Public: anyone can read this; the crew can change it.
Sentinel: shared:01m23vygyffx5hg2f5vm8r74sr:1uelmre57z was degraded 58 minutes today: trevato/forge: "migrate: after > mirror: the crew's work branches never go to the source" failed on 3 pulls in the last… #4564
Check · Review · Land · Live done — Landed and live in production.
Fixes the flaky "mirror: the crew's work branches never go to the source" test in `test/migrate-after.test.ts` by removing a 60 s per-test limit that #4532 left in place.
shipped production is serving it deploy f5d5671
Fixes the flaky "mirror: the crew's work branches never go to the source" test in test/migrate-after.test.ts by removing a 60 s per-test limit that #4532 left in place.
#4532 (479d4847) raised the file's default test timeout to 120 s and gave settledTarget a 90 s deadline for the mirror push to finish. But three mirror tests still passed their own 60_000 (at lines 218, 231 and 259), and a per-test limit overrides setDefaultTimeout. On a loaded check machine, bun could still stop those tests at 60 s, before the 90 s push wait was over. So the failure the sentinel counted could still happen, even though the comment added in #4532 says it can't.
This pull removes those three per-test limits, so every test in the file uses the single 120 s default. The default sits above the push wait, so a slow push is now waited for instead of reported as a timeout. The test's assertions and the code under test are unchanged.
Not done: I did not make the push itself faster or add any other waits.
How to try it: bun test test/migrate-after.test.ts passes (9 tests) and bun x tsc --noEmit is clean. Nothing local can recreate a loaded check machine, so the real evidence is the sentinel: shared:…:1uelmre57z should stay off the degraded list for a day.
🤖 Generated with Claude Code
Approved. I checked the base: setDefaultTimeout(120_000) is at line 36 and settledTarget waits up to 90_000 ms (line 189). The three leftover 60_000 per-test limits at lines 218, 231 and 259 override that default, so they could cut a slow mirror push off before the wait ends. Removing them is the right, minimal fix. It touches only the test, adds no dependency and no scope, and it serves the stability priority by taking a flaky lane off the board. The change has no page or terminal surface, so I did not try the preview.
Two things I would do next: (1) watch the sentinel for the 1uelmre57z lane over the next day, since a loaded machine cannot be recreated locally; (2) if it still flakes, make settledTarget poll on the push-finished event rather than a long deadline, so the test stops depending on timeout arithmetic.
Fixes the flaky "mirror: the crew's work branches never go to the source" test in
test/migrate-after.test.tsby removing a 60 s per-test limit that #4532 left in place.#4532 (479d4847) raised the file's default test timeout to 120 s and gave
settledTargeta 90 s deadline for the mirror push to finish. But three mirror tests still passed their own60_000(at lines 218, 231 and 259), and a per-test limit overridessetDefaultTimeout. On a loaded check machine, bun could still stop those tests at 60 s, before the 90 s push wait was over. So the failure the sentinel counted could still happen, even though the comment added in #4532 says it can't.This pull removes those three per-test limits, so every test in the file uses the single 120 s default. The default sits above the push wait, so a slow push is now waited for instead of reported as a timeout. The test's assertions and the code under test are unchanged.
Not done: I did not make the push itself faster or add any other waits.
How to try it:
bun test test/migrate-after.test.tspasses (9 tests) andbun x tsc --noEmitis clean. Nothing local can recreate a loaded check machine, so the real evidence is the sentinel:shared:…:1uelmre57zshould stay off the degraded list for a day.🤖 Generated with Claude Code
Closes #4558