Failed CI run: flaky or real?

Rerunning a failed pipeline until it turns green is the quick fix, but it treats a broken test like a network hiccup and burns build time. 2501 works out why the run failed, retries it once when the cause was transient, and sends a real failure to whoever owns the fix.

A run can fail because a download timed out or a build machine dropped out, or because of a broken test, an expired credential or a deployment that failed its checks. Each has a different owner.

What 2501 does

The CI specialty reads the failed run before it touches anything and classifies the failure. A transient one gets a single retry, and the agent waits for the result before closing the ticket; if the same failure comes back, it is real. A real failure goes with the evidence to its owner: the author of the change for a failing test or build, the platform team for credentials and build machines, the service owner for a deployment. It reports what failed, why and what it did, across GitHub Actions, GitLab CI, Jenkins, CircleCI and Azure DevOps.

You stay in control

• You choose the level: investigate only, act after a person confirms, or act on its own within your rules. The level is set per specialty, not per ticket.
• One retry, never a loop: a failure is retried once at most, and one that fails the same way every time, such as a broken test, is not retried at all.
• Pipelines and credentials stay yours: it never edits pipeline definitions or stored credentials, and never restarts shared production build machines on its own.
• Your rules apply: which pipelines it may retry unattended, and which failures go straight to a person.

Frequently asked

Why not rerun a failed deployment straight away? A deployment that failed partway can leave the environment half updated, and running it again on top can make things worse. The environment's state is confirmed first.

Can it classify failures without retrying anything? Yes. In investigate-only mode it reports why the run failed and who owns the fix, and retries nothing.

Related

• CI Manager
• Infrastructure drifting from its Terraform configuration
• Read-only AI investigation