Expand description
Flaky verification: a verify.e2e failure that did not reproduce.
graph::run_e2e_with_retry runs a failed e2e command once more on the same
commit before anything is decided from it. When that second run passes, the
first failure was not a verdict on the patch, and this module does the two
things that follow:
- identifies the failure (
fingerprint) without knowing the test framework, from the failure lines the saved output excerpt keeps; - files one queue task per distinct failure asking for the root cause
(
record), so ten runs hitting the same flaky test leave one task.
Best effort by construction, like followup::after_merge: a filing failure
becomes a run event and never changes a verdict.
Structs§
- Signature
- What identifies one failure: the normalised failure lines, sorted and distinct, plus a line to show a person.
Constants§
- NODE
- Node name carried by
Source::Agenton a filed flake task.
Functions§
- file
- File the task for
flake, or find the unfinished one that already tracks the same failure. Returns the task id and whether this call created it. - fingerprint
- A stable identifier of the failure in
repo: 40 bits of FNV-1a over the repository and theSignaturelines, as ten hex digits. The command is deliberately not part of it: the same failing test reached through a different command line is the same defect. - record
- Record a flake in the run’s ledger and file (or find) its task. Never
fails: the outcome is part of the record and of a
verifyevent. - signature
- The
Signatureof a failing command’s output (the excerpt magi kept) and exit code.