Expected Behavior

When task failed, the finally tasks should be trigger immediately.

Actual Behavior

In parallel case, finally tasks not trigger only all parallel tasks done. See the code snippet:

      - name: job-activity-job-1
        runAfter:
          - job-activity-fake-run-cmd
        taskSpec:
          results:
            - name: err-source
              description: indicates exit code for fake command
          steps:
            - name: run
              image: ubuntu
              script: |
                #!/bin/sh
                sleep 60
                printf "job 1 found issue" | tee /tekton/results/err-source
                echa "job-1 start to run..."
      - name: job-activity-job-2
        runAfter:
          - job-activity-fake-run-cmd
        taskSpec:
          results:
            - name: err-source
              description: indicates exit code for fake command
          steps:
            - name: run
              image: ubuntu
              script: |
                #!/bin/sh
                printf "job 2 found issue" | tee /tekton/results/err-source
                echa "job-2 start to run..."
    finally:
      - name: exception-handler
        taskRef:
          apiVersion: custom.tekton.dev/v1alpha1
          kind: Exception
          name: exception
        params:
        - name: pipelinerun_name
          value: guarded-pr

See, the job-activity-job-1 and job-activity-job-2 will be run as parallel and job-activity-job-2 will failed immediate(wrong command:echa) and job-activity-job-1 will failed after 60s.

The finally is triggered exactly after job-activity-job-1 failed.

Steps to Reproduce the Problem

Additional Info

• Kubernetes version:

  Output of kubectl version:

  (paste your output here)
  

• Tekton Pipeline version:

  Output of tkn version or kubectl get pods -n tekton-pipelines -l app=tekton-pipelines-controller -o=jsonpath='{.items[0].metadata.labels.version}'

See the code: https://github.com/tektoncd/pipeline/blob/f764c3ba94294d5675b2d5490240336bd66c1972/pkg/reconciler/pipelinerun/resources/pipelinerunstate.go#L387-L398

Seems if one taskrun is not “done” (skip, success,failed), the finally will not be trigger, and it’s also depends on the order of taskruns, I think this not make sense.