CI: render-chrome test fails when Chrome does not start on a busy runner #207

packages/cli/test/render-chrome.test.ts › "render_page checks in Chrome" › "finds each problem and names the section and component" failed on CI with Error: Chrome did not start (src/render-chrome.ts:95, the 20 s start timer) on the Node 24 job of release PR #204 (run 43005028930, commit 850e04f). Nothing in that PR touches the renderer: it only bumps versions. A rerun of the failed job is the workaround. This is the second time this test has failed on a busy runner: it timed out once before, during the 0.7.x releases.

session() already relaunches Chrome once after a hung start. Two launches on a runner busy with turbo run test --concurrency=2 can still both exceed 20 s.

Options:

• Raise the start timer under CI, or let session() retry more than once with backoff.
• Run the Chrome tests in their own vitest project or file pool, so they don't compete with the D1/workerd suites.
• Skip the test with a clear message when Chrome cannot start within the budget, and keep the in-page checks covered by the unit tests that don't need Chrome.

Done when the test passes reliably on CI, or fails only for a real renderer regression.

🤖 Generated with Claude Code