Argo: Nested steps run with incorrect parameters

[markterm]

Is this a BUG REPORT or FEATURE REQUEST?:
Bug report

What happened:
If you execute the parallelism-nested.yaml example then the pods with a seq-id of b, c or d all have a parallel-id of 1

What you expected to happen:
pods with a seq-id of b, c or d all to have all parallel ids (1, 2, 3 and 4).

How to reproduce it (as minimally and precisely as possible):
argo submit parallelism-nested.yaml
then:
argo get <workflow-id> -o yaml

There will be multiple pods in the Status section with seq-id of b and parallel id of 1, there should only be one.

Anything else we need to know?:
This looks to be related to https://github.com/argoproj/argo/pull/1552

storedTemplates contains the following, with seq-id hardcoded:

    /seq-worker:
      arguments: {}
      inputs: {}
      metadata: {}
      name: seq-worker
      outputs: {}
      parallelism: 1
      steps:
      - - arguments:
            parameters:
            - name: parallel-id
              value: "1"
            - name: seq-id
              value: '{{item}}'
          name: seq-step
          template: one-job
          withParam: |
            ["a","b","c","d"]

Environment:

• Argo version: master

$ argo version

• Kubernetes version : 1.12

[markterm]

@dtaniwaki fyi, I think this relates to your PR

[NirvanaZA]

Hi sir, When this bug fix will release? Hope soon. Many thanks.

[sarabala1979]

@dtaniwaki is refactoring the store template feature. This will close most of the bugs.

[NirvanaZA]

I tried to downgrade the version to 2.4.0, from 2.4.2, and seems the retry feature has some bug as well.

[simster7]

@NirvanaZA could you elaborate on the bug with the retry feature?

[samath117]

Another minimal example to illustrate what appears to be the same problem:

apiVersion: argoproj.io/v1alpha1
kind: Workflow
metadata:
  generateName: steps-
spec:
  entrypoint: steps
  templates:
  - name: print
    inputs:
      parameters:
      - name: message
    container:
      image: alpine:latest
      command: [sh, -c]
      args: ["echo {{inputs.parameters.message}}"]
  - name: print-twice
    inputs:
      parameters:
      - name: message
    steps:
    - - name: print1
        template: print
        arguments:
          parameters:
          - name: message
            value: "{{inputs.parameters.message}}"
    - - name: print2
        template: print
        arguments:
          parameters:
          - name: message
            value: "{{inputs.parameters.message}}"
  - name: steps
    steps:
    - - name: hello
        template: print-twice
        arguments:
          parameters:
          - name: message
            value: hello
    - - name: goodbye
        template: print-twice
        arguments:
          parameters:
          - name: message
            value: goodbye

On Argo 2.4.1, this has the expected behavior. Workflow logs:

print1: hello
print2: hello
print1: goodbye
print2: goodbye

On Argo 2.4.2, the message for the last one is passed incorrectly:

print1: hello
print2: hello
print1: goodbye
print2: hello

[cchanley2003]

Just FYI we are also seeing issues with when statements getting the wrong parameters with a nested setup. It works with 2.4.0-rc1 but is broken after that, so pretty certain it is related to that same change that is affected this ticket.

[simonwa7]

FYI after creating and deploying an argo controller image with https://github.com/argoproj/argo/pull/1744 it did fix the issues we were seeing with parameter passing, but now the podSpecPatch does not get applied or even stored with a workflow.

I.E. after submitting a workflow with a task that has a podSpecPatch it is not found anywhere within the workflow obtained by argo get <workflow-ID> -o yaml

[simster7]

@simonwa7 Could you take a look at @dtaniwaki's comments regarding your issue? https://github.com/argoproj/argo/pull/1744#issuecomment-553455599

[simster7]

For those blocked by this regression: I would consider building #1744 into an image until the PR goes through review and a release is made hopefully sometime next week.

[NirvanaZA]

For those blocked by this regression: I would consider building #1744 into an image until the PR goes through review and a release is made hopefully sometime next week.

👍 Waiting for the new release.

[simster7]

Hello everyone v2.4.3 was just released with a fix for this issue, please take a look! Thank you for your patience and thank you to @dtaniwaki for working on this fix!