MarkUsProject / Markus / 41502651755

91%

push

github

Configured ActionCable Redis reconnection retries (#8244)

When Redis restarted, every MarkUs web process lost its pub/sub
subscription and never got it back: ActionCable's Redis adapter
defaults to a single, immediate reconnect attempt, which fails while
Redis is still starting, after which the listener thread exits for the
lifetime of the process. Realtime updates (submission collection,
autotest results, PDF splitting, group creation) then silently stop
until MarkUs itself is restarted.

Set reconnect_attempts to a list of delays so the listener keeps
retrying for roughly 19 minutes (enough to cover a host reboot) and
resubscribes to all channels once Redis is back.

ActionCable passes the same option to the connection it publishes
broadcasts on, which would make a broadcast issued during an outage
block the calling thread for the whole window. Override the adapter's
redis_connector so publishing connections keep a single quick retry and
fail fast, while the listener keeps the long schedule.

Co-authored-by: Donny Wong <donnywong@cs.toronto.edu>
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>

1292 of 2569 branches covered (50.29%)

Branch coverage included in aggregate %.

12 of 12 new or added lines in 2 files covered. (100.0%)

50618 of 54186 relevant lines covered (93.42%)

129.39 hits per line