Rotorsoft / act-root / 36345636636

100%

push

github

fix(act): park a no-backoff retry until its own lease lapses (#1681)

* fix(act): park a no-backoff retry until its own lease lapses

A no-progress failure submits no ack, so the lease stays held, and every
adapter's claim excludes a live-leased stream even from its holder. Any
drain issued inside that window therefore claimed nothing, which the
empty-claim short-circuit read as "fully caught up" and used to clear the
armed flag the error path had deliberately kept set. One ordinary
settle() was enough: its next pass drains inside the 10s default lease,
so the stream got exactly one attempt, was never retried, and never
reached its blockOnError verdict — so it never appeared in
blocked_streams() either.

A failing, non-blocked stream now parks in the defer timer at its backoff
time, or at the lease expiry when no backoff is configured. The timer's
wake re-arms drain, which is the mechanism the backoff path already used;
the two error paths stop diverging.

HandleResult.next_attempt_at already documented "undefined means no
backoff configured — drain re-attempts as soon as the lease expires".
That sentence is now true.

Closes #1670

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

* docs(act): correct the no-backoff retry timing and pin it

error-handling.md said a no-backoff retry re-claims "on the next drain
cycle — typically within milliseconds", which was only ever true with a
very short lease: the failed cycle keeps its lease, so leaseMillis sets
the floor. Adds the behavior-contract row for the retry-and-block path.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: rotorsoft <rotorsoft@outlook.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>

4357 of 4357 branches covered (100.0%)

Branch coverage included in aggregate %.

2 of 2 new or added lines in 1 file covered. (100.0%)

9420 of 9420 relevant lines covered (100.0%)

1192.94 hits per line