job depends on artifacts but not on success or failure

what I need

three gitlab-ci jobs - lets call them A, B and C.

• job A always succeeds
• job B always fails
• job C needs the artifacts of A and B

they are all in the same stage (if that matters)

so clearly job C needs to depend on A and B to finish (artifacts are generated when they finish). But the docs don't seem to have such an entry. They only have a trigger for success and failure. not for finished jobs in general.

minimal example

stages:
  - build
  - test
  - coverage

build_job:
  stage: build
  script:
    - echo building...

test_job_A:
  needs:
    - build_job
  stage: test
  script:
    - echo "logs of tests executed in A" > test_result_A.txt
  artifacts:
    paths:
      - test_result_*

test_job_B:
  needs:
    - build_job
  stage: test
  script:
    - echo "logs of tests executed in B" > test_result_B.txt
    - mimic failing pipeline
  artifacts:
    paths:
      - test_result_*

coverage_C:
  needs:
    - test_job_A
    - test_job_B
  stage: test
  script:
    - cat test_result_A.txt
    - cat test_result_B.txt

which will result in the coverage_C job not executed:

using dependencies and moving the job C into another stage:

...

coverage_C:
  dependencies:
    - test_job_A
    - test_job_B
  stage: coverage
  script:
    - cat test_result_A.txt
    - cat test_result_B.txt

results in the same - coverage_C job not executed:

coverage_C is in test, shouldn't it be in coverage? In any case, test_job_B fails, therefore any job that depends on its success (!) will not start. That is a reasonable default interpretation of what a dependency means.

Your mistake is that by using the default, you depend on the success of the test jobs, you need to configure coverage_C so that it runs even if one of its dependencies fails. Check out the docs for the jobs' rules section, you need when: always.

Now, there is still another challenge left: A failing job will by default not export any artifacts. You probably want test logs even if one of them fails, so you will have to configure that as well. As a suggestion, create an empty file first though, so if anything fails fatally and the file is not created, it doesn't cause the dependent job to fail.

Allow me to edit your answer for some more details. You can re-write this as you like.

...

test_job_B:
  needs:
    - build_job
  stage: test
  script:
    - echo "logs of tests executed in B" > test_result_B.txt
    - mimic failing pipeline
  artifacts:
    when: always
    paths:
      - test_result_*

coverage_C:
  needs:
    - test_job_A
    - test_job_B
  stage: coverage
  script:
    - cat test_result_A.txt
    - cat test_result_B.txt
  when: always

other_job_D:
  stage: test
  script:
    - sleep 200

• Important for the failing job is the artifacts:when:always to save the artifacts on failure.

• the needs is completely correct. Using dependencies will end up waiting for the complete prior stage - which is unwanted.

• the stage should be a stage that is after the jobs you want to wait for. So in this case it would be stage: coverage

• job C then also needs the when:always flag because it doesn't want to depend on job success of jobs A and B