Read a deploy step that failed on a regression
When proofline ci deploy exits 3, the last line of the step’s log names each deploy that caused it and why:
A deploy of OWNER/NAME at SHA failed or recorded REGRESSED: production deploy:web run 4812 attempt 1 recorded REGRESSED at HOUR_1.
Each deploy reads recorded REGRESSED at the checkpoints that regressed, or failed when its GitHub deployment failed. A deploy whose GitHub deployment failed after a check recorded REGRESSED reads both, as in failed and recorded REGRESSED at HOUR_1. When several deploys caused the exit, the line names each of them.
Find the deploy in the log#
Above the last line, find the line for that deploy and checkpoint. It ends with the deploy page. The log lists each check through the checkpoint the step requested, and also any later check that already recorded REGRESSED, since that regression gives exit 3 too. For example, a step that waited for the landing check prints these lines for a deploy that regressed at its hour check:
production deploy:web run 4812 attempt 1: LANDING HEALTHY; state NOT_YET_DUE; gaps none; <deploy page>
production deploy:web run 4812 attempt 1: HOUR_1 REGRESSED; state NOT_YET_DUE; gaps none; <deploy page>
Follow the page link at the end of that line. For a failed deploy, follow the link on any of its lines; their state reads FAILED.
Read the evidence on the deploy page#
Under Checks, find the checkpoint marked Regressed. For the example above, open the deploy and read its 1 hour check. Under Signals, look for the highlighted reading in that checkpoint’s column. Hover over the reading or focus it with the keyboard to see the time window it covers. Compare it with the baseline shown for that signal.
If the page reads Deploy failed, the GitHub deployment failure alone gives exit 3. A check that observed the failure after landing says “The deployment failed after it landed.” If GitHub supplied a log, select Deploy log to read it.
A signal regression tells you which signal changed and during which window. It does not identify the cause. A later healthy check does not clear the earlier regression. Even a deploy whose coverage state is VERIFIED can have a regressed check. Read a deploy’s verification explains the checks. To trace a published report to GitHub, Slack or Notifications, see Find where a deployment regression was reported.
Decide what the pipeline does next#
Proofline reported the result and made no change to your environment. Your pipeline decides whether to hold a promotion or run your team’s rollback. If you roll back, that rollback is a deploy of its own. Follow its page to read its checks.
For command options and exit codes, see the CI reference.