-
Notifications
You must be signed in to change notification settings - Fork 203
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add Sliding Sync /sync/e2ee
endpoint for To-Device messages
#17167
Add Sliding Sync /sync/e2ee
endpoint for To-Device messages
#17167
Conversation
Based on: - MSC3575: Sliding Sync (aka Sync v3): matrix-org/matrix-spec-proposals#3575 - MSC3885: Sliding Sync Extension: To-Device messages: matrix-org/matrix-spec-proposals#3885 - MSC3884: Sliding Sync Extension: E2EE: matrix-org/matrix-spec-proposals#3884
/sync/e2ee
endpoint for To-Device messages/sync/e2ee
endpoint for To-Device messages
We were using the enum just to distinguish /sync v2 vs Sliding Sync /sync/e2ee so we should just make an enum for that instead of trying to glom onto the existing `sync_type` (overloading it).
synapse/handlers/sync.py
Outdated
# TODO: Do we need to worry about these? All of this info is | ||
# normally calculated when we `_generate_sync_entry_for_rooms()` but we | ||
# probably don't want to do all of that work for this endpoint. | ||
newly_joined_rooms=frozenset(), | ||
newly_joined_or_invited_or_knocked_users=frozenset(), | ||
newly_left_rooms=frozenset(), | ||
newly_left_users=frozenset(), |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Since I'm not familiar exactly with how /sync
should operate or exactly what we care about, I need some advice on whether this part matters.
Does /sync/e2ee
need to care about the device changes that would come from doing all of that extra work with rooms?
Perhaps device_lists
(device changes) should be in their own endpoint? It seems like a lot of work but I haven't profiled or seen how fast all of things go.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Yeah, it's important that device_lists
get told about those things, which sucks a bit. I think you're right that in future we might want to reconsider bundling device_lists
and e2ee
, but I think for now we should keep it bundled (as that is what the Rust SDK expects).
We can always change it later once we can try it out in practice.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Updated to include all of the room derived information ✅
As far as I can tell, we don't have tests for the scenarios that this important information is necessary for.
It feels like all the benefits of this endpoint have fallen away though.
We're doing all the same work for this new endpoint as the old sync v2 so the performance is bound to be about the same. Which also means To-Device events will probably get backed up by other events above the same. I did add a filter to only work with membership events which might help.
Work | Sync v2 | /sync/e2ee |
---|---|---|
_generate_sync_entry_for_account_data() |
✅ | ❌ |
_generate_sync_entry_for_rooms() |
✅ | ✅ (sets up derived data for device_lists ) |
_generate_sync_entry_for_presence() |
✅ (not run for matrix.org anyway) |
❌ |
_generate_sync_entry_for_device_list() |
✅ | ✅ |
_generate_sync_entry_for_to_device() |
✅ | ✅ |
device_one_time_keys_count |
✅ | ✅ |
device_unused_fallback_key_types |
✅ | ✅ |
And it doesn't seem to relieve the Sliding Sync proxy if we're going to use the same device for sync v2 and this new endpoint. This point is being tracked in this discussion -> #17167 (comment)
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Hmm, if we can split up _generate_sync_entry_for_rooms
so that we can get just the membership changes then I think that would help a lot. Though equally happy to land it as is and do follow up in a separate PR
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
They're pretty gnarled and coupled at the moment. In order to decouple, instead of trying to re-use the data we pull out in _generate_sync_entry_for_rooms()
, it might be easier just to do our own dedicated lookup to figure out these membership changes in our own new function.
I would definitely prefer to land this and update in a follow-up 👍
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@erikjohnston Before trying to optimize /sync/e2ee
, it would be good to see just how fast/slow it goes. Traces from matrix.org
on good size account would probably give us the best indication. Not sure how hard it may be to get enough to_device
/device_lists
traffic for a good sample
@@ -689,43 +689,177 @@ def test_noop_sync_does_not_tightloop(self) -> None: | |||
|
|||
|
|||
class DeviceListSyncTestCase(unittest.HomeserverTestCase): |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Consolidated all of these device_lists
tests spread between test_devices.py
and test_sync.py
to here
tests/rest/client/test_sync.py
Outdated
class DeviceOneTimeKeysSyncTestCase(unittest.HomeserverTestCase): | ||
"""Tests regarding device one time keys (`device_one_time_keys_count`) changes.""" |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Added tests for device_one_time_keys_count
in /sync
because they didn't exist before
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Otherwise I think looks good
Conflicts: synapse/handlers/sync.py
