[Gitlab CI] play a job in failure case

GitLab CI/CD
1

Hello,

I am trying to achieve the following pattern in a CI pipeline:

Untitled Diagram

If and only if job B or C, which depends on B, failed, I want to play the failure job. If both jobs succeed, I don’t want to play the failure job.

I tried with “needs” and “when” but I didn’t find the right solution.
My last idea is to use “after_script” but it seems too complex for a basic need.

Does anyone have an idea or have they encountered this need before?

Best regards,

1 Like
2

Hi @Maphi

What you want to do should certainly be possible. I think you probably want something like this:


stages:
  - ...
  - B
  - C
  - D
  - ...

...

jobC:
  - stage: C
  - rules:
    if: ...
    when: on_success
    allow_failure: true
  - ...

jobDsucceed:
  - stage: D
  - rules:
    if: ...
    when: on_success
  - ...

jobDfailed:
  - stage: D
  - when: on_failure
  - ...

It probably doesn’t help that the when keyword seems to take an on_failure value when it’s used on its own, but not when it is part of a rule. I’m not sure how this inconsistency in the YAML reference came about!

3

Thank you for your help @snim2
It is not exactly my need, but my original post was unclear, I edited it.

In fact, for steps B and C, I want the pipeline to fail if one of them fails. And likewise if B fails C should not be played, but directly the failure job.
With your idea I end up with:

  stages:
    - B
    - C
    - failed

  B:
    stage: B
    allow_failure: true
    script:
      - echo "Job B"

  C:
    stage: C
    needs:
      - job: B
    when: on_success
    allow_failure: true
    script:
      - echo "Job C"

  failed:
    stage: failed
    needs:
    - job: B
    - job: C
    script:
      - echo "fail job"
    when: on_failure

The problem is that allowing B to fail, the pipeline run C that end up successfull and don’t run the failing job.

4

Hmm. I don’t think there’s any way to say “if this job or that job fails, then fail the whole pipeline”, so I think you are going to end up with some sort of kludge.

The way I would approach this is to have jobs B and C upload some sort of artifact and then have a job D which needs B and C, checks their artifacts, and only passes if the artifacts show that B and C also passed.

Your other option might be to just combine B and C into a single job, but I imagine you want to avoid that for the obvious reasons.