cause errors on push. ([\#8965](#8965)) - Fix a small bug in v2 state resolution algorithm, which could also cause performance issues for rooms with large numbers of power levels. ([\#8971](#8971)) - Add validation to the `sendToDevice` API to raise a missing parameters error instead of a 500 error. ([\#8975](#8975)) - Add validation of group IDs to raise a 400 error instead of a 500 eror. ([\#8977](#8977)) Improved Documentation ---------------------- - Fix the "Event persist rate" section of the included grafana dashboard by adding missing prometheus rules. ([\#8802](#8802)) - Combine related media admin API docs. ([\#8839](#8839)) - Fix an error in the documentation for the SAML username mapping provider. ([\#8873](#8873)) - Clarify comments around template directories in `sample_config.yaml`. ([\#8891](#8891)) - Moved instructions for database setup, adjusted heading levels and improved syntax highlighting in [INSTALL.md](../INSTALL.md). Contributed by fossterer. ([\#8987](#8987)) - Update the example value of `group_creation_prefix` in the sample configuration. ([\#8992](#8992)) - Link the Synapse developer room to the development section in the docs. ([\#9002](#9002)) Deprecations and Removals ------------------------- - Deprecate Shutdown Room and Purge Room Admin APIs. ([\#8829](#8829)) Internal Changes ---------------- - Properly store the mapping of external ID to Matrix ID for CAS users. ([\#8856](#8856), [\#8958](#8958)) - Remove some unnecessary stubbing from unit tests. ([\#8861](#8861)) - Remove unused `FakeResponse` class from unit tests. ([\#8864](#8864)) - Pass `room_id` to `get_auth_chain_difference`. ([\#8879](#8879)) - Add type hints to push module. ([\#8880](#8880), [\#8882](#8882), [\#8901](#8901), [\#8940](#8940), [\#8943](#8943), [\#9020](#9020)) - Simplify logic for handling user-interactive-auth via single-sign-on servers. ([\#8881](#8881)) - Skip the SAML tests if the requirements (`pysaml2` and `xmlsec1`) aren't available. ([\#8905](#8905)) - Fix multiarch docker image builds. ([\#8906](#8906)) - Don't publish `latest` docker image until all archs are built. ([\#8909](#8909)) - Various clean-ups to the structured logging and logging context code. ([\#8916](#8916), [\#8935](#8935)) - Automatically drop stale forward-extremities under some specific conditions. ([\#8929](#8929)) - Refactor test utilities for injecting HTTP requests. ([\#8946](#8946)) - Add a maximum size of 50 kilobytes to .well-known lookups. ([\#8950](#8950)) - Fix bug in `generate_log_config` script which made it write empty files. ([\#8952](#8952)) - Clean up tox.ini file; disable coverage checking for non-test runs. ([\#8963](#8963)) - Add type hints to the admin and room list handlers. ([\#8973](#8973)) - Add type hints to the receipts and user directory handlers. ([\#8976](#8976)) - Drop the unused `local_invites` table. ([\#8979](#8979)) - Add type hints to the base storage code. ([\#8980](#8980)) - Support using PyJWT v2.0.0 in the test suite. ([\#8986](#8986)) - Fix `tests.federation.transport.RoomDirectoryFederationTests` and ensure it runs in CI. ([\#8998](#8998)) - Add type hints to the crypto module. ([\#8999](#8999))

