When CI goes red, Sinter reruns the failed jobs once, checks every failing test against your history, and leaves a comment on the pull request that says which failures are flakes and which ones are yours.
Elixir and ExUnit · GitHub Actions · private alphaSinter: 2 of 3 failing tests are flaky
Latest commit a41c9e2.
Flaky on this commit (1)
• dedupes rows across partitions
in
ImportTest:
failed, then passed on Sinter's rerun
Known flaky (1)
• parses 13F with missing CUSIP
in
ParserTest:
flaked on 6 other commits in the last 30 days
New failures (1)
• rejects expired invitations
in
InvitationTest
From a red X to an answer, without anyone clicking “Re-run”.
Temper, an ExUnit formatter, records every test outcome in CI, and mix temper.push sends it to Sinter. Sinter never runs your tests and never sees your code.
Only the failed jobs, only on the first attempt, and only when tests actually failed. A compile error or a linter is not something a rerun can explain, so Sinter leaves those alone. Repository owners turn it on.
Passed on the rerun: flaky here. Failed again: real. Flaked on other commits in the last 30 days: known flaky, even if this branch never touched it. Anything else is new, and probably yours.
A neutral “Sinter” check on the head commit, plus an optional comment with each test, its evidence and its failure message. Both update in place as workflows report.
Every failure is a flaky test, so re-running the failed jobs is safe.
See how flaky your suite is, and what to fix first.
The dashboard Sinter started with: every run, every retry, every partition, per repository and across the org.
GitHub Actions first. Leave an address and you will hear back personally.
No list, no newsletter. One email when your invite is ready.