Flaky Tests: Flaky test detection on merge branches

We’re excited to announce improved flaky test detection for merge queue users. In a merge queue, a single flaky failure will force every enqueued PR behind it to be retested, affecting every engineer…

We’re excited to announce improved flaky test detection for merge queue users. In a merge queue, a single flaky failure will force every enqueued PR behind it to be retested, affecting every engineer involved and wasting tons of CI resources. If you run a merge queue like Trunk Merge to test your changes before they’re merged, tests run on the merge branches are an important signal about your tests’ health. Flaky Tests can now detect flaky tests using signals on your PR branches, merge branches, and protected branches.

Tests are expected to behave differently on protected branches such as main or master compared to tests run on PR branches. Trunk adjusts for this inherent volatility of PR branches by applying a slightly different set of rules to them:

• Inconsistent test results on the same commit hash, where results are different on identical code is considered flaky
• Inconsistent test results on different commit hashes, but across a large number of PRs is considered flaky

Merge branches created by merge queues, on the other hand, are treated similarly to protected branches. We make the assumption that before PRs are submitted to a merge queue, tests have already been run once on the PR and are passing.

If you’d like to learn more about how Trunk detects flaky tests, see the Detection page in docs.

Trunk Flaky Test supports detection on merge branch for Trunk Merge, GitLab Merge Trains, Graphite Merge Queues, and GitHub merge queue. If you’re using an unsupported merge queue and would like to apply detection on merge branches, contact us on Slack.