From e5d7bda11576a9aa4cc7a6ecd5ca815d6bc88eda Mon Sep 17 00:00:00 2001 From: Zoltan Csizmadia Date: Sun, 20 Sep 2026 09:25:59 -0500 Subject: [PATCH] ci: turn the nightly e2e schedule off until #429 is fixed Tracked by #448, which exists so this does not quietly stay off. Two nights, two red runs, both on #429 -- a known and triaged bug in `restart --supervisor`. A nightly that is EXPECTED to be red is a nightly nobody reads, and the next real regression would have arrived as one more red run in a row of them. That is worse than no nightly, because it looks like coverage. workflow_dispatch is untouched, and RELEASING.md still requires a run on the release commit -- which is the moment it has to be true. Worth recording that it earned its keep before being switched off. In three runs it found #444 (install failing on a profile with no Run key, which no unit test could catch because the test helper created the key the product assumed), the audit flag-parsing bug in #428 within 45 seconds of its first run, and #429 itself. The alternative was skipping the one stage with a link to #429. I did not take it: this release spent a lot of effort removing exactly that pattern, and a skipped stage hides a real bug where a disabled schedule with a tracking issue does not. --- .github/workflows/e2e.yml | 20 +++++++++++++++----- 1 file changed, 15 insertions(+), 5 deletions(-) diff --git a/.github/workflows/e2e.yml b/.github/workflows/e2e.yml index be231fd..f2ad1f7 100644 --- a/.github/workflows/e2e.yml +++ b/.github/workflows/e2e.yml @@ -13,19 +13,29 @@ name: e2e # # Why not a PR gate: it provisions a real WSL2 distro, downloads the published # rootfs and runs containers -- minutes, not seconds, and it mutates the runner. -# Nightly plus manual is the honest trade: RELEASING.md requires a green run on -# the release commit, which is the moment it actually has to be true. +# Manual only for now — see the note on the trigger below. RELEASING.md +# requires a run on the release commit, which is the moment it has to be true. # # The platform excuse is retired and proven retired. PLAN.md said hosted # `windows-latest` exposes nested virtualization; the sibling setup-skrog repo # installs the engine and runs `docker run --rm hello-world` on that runner on # every push. No self-hosted runner is needed on x64. +# The nightly schedule is DISABLED, deliberately and temporarily (#448). +# +# It ran for two nights and both were red on #429 — a known, triaged bug in +# `restart --supervisor`. A nightly that is expected to be red is a nightly +# nobody reads, and the next real regression would have arrived as one more +# red run in a row of them. That is worse than no nightly, because it looks +# like coverage. +# +# Manual dispatch is unaffected, and RELEASING.md still requires a run on the +# release commit — which is the moment it actually has to be true. +# +# Re-enable when #429 is fixed. #448 tracks that, and exists so this does not +# quietly stay off. on: workflow_dispatch: - schedule: - # 04:20 UTC daily. Off-peak, and far from the release.yml window. - - cron: '20 4 * * *' permissions: contents: read