@Bubu

Synapse 1.25.0 (2021-01-13) =========================== Ending Support for Python 3.5 and Postgres 9.5 ---------------------------------------------- With this release, the Synapse team is announcing a formal deprecation policy for our platform dependencies, like Python and PostgreSQL: All future releases of Synapse will follow the upstream end-of-life schedules. Which means: * This is the last release which guarantees support for Python 3.5. * We will end support for PostgreSQL 9.5 early next month. * We will end support for Python 3.6 and PostgreSQL 9.6 near the end of the year. Crucially, this means __we will not produce .deb packages for Debian 9 (Stretch) or Ubuntu 16.04 (Xenial)__ beyond the transition period described below. The website https://endoflife.date/ has convenient summaries of the support schedules for projects like [Python](https://endoflife.date/python) and [PostgreSQL](https://endoflife.date/postgresql). If you are unable to upgrade your environment to a supported version of Python or Postgres, we encourage you to consider using the [Synapse Docker images](./INSTALL.md#docker-images-and-ansible-playbooks) instead. ### Transition Period We will make a good faith attempt to avoid breaking compatibility in all releases through the end of March 2021. However, critical security vulnerabilities in dependencies or other unanticipated circumstances may arise which necessitate breaking compatibility earlier. We intend to continue producing .deb packages for Debian 9 (Stretch) and Ubuntu 16.04 (Xenial) through the transition period. Removal warning --------------- The old [Purge Room API](https://github.com/matrix-org/synapse/tree/master/docs/admin_api/purge_room.md) and [Shutdown Room API](https://github.com/matrix-org/synapse/tree/master/docs/admin_api/shutdown_room.md) are deprecated and will be removed in a future release. They will be replaced by the [Delete Room API](https://github.com/matrix-org/synapse/tree/master/docs/admin_api/rooms.md#delete-room-api). `POST /_synapse/admin/v1/rooms/<room_id>/delete` replaces `POST /_synapse/admin/v1/purge_room` and `POST /_synapse/admin/v1/shutdown_room/<room_id>`. Bugfixes -------- - Fix HTTP proxy support when using a proxy that is on a blacklisted IP. Introduced in v1.25.0rc1. Contributed by @Bubu. ([\#9084](matrix-org/synapse#9084)) Synapse 1.25.0rc1 (2021-01-06) ============================== Features -------- - Add an admin API that lets server admins get power in rooms in which local users have power. ([\#8756](matrix-org/synapse#8756)) - Add optional HTTP authentication to replication endpoints. ([\#8853](matrix-org/synapse#8853)) - Improve the error messages printed as a result of configuration problems for extension modules. ([\#8874](matrix-org/synapse#8874)) - Add the number of local devices to Room Details Admin API. Contributed by @dklimpel. ([\#8886](matrix-org/synapse#8886)) - Add `X-Robots-Tag` header to stop web crawlers from indexing media. Contributed by Aaron Raimist. ([\#8887](matrix-org/synapse#8887)) - Spam-checkers may now define their methods as `async`. ([\#8890](matrix-org/synapse#8890)) - Add support for allowing users to pick their own user ID during a single-sign-on login. ([\#8897](matrix-org/synapse#8897), [\#8900](matrix-org/synapse#8900), [\#8911](matrix-org/synapse#8911), [\#8938](matrix-org/synapse#8938), [\#8941](matrix-org/synapse#8941), [\#8942](matrix-org/synapse#8942), [\#8951](matrix-org/synapse#8951)) - Add an `email.invite_client_location` configuration option to send a web client location to the invite endpoint on the identity server which allows customisation of the email template. ([\#8930](matrix-org/synapse#8930)) - The search term in the list room and list user Admin APIs is now treated as case-insensitive. ([\#8931](matrix-org/synapse#8931)) - Apply an IP range blacklist to push and key revocation requests. ([\#8821](matrix-org/synapse#8821), [\#8870](matrix-org/synapse#8870), [\#8954](matrix-org/synapse#8954)) - Add an option to allow re-use of user-interactive authentication sessions for a period of time. ([\#8970](matrix-org/synapse#8970)) - Allow running the redact endpoint on workers. ([\#8994](matrix-org/synapse#8994)) Bugfixes -------- - Fix bug where we might not correctly calculate the current state for rooms with multiple extremities. ([\#8827](matrix-org/synapse#8827)) - Fix a long-standing bug in the register admin endpoint (`/_synapse/admin/v1/register`) when the `mac` field was not provided. The endpoint now properly returns a 400 error. Contributed by @edwargix. ([\#8837](matrix-org/synapse#8837)) - Fix a long-standing bug on Synapse instances supporting Single-Sign-On, where users would be prompted to enter their password to confirm certain actions, even though they have not set a password. ([\#8858](matrix-org/synapse#8858)) - Fix a longstanding bug where a 500 error would be returned if the `Content-Length` header was not provided to the upload media resource. ([\#8862](matrix-org/synapse#8862)) - Add additional validation to pusher URLs to be compliant with the specification. ([\#8865](matrix-org/synapse#8865)) - Fix the error code that is returned when a user tries to register on a homeserver on which new-user registration has been disabled. ([\#8867](matrix-org/synapse#8867)) - Fix a bug where `PUT /_synapse/admin/v2/users/<user_id>` failed to create a