Repository navigation
CI: timestamp the browser lane's test server lines and explain the empty fork-PR failures #897
Description
Split from #552: main already prints per-script timings in scripts/ci-browser.sh and keeps screenshots from the browser lane on failure (#627).
Remaining
- Timestamps on the test servers' own lines. The web UI servers that the browser scripts start log to the lane's output. Their lines are mixed with the PASS lines and carry no timestamps of their own. Add timestamps so a server line can be placed against a PASS line. The frames the page received are also not kept today; herdr's
test-server.logis the only server log uploaded on failure. - Empty CI failures of fork PRs. Seven CI runs of fork PRs on 2026-10-06 ended as
failurewith no jobs and no log: 37455493445, 37452965707, 37443957584, 37437006600, 37436949778, 37411882568, 37411593274. They are not test failures. The likely cause is an approval-gated run that was closed or superseded, but the run metadata does not say. Look into them once, so the CI workflow's failure count can be read without subtracting them by hand.
No retries: a failure that a retry turns green is the kind this is meant to explain.
Activity
ci: timestamp the browser lane's test server lines (#1789)
devswha commentedon Oct 23, 2026
Looked into the second item (the empty CI failures of fork PRs). The first item, timestamps on the browser lane's lines, merged in #1789.
What the seven runs are. Each is a
pull_requestrun of an earlier commit of a fork PR that never started: 0 jobs, a check suite with 0 check runs,run_attempt1. GitHub concluded itfailurein the same second the PR was closed or merged, long after it was created:What that means. These runs sat waiting, most likely for the approval GitHub requires before a fork PR's workflow runs. The PR's later commits each got their own run, which ran and passed (#515: 37459272358 success; #514: 37459746193 and 37461135205 success; #504: 37430165753 success). When the PR closed, GitHub ended the old waiting runs as
failure.What it is not. It is not a broken workflow file: actionlint passes on
ci.ymlat those runs' commits (26e5896, f74042f, 8188793). It is not a test failure either: no job ever started.So when the CI workflow's failures are counted, a
pull_requestrun withconclusion: failureand 0 jobs, finished at its PR's close time, is one of these and can be left out. It still happens: #1380's run above is from 2026-10-19.devswha commentedon Oct 25, 2026
Fixed by #1789 (96eff68): the browser lane now timestamps its test server's output (
scripts/ci-browser-parallel.ts,scripts/ci-browser.sh), and the empty fork-PR failures are explained in the comment above (zero jobs, concluded when the PR closed). Shipped in v0.4.14.