Source: PR #55 R2 made this test's deadline failures loud (by design); the flakiness itself persists — it recurred on PR #59's first macOS CI run (green on --failed rerun) and was unreproducible locally across clean runs during PR #57 review.
The SIGKILL crash-recovery test in tests/lib/extraction-semaphore.test.ts intermittently exceeds its 10s worker-startup poll window under CI load.
Fix sketch: investigate worker spawn latency under CI load; consider a longer env-aware deadline or a startup handshake instead of DB polling. The loud-failure diagnostics from PR #55 (stderr capture) are the starting evidence.
Source: PR #55 R2 made this test's deadline failures loud (by design); the flakiness itself persists — it recurred on PR #59's first macOS CI run (green on --failed rerun) and was unreproducible locally across clean runs during PR #57 review.
The SIGKILL crash-recovery test in
tests/lib/extraction-semaphore.test.tsintermittently exceeds its 10s worker-startup poll window under CI load.Fix sketch: investigate worker spawn latency under CI load; consider a longer env-aware deadline or a startup handshake instead of DB polling. The loud-failure diagnostics from PR #55 (stderr capture) are the starting evidence.