Two e2e runs failed with ✗ Server failed to start after 180 seconds (the wrangler pages dev startup loop in .github/actions/e2e-env), cascading into total suite failure:
Both PRs auto-merged over the red e2e because e2e is not a required check (process now fixed on the agent side; ruleset hardening under discussion). Main's subsequent e2e runs are green, so the failures were infra, not code.
Investigate: wrangler startup logs are dumped on failure (the action cats /tmp/wrangler.log) — pull both runs' logs and diff; likely suspects are npm/registry latency during wrangler pages dev boot, D1 local init contention, or the 180s budget simply being too tight on slow runners. Consider: longer budget + earlier log surfacing, retry-once on startup failure (NOT on test failure — never mask real reds), and/or pre-warming wrangler in the composite action.
Two e2e runs failed with
✗ Server failed to start after 180 seconds(the wrangler pages dev startup loop in .github/actions/e2e-env), cascading into total suite failure:Both PRs auto-merged over the red e2e because e2e is not a required check (process now fixed on the agent side; ruleset hardening under discussion). Main's subsequent e2e runs are green, so the failures were infra, not code.
Investigate: wrangler startup logs are dumped on failure (the action cats /tmp/wrangler.log) — pull both runs' logs and diff; likely suspects are npm/registry latency during
wrangler pages devboot, D1 local init contention, or the 180s budget simply being too tight on slow runners. Consider: longer budget + earlier log surfacing, retry-once on startup failure (NOT on test failure — never mask real reds), and/or pre-warming wrangler in the composite action.