new user when `avatar_url` is specified. Bug introduced in Synapse v1.9.0. ([\#8872](matrix-org/synapse#8872)) - Fix a 500 error when attempting to preview an empty HTML file. ([\#8883](matrix-org/synapse#8883)) - Fix occasional deadlock when handling SIGHUP. ([\#8918](matrix-org/synapse#8918)) - Fix login API to not ratelimit application services that have ratelimiting disabled. ([\#8920](matrix-org/synapse#8920)) - Fix bug where we ratelimited auto joining of rooms on registration (using `auto_join_rooms` config). ([\#8921](matrix-org/synapse#8921)) - Fix a bug where deactivated users appeared in the user directory when their profile information was updated. ([\#8933](matrix-org/synapse#8933), [\#8964](matrix-org/synapse#8964)) - Fix bug introduced in Synapse v1.24.0 which would cause an exception on startup if both `enabled` and `localdb_enabled` were set to `False` in the `password_config` setting of the configuration file. ([\#8937](matrix-org/synapse#8937)) - Fix a bug where 500 errors would be returned if the `m.room_history_visibility` event had invalid content. ([\#8945](matrix-org/synapse#8945)) - Fix a bug causing common English words to not be considered for a user directory search. ([\#8959](matrix-org/synapse#8959)) - Fix bug where application services couldn't register new ghost users if the server had reached its MAU limit. ([\#8962](matrix-org/synapse#8962)) - Fix a long-standing bug where a `m.image` event without a `url` would cause errors on push. ([\#8965](matrix-org/synapse#8965)) - Fix a small bug in v2 state resolution algorithm, which could also cause performance issues for rooms with large numbers of power levels. ([\#8971](matrix-org/synapse#8971)) - Add validation to the `sendToDevice` API to raise a missing parameters error instead of a 500 error. ([\#8975](matrix-org/synapse#8975)) - Add validation of group IDs to raise a 400 error instead of a 500 eror. ([\#8977](matrix-org/synapse#8977)) Improved Documentation ---------------------- - Fix the "Event persist rate" section of the included grafana dashboard by adding missing prometheus rules. ([\#8802](matrix-org/synapse#8802)) - Combine related media admin API docs. ([\#8839](matrix-org/synapse#8839)) - Fix an error in the documentation for the SAML username mapping provider. ([\#8873](matrix-org/synapse#8873)) - Clarify comments around template directories in `sample_config.yaml`. ([\#8891](matrix-org/synapse#8891)) - Move instructions for database setup, adjusted heading levels and improved syntax highlighting in [INSTALL.md](../INSTALL.md). Contributed by @fossterer. ([\#8987](matrix-org/synapse#8987)) - Update the example value of `group_creation_prefix` in the sample configuration. ([\#8992](matrix-org/synapse#8992)) - Link the Synapse developer room to the development section in the docs. ([\#9002](matrix-org/synapse#9002)) Deprecations and Removals ------------------------- - Deprecate Shutdown Room and Purge Room Admin APIs. ([\#8829](matrix-org/synapse#8829)) Internal Changes ---------------- - Properly store the mapping of external ID to Matrix ID for CAS users. ([\#8856](matrix-org/synapse#8856), [\#8958](matrix-org/synapse#8958)) - Remove some unnecessary stubbing from unit tests. ([\#8861](matrix-org/synapse#8861)) - Remove unused `FakeResponse` class from unit tests. ([\#8864](matrix-org/synapse#8864)) - Pass `room_id` to `get_auth_chain_difference`. ([\#8879](matrix-org/synapse#8879)) - Add type hints to push module. ([\#8880](matrix-org/synapse#8880), [\#8882](matrix-org/synapse#8882), [\#8901](matrix-org/synapse#8901), [\#8940](matrix-org/synapse#8940), [\#8943](matrix-org/synapse#8943), [\#9020](matrix-org/synapse#9020)) - Simplify logic for handling user-interactive-auth via single-sign-on servers. ([\#8881](matrix-org/synapse#8881)) - Skip the SAML tests if the requirements (`pysaml2` and `xmlsec1`) aren't available. ([\#8905](matrix-org/synapse#8905)) - Fix multiarch docker image builds. ([\#8906](matrix-org/synapse#8906)) - Don't publish `latest` docker image until all archs are built. ([\#8909](matrix-org/synapse#8909)) - Various clean-ups to the structured logging and logging context code. ([\#8916](matrix-org/synapse#8916), [\#8935](matrix-org/synapse#8935)) - Automatically drop stale forward-extremities under some specific conditions. ([\#8929](matrix-org/synapse#8929)) - Refactor test utilities for injecting HTTP requests. ([\#8946](matrix-org/synapse#8946)) - Add a maximum size of 50 kilobytes to .well-known lookups. ([\#8950](matrix-org/synapse#8950)) - Fix bug in `generate_log_config` script which made it write empty files. ([\#8952](matrix-org/synapse#8952)) - Clean up tox.ini file; disable coverage checking for non-test runs. ([\#8963](matrix-org/synapse#8963)) - Add type hints to the admin and room list handlers. ([\#8973](matrix-org/synapse#8973)) - Add type hints to the receipts and user directory handlers. ([\#8976](matrix-org/synapse#8976)) - Drop the unused `local_invites` table. ([\#8979](matrix-org/synapse#8979)) - Add type hints to the base storage code. ([\#8980](matrix-org/synapse#8980)) - Support using PyJWT v2.0.0 in the test suite. ([\#8986](matrix-org/synapse#8986)) - Fix `tests.federation.transport.RoomDirectoryFederationTests` and ensure it runs in CI. ([\#8998](matrix-org/synapse#8998)) - Add type hints to the crypto module. ([\#8999](matrix-org/synapse#8999))

