Backend: Chains of needs is not automatically triggered after a failure

Summary

Not all jobs in a chain of needs are triggered automatically if one of the previous jobs fails.

Steps to reproduce

1. Create a stage with multiple jobs, each depending on the previous one using needs. The third job needs the second, the second job needs the first etc. Sample GitLab ci file below.
2. Trigger the pipeline and let the first job fail.
3. After the failure of the first job, retry the failed job and let it pass.
4. Wait for the next jobs to be triggered.

gitlab-ci.yml:

stages:
  - my_stage

job_1:
  stage: my_stage
  tags:
    - my-runner
  script:
    - exit $(( $RANDOM % 2)) # Exit randomly with 0 or 1

job_2:
  stage: my_stage
  needs:
    - job_1
  tags:
    - my-runner
  script:
    - exit 0

job_3:
  stage: my_stage
  needs:
    - job_2
  tags:
    - my-runner
  script:
    - exit 0

What is the current bug behavior?

If job_1 failed, the next steps are marked as skipped:

After successfully retrying job_1, next in the line job_2 gets automatically triggered but job_3 remains skipped:

After manually retrying job_2, job_3 finally gets automatically triggered:

What is the expected correct behavior?

After job_1 fails, press retry on job_1 and both job_2 and job_3 are automatically triggered.

Output of checks

This bug happens on GitLab.com