Gitlab - atlantis/apply goes into "running" state prematurely

[cwstrommer]

Community Note

• Please vote on this issue by adding a 👍 reaction to the original issue to help the community and maintainers prioritize this request. Searching for pre-existing feature requests helps us consolidate datapoints for identical requirements into a single place, thank you!
• Please do not leave "+1" or other comments that do not add relevant new information or questions, they generate extra noise for issue followers and do not help prioritize the request.
• If you are interested in working on this issue or have submitted a pull request, please leave a comment.

---

Overview of the Issue

The atlantis/apply job for the pipeline enters the running state immediately when the atlantis/plan workflow runs. It should enter pending state first (to prevent the merge request from being merged before atlantis apply has been run) and only enter the running state when the apply workflow begins. I suspect that this may have occurred as a result of #1971 . I believe that ideally all jobs in the pipeline would be set to pending on init, running when their respective workflow begins, and failed/success when the workflow concludes.

Reproduction Steps

Open a merge request in Gitlab, wait for Atlantis to respond with the terraform plan. At this point the Atlantis pipeline will have three jobs with the following respective states:

• atlantis/policy_check, success
• atlatnis/plan, success
• atlantis/apply, running

This is despite the atlantis/apply job not having been triggered yet.

Environment details

• Atlantis versions tested: v0.19.1 and v0.19.2, deployed with the respective Helm chart
• Using custom workflows (init, plan, apply)

Additional Context

#1971

[ichasco-heytrade]

I have the same error:

atlantis-0 atlantis {"level":"warn","ts":"2022-03-16T10:50:32.883Z","caller":"events/plan_command_runner.go:187","msg":"unable to update plan commit status: POST https://gitlab.com/api/v4/projects/xxxxxxxxx/statuses/6cbaba6fb71ece24062a55b053b74ee8b4f17ff3: 400 {message: Cannot transition status via :run from :running (Reason(s): Status cannot transition via \"run\")}","json":{"repo":"heytrade/devops/terraform/projects/fortinet/fortigate","pull":"40"},"stacktrace":"github.com/runatlantis/atlantis/server/events.(*PlanCommandRunner).run\n\tgit.luolix.top/runatlantis/atlantis/server/events/plan_command_runner.go:187\ngit.luolix.top/runatlantis/atlantis/server/events.(*PlanCommandRunner).Run\n\tgit.luolix.top/runatlantis/atlantis/server/events/plan_command_runner.go:234\ngit.luolix.top/runatlantis/atlantis/server/events.(*DefaultCommandRunner).RunCommentCommand\n\tgit.luolix.top/runatlantis/atlantis/server/events/command_runner.go:259"}

Thanks!

[sujeets-toast]

Same issue I am facing on github. Its shoing atlantis/apply pending. I closed the pull request and created the new pullrequet there without running atlantis command in pullrequest its showing pending. What could be issue.

[hhsnow]

I think this is also an issue for Atlantis setups that have --disable-apply/ATLANTIS_DISABLE_APPLY=true configured. I'm seeing apply checks pending in GitHub that never clear and block merging.

[jamengual]

Does anyone try an older version of atlantis?

I wonder if there was a change on the Github Api or Github go library.

PRs are welcome.

[jamengual]

0.18.5 was reported to work.

[cwstrommer]

@jamengual 1.17.6 also works. FWIW since this also affects Gitlab I think the likelihood of this being a Github-specific API issue is very low. There's 2 symptoms that are being reported here:

1. The apply pipeline starts as "running" even if plan is still in process. This was the original intent of my submitted issue. I don't know if this happens in Github, but it is reproducible in Gitlab. This is a "cosmetic" issue in that it does not affect the merge request or Atlantis workflows.
2. There is more than one apply pipeline being created (equivalent to the multiple plan pipelines in this comment), but only one of them gets updated when the apply workflow finishes. I consider this to be a separate issue, but I think this is what @sujeets-toast and @hhsnow are referring to. This is also a breaking issue, because even after a successful "apply" workflow the request will be unmergeable because at least one pipeline is considered pending/running.

[jamengual]

but the gitlab issue mentioned was merged and since then we did not hear about this until now but we did some security updates to a bunch of client libraries so we are wondering if this could have been something related to that.

[gtg3vv]

Can confirm that we see this for bitbucket server as well (not github specific). The apply state shows as running as soon as the PR is planned and remains that way until actually applied.

[jamengual]

We just released a fix for the VCS status issue some of you have reported on our dev image docker pull ghcr.io/runatlantis/atlantis:dev Please test it and report back, Thanks.

[notnmeyer]

can confirm ghcr.io/runatlantis/atlantis:dev fixed the issue for me. appreciate it!

[gtg3vv]

We are still seeing this issue on the most recent release (0.19.4). Can anyone confirm if the fix made it into that version? All prs show an in-progress apply from the moment the first plan is initialized.