supabase / supabase-swift / 31604397264

85%

Build Type

push

github

Committed by web-flow

Commit Message

fix(realtime): retry auto-reconnect with backoff instead of giving up after one attempt (#1198)

* fix(realtime): retry auto-reconnect with backoff instead of giving up after one attempt

ConnectionManager.initiateReconnect scheduled exactly one reconnect
attempt. If that attempt failed (network still down, the common case
during a real outage), the client stayed parked in .disconnected
forever with nothing scheduling another try.

initiateReconnect now retries with exponential backoff (capped at 30s)
until a connection succeeds or disconnect()/deinit cancels the task.
State stays .reconnecting across every attempt instead of bouncing
through .disconnected between attempts, so RealtimeClientV2's
reconnect-completion latch isn't misled into treating the eventual
success as a fresh connect() and skipping rejoinChannels().

Fixes #1147

* fix(realtime): don't retain ConnectionManager across the reconnect backoff sleep

The retry loop's `guard let self` sat before `clock.sleep`, so `self`
was held strongly for the whole backoff wait, not just the connection
attempt. A client released mid-outage would keep the actor (and
everything it captures) alive for up to the current attempt's delay.
deinit also never cancelled a pending `.reconnecting` task.

Capture `clock`/`reconnectDelay`/`logger` by value so the sleep needs
no strong self, reacquiring self only right before the connect
attempt, and cancel the task in deinit if state is still .reconnecting.

Coverage Stats

43 of 49 new or added lines in 1 file covered. (87.76%)

2 existing lines in 2 files now uncovered.

8848 of 10462 relevant lines covered (84.57%)

39.88 hits per line