Complement Postgres & Complement workers: de-flaking

[reivilibre]

The Postgres and workers Complement runs, which have recently been enabled, are a bit flakey at times.

Let's track the list of flakey tests here and get them cleaned up.

• TestFederationRoomsInvite/Parallel/Invited_user_can_reject_invite_over_federation_for_empty_room
• TestRestrictedRoomsLocalJoinInMSC3787Room/Join_should_fail_when_left_allowed_room and TestRestrictedRoomsLocalJoin}/Join_should_fail_when_left_allowed_room: https://github.com/matrix-org/synapse/runs/7179299458?check_suite_focus=true
• TestRestrictedRoomsLocalJoin/Join_should_succeed_when_joined_to_allowed_room and TestRestrictedRoomsLocalJoinInMSC3787Room/Join_should_succeed_when_joined_to_allowed_room: https://github.com/matrix-org/synapse/runs/7179299458?check_suite_focus=true
  • Race condition where a second join races with cache invalidation over replication and erroneously rejects joining Restricted Room #13185
  • Deflake Join_should_succeed_when_joined_to_allowed_room subtest of restricted rooms tests complement#413

Once all this is done, we can close #13264.

[squahtx]

(This comment referred to an MSC3787 tests. I've moved this to the top-level one with the other restricted rooms failures, as the underlying cause is the same. -- dmr.)

[DMRobertson]

• Complement Existing members see new members' presence is flaky #13199

[DMRobertson]

• TestWriteMDirectAccountData, see Complement TestWriteMDirectAccountData is flakey #13334

[DMRobertson]

• TestSendJoinPartialStateResponse: see TestSendJoinPartialStateResponse is flaky #13256

[squahtx]

• TestSync/parallel/sync_should_succeed_even_if_the_sync_token_points_to_a_redaction_of_an_unknown_event
  • https://github.com/matrix-org/synapse/runs/7197178106?check_suite_focus=true#step:4:6887
  • https://github.com/matrix-org/synapse/runs/7435035090?check_suite_focus=true (This particular case should be solved by Fix an out of bounds error in TestSync 'sync should succeed even if the sync token points to a redaction of an unknown event' complement#427)

[DMRobertson]

• TestKnockRoomsInPublicRoomsDirectory https://github.com/matrix-org/synapse/runs/7300817542?check_suite_focus=true
  • see PR Retry /publicRooms a few times to allow for propagation delay before a new room appears in the room directory complement#415

[clokep]

#13266 made it so that complement + workers won't stop us from merging PRs, we should revert this PR once these builds are no longer flakey.

[DMRobertson]

• TestRestrictedRoomsLocalJoinInMSC3787Room/Join_should_succeed_when_invited
  • caused by _should_perform_remote_join inspects the current state of the room, not the state of the would-be join #13269
  • some investigation in Deflake Join_should_succeed_when_joined_to_allowed_room subtest of restricted rooms tests complement#413

[DMRobertson]

• TestRoomState/Parallel/GET_/publicRooms_lists_newly-created_room
  • https://github.com/matrix-org/synapse/runs/7337130834?check_suite_focus=true
  • https://github.com/matrix-org/synapse/runs/7406999068?check_suite_focus=true (postgres monolith)
  • see PR at Retry /publicRooms a few times to allow for propagation delay before a new room appears in the room directory complement#415

[DMRobertson]

• TestRestrictedRoomsRemoteJoinFailOverInMSC3787Room
  • https://github.com/matrix-org/synapse/runs/7397381295?check_suite_focus=true (postgres monolith)
    • see Faster joins: improve handling of race where we can fail to persist a client event with partial state after _sync_partial_state_room clears the partial state flag for a room #13173 (comment) (429 due to partial state conflict)

[reivilibre]

• TestImportHistoricalMessages/parallel/Federation/When_messages_have_already_been_scrolled_back_through,_new_historical_messages_are_visible_in_next_scroll_back_on_federated_server https://github.com/matrix-org/synapse/runs/7609443787?check_suite_focus=true

[reivilibre]

Superseded by:

• Complement TestImportHistoricalMessages/parallel/Federation/When_messages_have_already_been_scrolled_back_through,_new_historical_messages_are_visible_in_next_scroll_back_on_federated_server is potentially flaky. #13508
• Faster joins: improve handling of race where we can fail to persist a client event with partial state after _sync_partial_state_room clears the partial state flag for a room #13173 (comment)
• Complement TestWriteMDirectAccountData is flakey #13334