pneumo

M5: Tangled CI for the e2e lanes (ENG-278) #29

Merged

demi.dev 1w ago

targeting main from stoopkidddd/eng-278-m5-tangled-ci-for-the-e2e-lanes-server-web-docker

Merged opened by and assigned to

demi.dev 1w ago targeting main from stoopkidddd/eng-278-m5-tangled-ci-for-the-e2e-lanes-server-web-docker

Linear: https://linear.app/engineereddev/issue/ENG-278

What's in this PR (3 files):

.tangled/workflows/e2e.yml (nixery): one workflow sharing the Nix and pnpm install --frozen-lockfile setup, then e2e:server with native MinIO (PNEUMO_E2E_MINIO=native) and e2e:web with Playwright 1.64.1 Chromium from a pinned nixpkgs revision (Chromium and headless shell only, PLAYWRIGHT_BROWSERS_PATH), --workers=2. PR trigger unfiltered (hosted quirk); push trigger path-filtered. pnpm store cached.
MinIO for the test fixture comes from an archived nixpkgs pin, because unstable refuses the abandoned package; loopback-only, documented.
scripts/ci/run-nixery.sh: writes a minimal /etc/passwd when missing (server subprocesses without HOME still call os.homedir() via the AWS SDK).
docs/agents/tangled.md: CI coverage table rows for e2e server/web (CI), e2e docker and e2e android (local-only, with reasons), plus the timing budget.

Verify (implementer, local): native e2e:server 17 passed; native e2e:web 42 passed (also with CI=true --workers=2); the exact workflow commands on native Linux arm64 with Nix passed (server 11 s, web 57 s); e2e:docker cold amd64 270 s (why it stays local-only). Lead: mise run check:infra exit 0 (4 workflows valid).

Hosted verification is this PR's acceptance: the lead reads this PR's pipeline (cold and warm) and updates the coverage table with measured durations before review.

🤖 Generated with Claude Code

Commits 0

Could not load commits.

Comments

v0 1w ago

v1 1w ago

demi.dev 1w ago

pneumo-reviewer: round 1, APPROVE #

One nixery workflow, e2e.yml, runs e2e:server and e2e:web with native MinIO, Chromium 1248 from a pinned nixpkgs revision and its own pnpm cache; run-nixery.sh writes a minimal /etc/passwd when missing; docs/agents/tangled.md covers all 10 lanes. Nothing must change before merge.

Workflow valid; every step wrapped; no secrets; PR trigger unfiltered, push paths cover everything the suites read.
Playwright pin verified against upstream nixpkgs (1.64.1, Chromium/headless shell 1248; link names and FONTCONFIG_FILE correct; no silent download fallback).
MinIO pin verified (nixos-25.05 ac62194c…, matches the hosted --version); loopback test fixture only, limitation documented, nothing in production uses it.
/etc/passwd change is guarded and shellcheck-clean; the other workflows passed twice with it.
Local: check:infra exit 0; native e2e:server 17 passed; CI=true native e2e:web --workers=2 42 passed.

FILE (score < 5, recorded on ENG-275)

score=4 — timeout headroom: the doc's "at least 5 m 35 s" bound is out of date (PR #29's cold pipeline shows ≥ 6 m 18 s), and it doesn't say that a timeout status most likely means a slow cold run and a retry is the first response.
score=3 — the ts.yml step timings in the doc are e2e.yml's (lead's misread). Real ts.yml: cold Nix install 33 s, pnpm 15 s, protocol 10 s, server 18 s, web 9 s; warm 26 s, 7 s, 12 s, 10 s, 11 s. Only e2e.yml pays ~1 m for Nix (the Chromium closure).
score=3 — run-nixery.sh puts every *-gcc-*-lib/lib on LD_LIBRARY_PATH in hash order; with the Chromium pin a future GCC major in unstable could surface an older libstdc++ first.

Hosted: d2a8dc3 cold 4/4 (6 m 18 s) and warm 4/4 (4 m 43 s); c513062 4/4 (5 m 5 s).

Versions

pneumo-reviewer: round 1, APPROVE #