Opus finding
Abort signal listener on retry backoff is never removed, leaking listeners across attempts
mediumbughigh
- src/providers/orchestrator.ts:285-296
When the timer fires normally (no abort), the 'abort' listener installed with { once: true } is NOT removed because it was never invoked. Across multiple retry attempts on a long-lived signal (e.g. options.signal passed by caller), each backoff adds a new dangling listener. Over many calls in a long-running orchestrator, this accumulates listeners on the user-supplied AbortSignal (which may also be the watchdog signal). Even with once:true, the listener still holds a closure over timer/resolve until garbage-collected with the signal. If the signal aborts later, all stale listeners fire and call resolve() (harmless) and clearTimeout (harmless), but Node will emit MaxListenersExceededWarning after 10 attempts on the same signal.
Recommendation
Track the listener and removeEventListener on the timer's resolve path, or use AbortSignal.timeout / a helper that wires/unwires both directions.