Why it matters: When every e2e test fails with “Executable doesn’t exist”, the real failure is one step earlier, and whether you hit it depends on which runner region the job landed in.

Context

A CI job ran npx playwright install --with-deps chromium and then the e2e suite. On several consecutive runs:

  • the install step failed in apt with Temporary failure resolving '<regional package mirror>' and exit code 100;
  • the job carried on;
  • every test then failed with browserType.launch: Executable doesn't exist.

The same workflow passed on another branch in between, which made it look like a problem with the change.

The insight

The outcome depended on runner placement, not on the code. Both jobs used the same hosted runner label, but the passing job’s apt used a different regional package mirror than the failing ones. Reruns, and a fresh run after updating the branch, kept landing on affected runners, and the provider’s status page showed no incident.

How to triage

Look at the install step first, before touching any test:

Terminal window
gh run view --job <id> --log-failed | grep -E 'Temporary failure resolving|exited with code'

Then compare the apt mirror lines of the install step in a passing and a failing job:

Terminal window
gh run view --job <id> --log | grep 'archive.ubuntu.com' | head

If the mirrors differ, it is the runner, and rerunning the tests proves nothing about the change.