You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Some data on the hangs, from CI runs on two unrelated PRs of mine (#1531 touches packages/project, #1534 touches ws-worker — neither goes near the integration suite).
The worker integration job fails by hanging rather than erroring, and it's nondeterministic across runs:
build 18318: 1 test pending in test/autoinstall.test.ts
build 18317: 15 tests pending in test/integration.test.ts
build 18256: 3 tests pending in test/autoinstall.test.ts
48 tests pass each time; a handful just never resolve and ava times out. In-repo PRs #1527 and #1530 fail the same job, while #1528, #1529, #1532 and #1533 pass, so it isn't fork-specific.
A likely mechanism, at least for these files: the run() helper wraps everything in a promise that only settles on a run:complete event —
reject is destructured but never called, and there's no setTimeout anywhere in either file (grep -c setTimeout autoinstall.test.ts → 0). So if a run never completes — and autoinstall waits on real npm installs of @openfn/language-*@version — nothing settles the promise and the test sits pending until ava's own timeout kills the process. That would also explain why the count varies with network speed.
Not sending a PR for it: adding timeouts across the integration harness overlaps whatever is on your fix-ava branch, and I can't verify a fix locally without reproducing the flake itself. Happy to take it on if you'd like it separated from the ava upgrade.
Activity
CaptainAni187 commentedon Sep 8, 2026
Some data on the hangs, from CI runs on two unrelated PRs of mine (#1531 touches
packages/project, #1534 touchesws-worker— neither goes near the integration suite).The worker integration job fails by hanging rather than erroring, and it's nondeterministic across runs:
48 tests pass each time; a handful just never resolve and ava times out. In-repo PRs #1527 and #1530 fail the same job, while #1528, #1529, #1532 and #1533 pass, so it isn't fork-specific.
A likely mechanism, at least for these files: the
run()helper wraps everything in a promise that only settles on arun:completeevent —rejectis destructured but never called, and there's nosetTimeoutanywhere in either file (grep -c setTimeout autoinstall.test.ts→ 0). So if a run never completes — andautoinstallwaits on real npm installs of@openfn/language-*@version— nothing settles the promise and the test sits pending until ava's own timeout kills the process. That would also explain why the count varies with network speed.Not sending a PR for it: adding timeouts across the integration harness overlaps whatever is on your
fix-avabranch, and I can't verify a fix locally without reproducing the flake itself. Happy to take it on if you'd like it separated from the ava upgrade.