diff options
author | Erik Johnston <erik@matrix.org> | 2016-02-12 11:13:06 +0000 |
---|---|---|
committer | Erik Johnston <erik@matrix.org> | 2016-02-12 11:13:06 +0000 |
commit | 4d54d87c3e640819a5e1707f58529af8e1c57c7d (patch) | |
tree | 1d984b273d068180ad9727a4ffdb28104dcdca97 /synapse/handlers/sync.py | |
parent | Merge branch 'release-v0.13.3' of github.com:matrix-org/synapse (diff) | |
download | synapse-4d54d87c3e640819a5e1707f58529af8e1c57c7d.tar.xz |
Mitigate against incorrect old state in /sync.
There have been reports from the field that servers occasionally have incorrect notions of what the old state of a room is. This proves problematic as /sync relies on a correct old state. This patch makes /sync specifically include in the 'state' key any current state events that haven't been correctly included.
Diffstat (limited to 'synapse/handlers/sync.py')
-rw-r--r-- | synapse/handlers/sync.py | 17 |
1 files changed, 12 insertions, 5 deletions
diff --git a/synapse/handlers/sync.py b/synapse/handlers/sync.py index 1d0f0058a2..b2b9b928c9 100644 --- a/synapse/handlers/sync.py +++ b/synapse/handlers/sync.py @@ -823,15 +823,17 @@ class SyncHandler(BaseHandler): # TODO(mjark) Check for new redactions in the state events. with Measure(self.clock, "compute_state_delta"): + current_state = yield self.get_state_at( + room_id, stream_position=now_token + ) + if full_state: if batch: state = yield self.store.get_state_for_event( batch.events[0].event_id ) else: - state = yield self.get_state_at( - room_id, stream_position=now_token - ) + state = current_state timeline_state = { (event.type, event.state_key): event @@ -842,6 +844,7 @@ class SyncHandler(BaseHandler): timeline_contains=timeline_state, timeline_start=state, previous={}, + current=current_state, ) elif batch.limited: state_at_previous_sync = yield self.get_state_at( @@ -861,6 +864,7 @@ class SyncHandler(BaseHandler): timeline_contains=timeline_state, timeline_start=state_at_timeline_start, previous=state_at_previous_sync, + current=current_state, ) else: state = {} @@ -920,7 +924,7 @@ def _action_has_highlight(actions): return False -def _calculate_state(timeline_contains, timeline_start, previous): +def _calculate_state(timeline_contains, timeline_start, previous, current): """Works out what state to include in a sync response. Args: @@ -928,6 +932,7 @@ def _calculate_state(timeline_contains, timeline_start, previous): timeline_start (dict): state at the start of the timeline previous (dict): state at the end of the previous sync (or empty dict if this is an initial sync) + current (dict): state at the end of the timeline Returns: dict @@ -938,14 +943,16 @@ def _calculate_state(timeline_contains, timeline_start, previous): timeline_contains.values(), previous.values(), timeline_start.values(), + current.values(), ) } + c_ids = set(e.event_id for e in current.values()) tc_ids = set(e.event_id for e in timeline_contains.values()) p_ids = set(e.event_id for e in previous.values()) ts_ids = set(e.event_id for e in timeline_start.values()) - state_ids = (ts_ids - p_ids) - tc_ids + state_ids = ((c_ids | ts_ids) - p_ids) - tc_ids evs = (event_id_to_state[e] for e in state_ids) return { |