- Added `room.account_data` and `room.presence` to avoid extra work in `_generate_sync_entry_for_rooms()` - Added a comment to the top-level `account_data` and `presence` filters that `(This is just here for good measure)` See #17167 (comment)
As suggested in #17167 (comment)
Thanks for the review @erikjohnston and tips/context @anoadragon453 @kegsay @Hywan @MatMaul 🐧 |
🎉 thanks for your work! |
…t-hq#17167) This is being introduced as part of Sliding Sync but doesn't have any sliding window component. It's just a way to get E2EE events without having to sit through a big initial sync (`/sync` v2). And we can avoid encryption events being backed up by the main sync response or vice-versa. Part of some Sliding Sync simplification/experimentation. See [this discussion](element-hq#17167 (comment)) for why it may not be as useful as we thought. Based on: - matrix-org/matrix-spec-proposals#3575 - matrix-org/matrix-spec-proposals#3885 - matrix-org/matrix-spec-proposals#3884
…t-hq#17167) This is being introduced as part of Sliding Sync but doesn't have any sliding window component. It's just a way to get E2EE events without having to sit through a big initial sync (`/sync` v2). And we can avoid encryption events being backed up by the main sync response or vice-versa. Part of some Sliding Sync simplification/experimentation. See [this discussion](element-hq#17167 (comment)) for why it may not be as useful as we thought. Based on: - matrix-org/matrix-spec-proposals#3575 - matrix-org/matrix-spec-proposals#3885 - matrix-org/matrix-spec-proposals#3884
# Synapse 1.109.0 (2024-06-18) - Add the ability to auto-accept invites on the behalf of users. See the [`auto_accept_invites`](https://element-hq.github.io/synapse/latest/usage/configuration/config_documentation.html#auto-accept-invites) config option for details. ([\#17147](element-hq/synapse#17147)) - Add experimental [MSC3575](matrix-org/matrix-spec-proposals#3575) Sliding Sync `/sync/e2ee` endpoint for to-device messages and device encryption info. ([\#17167](element-hq/synapse#17167)) - Support [MSC3916](matrix-org/matrix-spec-proposals#3916) by adding unstable media endpoints to `/_matrix/client`. ([\#17213](element-hq/synapse#17213)) - Add logging to tasks managed by the task scheduler, showing CPU and database usage. ([\#17219](element-hq/synapse#17219)) # Synapse 1.108.0 (2024-05-28) - Add a feature that allows clients to query the configured federation whitelist. Disabled by default. ([\#16848](element-hq/synapse#16848), [\#17199](element-hq/synapse#17199)) - Add the ability to allow numeric user IDs with a specific prefix when in the CAS flow. Contributed by Aurélien Grimpard. ([\#17098](element-hq/synapse#17098)) Synapse 1.107.0 (2024-05-14) - Add preliminary support for [MSC3823: Account Suspension](matrix-org/matrix-spec-proposals#3823). ([\#17051](element-hq/synapse#17051)) - Declare support for [Matrix v1.10](https://matrix.org/blog/2024/03/22/matrix-v1.10-release/). Contributed by @clokep. ([\#17082](element-hq/synapse#17082)) - Add support for [MSC4115: membership metadata on events](matrix-org/matrix-spec-proposals#4115). ([\#17104](element-hq/synapse#17104), [\#17137](element-hq/synapse#17137)) # Synapse 1.106.0 (2024-04-30) - Send an email if the address is already bound to an user account. ([\#16819](element-hq/synapse#16819)) - Implement the rendezvous mechanism described by [MSC4108](matrix-org/matrix-spec-proposals#4108). ([\#17056](element-hq/synapse#17056)) - Support delegating the rendezvous mechanism described [MSC4108](matrix-org/matrix-spec-proposals#4108) to an external implementation. ([\#17086](element-hq/synapse#17086))
- Fix the building of binary wheels for macOS by switching to macOS 12 CI runners. ([\#17319](element-hq/synapse#17319)) - When rolling back to a previous Synapse version and then forwards again to this release, don't require server operators to manually run SQL. ([\#17305](element-hq/synapse#17305), [\#17309](element-hq/synapse#17309)) - Use the release branch for sytest in release-branch PRs. ([\#17306](element-hq/synapse#17306)) - Fix bug where one-time-keys were not always included in `/sync` response when using workers. Introduced in v1.109.0rc1. ([\#17275](element-hq/synapse#17275)) - Fix bug where `/sync` could get stuck due to edge case in device lists handling. Introduced in v1.109.0rc1. ([\#17292](element-hq/synapse#17292)) - Add the ability to auto-accept invites on the behalf of users. See the [`auto_accept_invites`](https://element-hq.github.io/synapse/latest/usage/configuration/config_documentation.html#auto-accept-invites) config option for details. ([\#17147](element-hq/synapse#17147)) - Add experimental [MSC3575](matrix-org/matrix-spec-proposals#3575) Sliding Sync `/sync/e2ee` endpoint for to-device messages and device encryption info. ([\#17167](element-hq/synapse#17167)) - Support [MSC3916](matrix-org/matrix-spec-proposals#3916) by adding unstable media endpoints to `/_matrix/client`. ([\#17213](element-hq/synapse#17213)) - Add logging to tasks