The Windows worker-idle-timeout regression test is intermittently red on otherwise passing release follow-up CI.
Evidence:
The test uses a 150 ms idle budget and 400 ms wall budget. Its second task goes through MockExecutor, which emits several persisted events and sleeps for 50 ms. Scheduling/storage contention is a plausible explanation; the failed record’s terminal reason was not printed, so this is not yet a confirmed root cause.
For v0.9.13: include the full terminal record in failure diagnostics, reproduce under contention on Windows, and make the worker-reuse test deterministic without weakening production timeouts or removing the proof that a forcibly timed-out task releases its worker. Preserve the rule that failures are visible rather than silently retried.
The Windows worker-idle-timeout regression test is intermittently red on otherwise passing release follow-up CI.
Evidence:
task_manager::tests::forced_idle_timeout_releases_the_worker_for_later_tasksexpected the second task to be Completed but observed Failed (task_manager.rs:4114).The test uses a 150 ms idle budget and 400 ms wall budget. Its second task goes through MockExecutor, which emits several persisted events and sleeps for 50 ms. Scheduling/storage contention is a plausible explanation; the failed record’s terminal reason was not printed, so this is not yet a confirmed root cause.
For v0.9.13: include the full terminal record in failure diagnostics, reproduce under contention on Windows, and make the worker-reuse test deterministic without weakening production timeouts or removing the proof that a forcibly timed-out task releases its worker. Preserve the rule that failures are visible rather than silently retried.