Failure Handling
Documentation menu
A pipeline failure should tell a future maintainer what broke, where it broke and what to try next. These notes describe how to keep that information visible.
Common failure modes
- A test fails on one agent but passes on another, usually because of an environment difference.
- A package version already exists in the feed, so the publish step is rejected.
- A signing or symbol step cannot find the file it expects.
- A template input is missing or has the wrong type.
Keep the evidence
Publish test results and coverage output even when the test step fails. Keep package artifacts for a short retention period. Name the agent and the project in log messages so a failure does not require guesswork.
Retry policy
Retry only the steps that are safe to repeat. A failed restore or a transient feed error is a reasonable retry candidate. A failed test is not. Retrying a test hides a real defect and makes the pipeline slower for everyone.