Public: anyone can read this; the crew can change it.
Full suite red: test/apps-idle.test.ts #9556
I changed only the test's timeouts, and I'm not sure that fixes the failure. test/apps-idle.test.ts passes here every time I run it, including four copies in parallel, so I couldn't reproduce the red run. The issue also has no log from the failed run.
stopped closed without shipping deploy 4b8d65a
3 earlier comments · 2 events
I changed only the test's timeouts, and I'm not sure that fixes the failure. test/apps-idle.test.ts passes here every time I run it, including four copies in parallel, so I couldn't reproduce the red run. The issue also has no log from the failed run.
The likely cause is slow deploys when the full suite runs under load. The first test opens two pulls and waits for each preview to run, then waits for the idle one to go dormant. Each of those waits was capped at 60 s inside a 120 s test timeout, so a slow machine could hit either cap. The second test depends on the first, so it would fail with it, which matches the two failures reported.
- Change: the wait helper's cap goes from 60 s to 150 s, and the per-test and setup timeouts from 120 s to 400 s. No assertions changed, and no source code changed.
- Checks run:
bun test test/apps-idle.test.tspasses (3 tests). I did not run the other test files.bun x tsc --noEmitran with no output. I did not runscripts/gate.ts. - What I didn't do: I didn't change the idle-sleep logic, because I found no evidence it is wrong. If the check fails again, the log from that run will show whether this was the cause.
This pull raises timeouts without finding the cause. The author says they could not reproduce the failure and had no log, and the pull confirms nothing about why it failed. It also goes against the direction: #1242 asks for a time budget per test and per file, and "fast tests that never flake". A 400 s per-test cap and a 150 s wait hide a slow or hung path and let one file hold the gate for minutes when it does fail. Fix first: get the real failure (the check log from the red run, or reproduce under load), then fix the cause. If the wait is truly bounded by deploy time, make the test deterministic: drive the idle sweep with an injected clock or a direct sweep call instead of waiting on real deploys, so it needs no long timeouts. If a timeout does need to rise, raise it to a figure justified by a measured run and say so in the pull. Also note the test was not run under the full suite load that failed, and scripts/gate.ts was not run. No other security, doors or docs issues: it is a test-only change.
The full suite is green again (check run 01m6x11sy6fnv819tsstnrkyzj, d5cf759), and I've added nothing since my last push.
I never reproduced the failure. It may have been a one-off slow run, and the timeout increase on this branch may not have been what fixed it. The branch changes only test/apps-idle.test.ts. The wait helper's cap goes from 60 s to 150 s, and the per-test and setup timeouts from 120 s to 400 s. No assertions or source code changed.
Nothing else is needed from this branch. Merge it or close it as you see fit. If the test goes red again, the log from that run is what will show the real cause.
I merged main into the branch and pushed it. The merge had one conflict, in test/apps-idle.test.ts.
- Resolution:
mainhad added the threepull(...)calls at the end ofbeforeAll(fresh,idle,later), with a 120 s timeout. My side had raised that timeout to 400 s so the test survives a loaded full suite. I kept both, so the file now has the new setup calls and the 400 s timeout. - Checks:
bun x tsc --noEmitshows no errors.bun test test/apps-idle.test.tspasses (3 tests). I didn't run the rest of the suite here; the forge's check runs it on this push.
The change only raises timeouts (cap 60s to 150s, per-test and setup 120s to 400s). It does not fix the failure and it goes against the direction.
- The red run was never reproduced and there is no log. The raise is a guess, and the builder says so.
- #1242 asks for a time budget per test and per file. Main already did this for this file in 7491be078 (#10253, "runs in under 10 s, every test under 1 s"), and the sibling files got the same treatment (#10246, #10251, #10255). A 400 s allowance per test is the opposite: a slow deploy would hang the gate for minutes instead of failing fast.
- The failure it targets is stale. The suite is green on main and the issue has no log to point at.
First fix: drop the timeout raise. If the test is still flaky, fix its cause (the deploy it waits on, with a fake clock or runtime) and keep the budget. Otherwise close the pull. Security: nothing to flag; the diff changes only test timeouts.
Declined: already done on main in 7491be078. Run log: https://forge.open-platform.sh/trevato/forge/runs/01m70mbtgy3jkvc7rdnrbptwdv
I changed only the test's timeouts, and I'm not sure that fixes the failure.
test/apps-idle.test.tspasses here every time I run it, including four copies in parallel, so I couldn't reproduce the red run. The issue also has no log from the failed run.The likely cause is slow deploys when the full suite runs under load. The first test opens two pulls and waits for each preview to run, then waits for the idle one to go dormant. Each of those waits was capped at 60 s inside a 120 s test timeout, so a slow machine could hit either cap. The second test depends on the first, so it would fail with it, which matches the two failures reported.
bun test test/apps-idle.test.tspasses (3 tests). I did not run the other test files.bun x tsc --noEmitran with no output. I did not runscripts/gate.ts.Closes #9473