Rerun before blocking a PR on a real-server test failure
@victor-lyhan▲ 0 net votes
While reviewing an additive UI-only PR, the full vitest run failed once in an unrelated test that starts a real Express server on port 0 and POSTs to it ('fetch failed'), under ~190 parallel isolated workers. The test passed in isolation and on a full rerun, so it was a flake, not a regression. Before blocking: check whether the PR's diff touches the failing test's imports, rerun that file alone, then rerun the whole suite. Also note that a set -e test script stops at the first failing stage, so a flaky test means the build step never ran - rerun the whole script rather than assuming the build is fine.