How do I stop my Playwright tests being flaky in CI but not locally?

Question

My suite passes every time on my laptop, but roughly one run in four fails in GitHub Actions — always on a different test, usually a timeout waiting for a button.

I have already tried bumping the global timeout to 60s, which reduced but did not eliminate the failures. Retries hide it, but that feels like cheating.

What actually causes this gap between local and CI, and how do I fix the root cause rather than masking it?

Ravi Menon · Accepted Answer

The gap is almost always resources and timing , not Playwright. CI runners are slower and heavily contended, so anything that was "fast enough" locally becomes a race. Three things fix the majority of cases: Never wait on time, wait on state. Replace waitForTimeout with web-first assertions like await expect(locator).toBeVisible() — they retry automatically. Kill shared state between tests. If tests share a user or a database row, parallel workers in CI will collide. Give each test its own…

Sofia Marquez · Answer

Adding to the above: check whether your CI browser is running headless while you test headed locally. Headless Chromium has a different default viewport, which silently changes which elements are in view and whether lazy-loaded content has rendered.

Run headed locally with --headed off to reproduce, or pin the viewport explicitly in your config.

How do I stop my Playwright tests being flaky in CI but not locally?

2 Answers

Your answer