managed by the task scheduler, showing CPU and database usage. ([\#17219](element-hq/synapse#17219)) - Fix deduplicating of membership events to not create unused state groups. ([\#17164](element-hq/synapse#17164)) - Fix bug where duplicate events could be sent down sync when using workers that are overloaded. ([\#17215](element-hq/synapse#17215)) - Ignore attempts to send to-device messages to bad users, to avoid log spam when we try to connect to the bad server. ([\#17240](element-hq/synapse#17240)) - Fix handling of duplicate concurrent uploading of device one-time-keys. ([\#17241](element-hq/synapse#17241)) - Fix reporting of default tags to Sentry, such as worker name. Broke in v1.108.0. ([\#17251](element-hq/synapse#17251)) - Fix bug where typing updates would not be sent when using workers after a restart. ([\#17252](element-hq/synapse#17252)) - Update the LemonLDAP documentation to say that claims should be explicitly included in the returned `id_token`, as Synapse won't request them. ([\#17204](element-hq/synapse#17204)) - Improve DB usage when fetching related events. ([\#17083](element-hq/synapse#17083)) - Log exceptions when failing to auto-join new user according to the `auto_join_rooms` option. ([\#17176](element-hq/synapse#17176)) - Reduce work of calculating outbound device lists updates. ([\#17211](element-hq/synapse#17211)) - Improve performance of calculating device lists changes in `/sync`. ([\#17216](element-hq/synapse#17216)) - Move towards using `MultiWriterIdGenerator` everywhere. ([\#17226](element-hq/synapse#17226)) - Replaces all usages of `StreamIdGenerator` with `MultiWriterIdGenerator`. ([\#17229](element-hq/synapse#17229)) - Change the `allow_unsafe_locale` config option to also apply when setting up new databases. ([\#17238](element-hq/synapse#17238)) - Fix errors in logs about closing incorrect logging contexts when media gets rejected by a module. ([\#17239](element-hq/synapse#17239), [\#17246](element-hq/synapse#17246)) - Clean out invalid destinations from `device_federation_outbox` table. ([\#17242](element-hq/synapse#17242)) - Stop logging errors when receiving invalid User IDs in key querys requests. ([\#17250](element-hq/synapse#17250)) * Bump anyhow from 1.0.83 to 1.0.86. ([\#17220](element-hq/synapse#17220)) * Bump bcrypt from 4.1.2 to 4.1.3. ([\#17224](element-hq/synapse#17224)) * Bump lxml from 5.2.1 to 5.2.2. ([\#17261](element-hq/synapse#17261)) * Bump mypy-zope from 1.0.3 to 1.0.4. ([\#17262](element-hq/synapse#17262)) * Bump phonenumbers from 8.13.35 to 8.13.37. ([\#17235](element-hq/synapse#17235)) * Bump prometheus-client from 0.19.0 to 0.20.0. ([\#17233](element-hq/synapse#17233)) * Bump pyasn1 from 0.5.1 to 0.6.0. ([\#17223](element-hq/synapse#17223)) * Bump pyicu from 2.13 to 2.13.1. ([\#17236](element-hq/synapse#17236)) * Bump pyopenssl from 24.0.0 to 24.1.0. ([\#17234](element-hq/synapse#17234)) * Bump serde from 1.0.201 to 1.0.202. ([\#17221](element-hq/synapse#17221)) * Bump serde from 1.0.202 to 1.0.203. ([\#17232](element-hq/synapse#17232)) * Bump twine from 5.0.0 to 5.1.0. ([\#17225](element-hq/synapse#17225)) * Bump types-psycopg2 from 2.9.21.20240311 to 2.9.21.20240417. ([\#17222](element-hq/synapse#17222)) * Bump types-pyopenssl from 24.0.0.20240311 to 24.1.0.20240425. ([\#17260](element-hq/synapse#17260))
Add Sliding Sync
/sync/e2ee
endpoint for To-Device messages.This is being introduced as part of Sliding Sync but doesn't have any sliding window component. It's just a way to get E2EE events without having to sit through a big initial sync (
/sync
v2). And we can avoid encryption events being backed up by the main sync response or vice-versa.Splitting To-Device messages out to its own endpoint also helps when clients need to have 2 or more sync streams open at a time, e.g a push notification process and a main process. This can cause the two processes to race to fetch the To-Device events, resulting in the need for complex synchronisation rules to ensure the token is correctly and atomically exchanged between processes. (based on the words from MSC3885)
Part of the Sliding Sync simplification. See this discussion below for why it may not be as useful as we thought before implementing.
Based on:
Dev notes
Running relevant tests:
$ SYNAPSE_POSTGRES=1 SYNAPSE_POSTGRES_USER=postgres SYNAPSE_TEST_LOG_LEVEL=INFO poetry run trial tests.rest.client.test_sliding_sync # Or without running with poetry but in the poetry environment $ poetry shell SYNAPSE_POSTGRES=1 SYNAPSE_POSTGRES_USER=postgres SYNAPSE_TEST_LOG_LEVEL=INFO python -m twisted.trial tests.rest.client.test_sync tests.rest.client.test_sliding_sync tests.rest.client.test_sendtodevice $ SYNAPSE_POSTGRES=1 SYNAPSE_POSTGRES_USER=postgres SYNAPSE_TEST_LOG_LEVEL=INFO python -m twisted.trial tests.rest.client.test_sliding_sync
Notes on sharing/inheriting Twisted trial unit tests: #17167 (comment)
Putting quotes around a type is called a forward reference. (mypy)
Pull Request Checklist
EventStore
toEventWorkerStore
.".code blocks
.(run the linters)