erikjohnston commented Dec 11, 2020

View reviewed changes

synapse/storage/persist_events.py Outdated Show resolved Hide resolved

erikjohnston force-pushed the erikj/drop_stale_extremities branch from b6791b6 to c9d981f Compare December 11, 2020 15:35

erikjohnston requested a review from a team December 11, 2020 15:40

clokep reviewed Dec 11, 2020

View reviewed changes

erikjohnston requested a review from clokep December 11, 2020 16:59

richvdh reviewed Dec 11, 2020

View reviewed changes

changelog.d/8929.misc Outdated Show resolved Hide resolved

richvdh reviewed Dec 11, 2020

View reviewed changes

synapse/storage/persist_events.py Outdated Show resolved Hide resolved

richvdh reviewed Dec 11, 2020

View reviewed changes

synapse/storage/persist_events.py Show resolved Hide resolved

erikjohnston force-pushed the erikj/drop_stale_extremities branch from 275729c to d0b2ccf Compare December 11, 2020 17:46

richvdh reviewed Dec 11, 2020

View reviewed changes

synapse/storage/persist_events.py Outdated Show resolved Hide resolved

synapse/storage/persist_events.py Outdated Show resolved Hide resolved

synapse/storage/persist_events.py Outdated Show resolved Hide resolved

erikjohnston requested a review from a team December 11, 2020 17:54

richvdh reviewed Dec 11, 2020

View reviewed changes

erikjohnston requested a review from richvdh December 14, 2020 11:59

erikjohnston commented Dec 14, 2020

View reviewed changes

richvdh reviewed Dec 14, 2020

View reviewed changes

clokep removed their request for review December 14, 2020 14:38

erikjohnston and others added 14 commits December 14, 2020 16:04

Try and drop stale extremities.

f12cd82

If we see stale extremities while persisting events, and notice that they don't change the result of state resolution, we drop them.

Newsfile

b4d7414

Add tests and fixup

a24c850

Refactor tests

3f80eb5

Add test for 'old' events

3dae3ea

Add other server test

cb242bc

Fixups

c12c988

More tests

f6c0e6e

Update changelog.d/8929.misc

4b28064

Co-authored-by: Richard van der Hoff <1389908+richvdh@users.noreply.github.com>

Words

c79228d

Refactor

2a08d28

Add metrics and log lines

f833656

Update names/wording

4b8de3c

Apply suggestions from code review

082ecbe

Co-authored-by: Richard van der Hoff <1389908+richvdh@users.noreply.github.com>

erikjohnston added 4 commits December 14, 2020 16:04

There are 24 hours in a day

5d0629c

Add dummy event type to constants

9ce4277

Wait a few events when pruning extremities from the same server

62c30e8

Only drop dummy events if they reference remote events

a4d8bd9

erikjohnston force-pushed the erikj/drop_stale_extremities branch from f52b159 to a4d8bd9 Compare December 14, 2020 16:04

erikjohnston merged commit 70586aa into develop Dec 18, 2020

erikjohnston deleted the erikj/drop_stale_extremities branch December 18, 2020 09:49

erikjohnston added a commit that referenced this pull request Dec 18, 2020

Try and drop stale extremities. (#8929)

0cbac08

If we see stale extremities while persisting events, and notice that they don't change the result of state resolution, we drop them.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Try and drop stale extremities. #8929

Try and drop stale extremities. #8929

erikjohnston commented Dec 11, 2020 •

edited

Loading

clokep left a comment

clokep Dec 11, 2020

erikjohnston Dec 11, 2020

richvdh Dec 11, 2020

richvdh Dec 11, 2020

erikjohnston Dec 14, 2020

richvdh Dec 14, 2020

erikjohnston Dec 14, 2020

erikjohnston Dec 14, 2020

erikjohnston Dec 14, 2020

erikjohnston Dec 14, 2020

erikjohnston Dec 14, 2020

richvdh Dec 14, 2020

erikjohnston Dec 14, 2020

erikjohnston Dec 14, 2020

richvdh Dec 14, 2020

erikjohnston Dec 14, 2020

erikjohnston commented Dec 14, 2020

		@@ -0,0 +1 @@
		Automatically drop stale extremities under some specific conditions.

Try and drop stale extremities. #8929

Try and drop stale extremities. #8929

Conversation

erikjohnston commented Dec 11, 2020 • edited Loading

clokep left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

erikjohnston commented Dec 14, 2020

erikjohnston commented Dec 11, 2020 •

edited

Loading