Currently, the second join will block until Synapse finishes syncing state. While not incorrect, it's not a great experience and we should be able to allow some membership changes before we've finished syncing state.

I've left a WIP complement test at https://github.com/matrix-org/complement/tree/squah/faster_room_joins_handle_second_join_while_resyncing

Synapse blocks in update_membership_locked:

  File "/usr/local/lib/python3.9/site-packages/synapse/http/server.py", line 366, in _async_render_wrapper
    callback_return = await self._async_render(request)
  File "/usr/local/lib/python3.9/site-packages/synapse/http/server.py", line 572, in _async_render                                                                                                                                               callback_return = await raw_callback_return
  File "/usr/local/lib/python3.9/site-packages/synapse/rest/client/room.py", line 335, in on_POST
    await self.room_member_handler.update_membership(                                                                                                                                                                                          File "/usr/local/lib/python3.9/site-packages/synapse/handlers/room_member.py", line 524, in update_membership
    result = await self.update_membership_locked(
  File "/usr/local/lib/python3.9/site-packages/synapse/handlers/room_member.py", line 735, in update_membership_locked
    current_state_ids = await self.state_handler.get_current_state_ids(                                                                                                                                                                        File "/usr/local/lib/python3.9/site-packages/synapse/state/__init__.py", line 215, in get_current_state_ids
    ret = await self.resolve_state_groups_for_events(room_id, latest_event_ids)
  File "/usr/local/lib/python3.9/site-packages/synapse/util/metrics.py", line 113, in measured_func
    r = await func(self, *args, **kwargs)
  File "/usr/local/lib/python3.9/site-packages/synapse/state/__init__.py", line 423, in resolve_state_groups_for_events
    state_groups = await self._state_storage_controller.get_state_group_for_events(
  File "/usr/local/lib/python3.9/site-packages/synapse/storage/controllers/state.py", line 338, in get_state_group_for_events
    await self._partial_state_events_tracker.await_full_state(event_ids)
  File "/usr/local/lib/python3.9/site-packages/synapse/storage/util/partial_state_events_tracker.py", line 78, in await_full_state
    logger.info(
2022-06-06 17:14:28,273 - synapse.storage.util.partial_state_events_tracker - 78 - INFO - GET-8 - Awaiting un-partial-stating of events ['$-2iMGMrGcncRdZu8ZdY8GwzR2u0WPiFBfCwPLO7O_3Y']

update_membership_locked is a large-ish method that handles all membership changes and not just joins. So we have to consider the impact of using partial state for each of the membership transitions.

Uh oh!

Faster joins: Handle a second join while syncing state for the first one #12801

Description

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions