diff options
| author | Matthew Hodgson <matthew@matrix.org> | 2017-01-05 11:44:44 +0000 |
|---|---|---|
| committer | Mark Haines <mark.haines@matrix.org> | 2017-01-05 12:00:11 +0000 |
| commit | 468749c9fca61fabc9dc5da5521ead84b4825783 (patch) | |
| tree | 24230206fed49711294cb8a190525ac4c0818003 | |
| parent | limit total timeout for get_missing_events to 10s (diff) | |
| download | synapse-468749c9fca61fabc9dc5da5521ead84b4825783.tar.xz | |
fix comment
Diffstat (limited to '')
| -rw-r--r-- | synapse/federation/federation_server.py | 18 |
1 files changed, 17 insertions, 1 deletions
diff --git a/synapse/federation/federation_server.py b/synapse/federation/federation_server.py index 6d76e6f917..800f04189f 100644 --- a/synapse/federation/federation_server.py +++ b/synapse/federation/federation_server.py @@ -569,7 +569,23 @@ class FederationServer(FederationBase): ) # XXX: we set timeout to 10s to help workaround - # https://github.com/matrix-org/synapse/issues/1733 + # https://github.com/matrix-org/synapse/issues/1733. + # The reason is to avoid holding the linearizer lock + # whilst processing inbound /send transactions, causing + # FDs to stack up and block other inbound transactions + # which empirically can currently take up to 30 minutes. + # + # N.B. this explicitly disables retry attempts. + # + # N.B. this also increases our chances of falling back to + # fetching fresh state for the room if the missing event + # can't be found, which slightly reduces our security. + # it may also increase our DAG extremity count for the room, + # causing additional state resolution? See #1760. + # However, fetching state doesn't hold the linearizer lock + # apparently. + # + # see https://github.com/matrix-org/synapse/pull/1744 missing_events = yield self.get_missing_events( origin, |