child steps template with outputs breaks execution of the parent steps template

[ghost]

Pre-requisites

• I have double-checked my configuration
• I can confirm the issues exists when I tested with :latest
• I'd like to contribute the fix myself (see contributing guide)

What happened/what did you expect to happen?

Okay, so this is a weird one.

If you run the workflow I provided below, you'll see that step1, which is supposed to trigger template t2, never shows up in the UI. The workflow gets marked as successful, and after inspecting the workflow status manually using kubectl, I can see that step1 did get executed, but it is not visible in the UI. This might not be a bug in the UI since the controller could be doing something wrong, and the UI follows along. I don't have enough knowledge about internals to know right away.

This only happens when step0 calls another StepsTemplate that has outputs declared (in this case it is t1). When I remove outputs from t1 it works, the step1 does show up in the UI. If step0 calls t2 it works as well.

Workflow that does not work - step1 never shows up in the UI

apiVersion: argoproj.io/v1alpha1
kind: Workflow
metadata:
  generateName: lifecycle-hook-tmpl-level-
  labels:
    test: test
spec:
  entrypoint: t0
  templates:
    - name: t0
      steps:
        - - name: step0
            template: t1
        - - name: step1
            template: t2
    - name: t1
      outputs:
        parameters:
          - name: dummy
            valueFrom:
              expression: "1"
      steps:
        - - name: step2
            template: t2
    - name: t2
      container:
        image: alpine:3.6
        command: [sh, -c]
        args: ["echo this is t2"]

How the UI looks like

Workflow that does work - step1 shows up in the UI (outputs from t1 are removed)

apiVersion: argoproj.io/v1alpha1
kind: Workflow
metadata:
  generateName: lifecycle-hook-tmpl-level-
  labels:
    test: test
spec:
  entrypoint: t0
  templates:
    - name: t0
      steps:
        - - name: step0
            template: t1
        - - name: step1
            template: t2
    - name: t1
      # outputs:
      #   parameters:
      #     - name: dummy
      #       valueFrom:
      #         expression: "1"
      steps:
        - - name: step2
            template: t2
    - name: t2
      container:
        image: alpine:3.6
        command: [sh, -c]
        args: ["echo this is t2"]

How the UI looks like

I was getting around this issue #12283 by declaring dummy outputs, but then I noticed the bug I described above.

Version

v3.5.2

Paste a small workflow that reproduces the issue. We must be able to run the workflow; don't enter a workflows that uses private images.

apiVersion: argoproj.io/v1alpha1
kind: Workflow
metadata:
  generateName: lifecycle-hook-tmpl-level-
  labels:
    test: test
spec:
  entrypoint: t0
  templates:
    - name: t0
      steps:
        - - name: step0
            template: t1
        - - name: step1
            template: t2
    - name: t1
      outputs:
        parameters:
          - name: dummy
            valueFrom:
              expression: "1"
      steps:
        - - name: step2
            template: t2
    - name: t2
      container:
        image: alpine:3.6
        command: [sh, -c]
        args: ["echo this is t2"]

Logs from the workflow controller

-

Logs from in your workflow's wait container

-

[Joibel]

This sounds similar to something I've looked at recently. I'll take a look - thanks for narrowing it down.

[ghost]

Awesome, no problem.

[Joibel]

I'm pretty sure you've only reproduced this on 3.5.1 and it is fixed on 3.5.2 by #12203.

[ghost]

Hmm, it's possible I made a mistake when specifying the version, I definitively know the bug was there. I was working in a temporary environment so I can't 100% confirm what version I had checked out. I can confirm that I don't see the bug anymore when building from the main branch. Thanks for taking the time to help, and sorry if this wasted your time.

[agilgur5]

For posterity, marking as duplicate of #12165