#Failure summary

▍failure-summary120×32

failure-summaryA pipeline preview pane. The top block lists the pipeline's jobs grouped by stage, with two of them marked failed and carrying their failure reasons, script_failure and runner_system_failure. Below it, a First failure block names the job, the line number, and four lines of the trace around the assertion that failed.

#What you see

Three blocks, in the order you need them.

Which jobs failed. Grouped by stage, so a whole stage failing looks different from one flaky test. Allow-failure jobs are marked and rendered as warnings rather than errors, because the pipeline was told to tolerate them.

Why each one failed. GitLab returns a failureReason on every failed job:

Two rows on the same screen reading script_failure and runner_system_failure are two different incidents, and they render differently.

The first failure, in full. The job, the line number, and the lines around the error, extracted from the trace and shown inline. For most failures this is the whole answer.

#Where the lines come from

GitLab's GraphQL CiJobTrace type exposes one field, htmlSummary, and it is a truncated tail. The full log comes from REST, GET /projects/:id/jobs/:job_id/trace, which returns every line the runner produced.

The extraction runs over the tail of the trace and matches the shapes that common test runners and compilers emit. When nothing matches, it falls back to the last non-empty lines before the failure.

#What you can do

Key	Action	Key	Action
L	the full log, scrolled to this error	R	retry the pipeline
y	copy the failure block	o	open the job in the browser

L goes from a failed pipeline row to the failing job's log, already scrolled to the error. It does not open at line 1 of 40,000, and it does not stop at a job list for you to choose from.

#Where it appears

The failure summary appears in the preview whenever the selected pipeline has a failed job. The columns that go with it, failure reason and queued duration and coverage, sit in the pipeline table under the same column priority rules as everywhere else.

#Repeat failures

A pipeline you already retried, that failed again, is a different queue from one that failed once. status:failed retried:true updated < 7d in the filter bar, then P, keeps that queue as a tab.