fix(ci): unblock the Playwright install and move the landing suite's port out of the ephemeral range

[AminDhouib]

Two independent CI fixes. The first unblocks every Playwright job in the repo; the second moves one more fixed listen port below the Linux ephemeral range. The storybook / cross-framework half of the port sweep is deliberately NOT here — see "Not done" below; it needs a decision that belongs to the maintainer, not to this PR.

1 — drop Google's chrome-stable apt source before playwright install

Every Playwright job currently dies before it runs a test. Nightly run 34382957429 on master, attempts 1 and 2, three separate jobs, step "Install Playwright Chromium":

Get:29 https://dl.google.com/linux/chrome-stable/deb stable/main amd64 Packages
Err:29  https://dl.google.com/linux/chrome-stable/deb stable/main amd64 Packages
  Hash Sum mismatch
E: Failed to fetch .../binary-amd64/Packages.gz  Hash Sum mismatch
Failed to install browsers
Error: Installation process exited with code: 100

Every Ubuntu archive source fetched fine; only Google's chrome-stable index is broken, and --with-deps runs apt-get update across all sources, so one bad third-party repo fails the step. Playwright downloads its own Chromium and takes system deps from the Ubuntu archives — it never needs that repo — so the fix is to remove the source list, not to retry a broken mirror.

Applied to all four install steps: e2e.yml (E2E, Docs-E2E) and nightly.yml (E2E-Full, Landing-Feedback — its steps.creds.outputs.present guard is kept). Nothing in docs/, scripts/, or CONTRIBUTING.md documents this step, so there was nothing to keep in sync (grep -rn "playwright install\|with-deps\|chrome-stable" docs/ scripts/ CONTRIBUTING.md local-dev/*.md → no hits).

2 — 53080 → 31080 (landing suite webServer)

Port	Old	New	Where
Landing e2e webServer	53080	31080	apps/e2e-test/playwright.landing.config.ts

Same class as the 53061/53062 fix in #378: Linux draws ephemeral source ports from 32768–60999, so a fixed listen port in that window can be transiently held by one of the runner's own outbound connections, and Playwright then polls a URL that never comes up. The reason is recorded beside the constant, pointing at the resume harness that documented the rule.

Method: a small Node codemod over git ls-files — exact-substring, longest-key-first, with an explicit KEEP list (53000, 53050–53056, 53060, 53070). It rewrote 1 file; git grep 53080 now matches only the explanatory comment. 31050–31055, 31060, 31080 were verified free repo-wide beforehand (only 31061/31062 are taken, by the resume harness). parity-fixtures.json contains no port numbers at all, so no fixture regen is involved.

Not done — the storybook ports (53050–53055) and the cf harness (53060)

Renumbering these would change external service configuration, which is the stop condition on this task:

• Each storybook's .env.example says: "Add this Storybook's origin (http://localhost:5305x) to each provider's list of authorized JavaScript origins / redirect URIs in its developer console", and for OneDrive/Dropbox/Box "redirectUri defaults to window origin". So the six origins are registered in four developer consoles (Google Cloud, Azure, Dropbox, Box).
• local-dev/.env.ports records the same fact: "53056 sits inside the OAuth clients' registered redirect range (53050-53060)".
• fix(ci): make nightly green — four independent causes behind #376 #378's own commit message states the ruling explicitly: ":53060 … sits inside the OAuth clients' registered redirect range (53050-53060), so it is pinned by external registration, not by convention."

Moving them silently breaks cloud-drive sign-in in the storybooks and the local playground until a human re-registers six origins in four consoles. Worth noting for whoever decides: the cf harness port is already env-overridable (UPUP_E2E_SERVER_PORT), while the six storybook ports are baked into each package's storybook script — so a CI-only override is a possible middle path that keeps the registered defaults intact.

53000 (the landing dev default) is also inside the ephemeral range but is left alone deliberately: it is the documented local-dev convention block (local-dev/LOCAL-DEV.md, local-dev/ports/landing.env, AGENTS_RULES.md), not a CI listen port.

Suspected, not proven

Attempt 1 of that nightly lost the cross-framework suite to a silent 900 s Timed out waiting 900000ms from config.webServer (job 102572070419) hours after a green ~52 s boot on dev. That is consistent with a storybook failing to bind a port inside the ephemeral range — Playwright only polls a URL and never surfaces the child's EADDRINUSE — but it is a hypothesis, not a demonstrated cause: the rerun died at the apt step before it ever reached the cross-framework job.

Gates (all via rtk proxy, raw exit codes captured)

Gate	Result
YAML parse of both workflows (yaml package; asserts all 4 install steps + the if: guard)	OK, exit 0
pnpm run typecheck	31/31 tasks, exit 0
pnpm --filter @useupup/e2e-test run typecheck	exit 0
pnpm run test:quality	392 test files + 5 workflows clean, exit 0
prettier --check on all 3 touched files	clean, exit 0
pre-commit (core + react + server suites)	green on both commits
pre-push (typecheck + lint + knip)	green

No package with unit tests was touched (packages/angular's server-mode-drive.spec.ts references :53060, which this PR does not move), so no suite needed re-running beyond the hooks.

[codesandbox]

Review or Edit in CodeSandbox

Open the branch in Web Editor • VS Code • Insiders

Open Preview