MotirBuilding in public
MOTIR · moooon
onMotir
You’re viewing a public project. Anyone can view it — no account needed. Sign in to submit, upvote, or comment on requests.View-only — you can’t edit work items
MOTIR-1679

Playwright E2E bulk shards flake: webServer OOM → ERR_CONNECTION_RESET / page.goto timeouts

Done
Description

Symptom

The Playwright E2E bulk shards (bulk-1..4) flake on PRs: a RANDOM, diff-unrelated test fails each run with net::ERR_CONNECTION_RESET at http://localhost:3000 or a page.goto 30s timeout. A different test each run is the tell — on PR #1560: issue-detail-flow.spec.ts:485 (remove blocked-by link), then on re-run issue-detail-flow.spec.ts:569 (create with a link). Both pass on main.

Root cause (hypothesis)

The next dev webServer the E2E lane runs against OOMs under a heavy bulk shard's memory pressure (dev holds a resident webpack compiler + compiles routes on demand) and drops the connection mid-test → whichever test is navigating when the server dies fails. It is the harness dying, not any assertion about the code under test.

Impact

  • Masks real results and forces repeated gh run rerun <id> --failed cycles (PR #1560 needed several).
  • Makes a red check indistinguishable from a real regression — erodes trust that green means green.

Fix directions (investigate)

  • Give the E2E webServer more heap headroom (NODE_OPTIONS=--max-old-space-size=…) and/or a bigger CI runner.
  • Reduce per-shard weight — more, smaller shards (each with its own PG), not fewer heavy ones.
  • Pre-warm the heaviest routes before the shard runs so compile isn't competing with the browser.
  • Consider a production build for shards that seed in-process (no /api/_test routes) — next start is far lighter than next dev.

Related lower-severity flakes seen on PR #1560 (logged; don't chase with code)

  • acceptance-video test 1 — first-hit cold-compile of /items/[id]; the default 20s expect timeout is tight → "Acceptance" heading not found. FIXED in #1560 (bumped that first assertion to 60s; identical-seed test 2 passes warm, proving it's timing).
  • Vitest (integration + coverage) → ProjectRoadmapCanvas.test.tsx — one auto-layout assertion flakes in CI; passes 25/25 locally.
  • bulk-3 (and peers) setup — Installation process exited with code: 100 (pnpm install died); leg dies at setup, 0 tests run (~1m runtime is the tell). Pure infra.

Filed while shipping MOTIR-1665 (PR #1560), where these flakes repeatedly red-lit an otherwise-green diff.