fix: Correct SIGTERM handling. Fixes #10518 #10337 #10033 #10490

[alexec]

Signed-off-by: Alex Collins [email protected]

This MIGHT fix these:

Fixes #10518
Fixes #10337
Fixes #10033
Fixes #10490

Please do not open a pull request until you have checked ALL of these:

• Create the PR as draft .
• Run make pre-commit -B to fix codegen and lint problems.
• Sign-off your commits (otherwise the DCO check will fail).
• Use a conventional commit message (otherwise the commit message check will fail).
• "Fixes #" is in both the PR title (for release notes) and this description (to automatically link and close the issue).
• Add unit or e2e tests. Say how you tested your changes. If you changed the UI, attach screenshots.
• Github checks are green.
• Once required tests have passed, mark your PR "Ready for review".

If changes were requested, and you've made them, dismiss the review to get it reviewed again.

[sxllwx]

Do we consider using the cmd.Context() as the root-context ?

[alexec]

I've been working with this, and I'm not sure my PR will fix the issues.

[sxllwx]

Hope my work can help you 😄

#10490 (comment)

[alexec]

@sxllwx I think you're suggesting:

1. NotifyContext catches a SIGTERM (probably sent from the controller).
2. This block will exit.
3. stop() will be invoked, clearing the signal handler.
4. Another SIGTERM occurs (controller again), but without a handler we get default behaviour.
5. Default behaviour is to exit (does this mean crash?).

If that is what you think is correct, then my fix should work.

[sxllwx]

Yes. I tried to look at the default processing method of Golang's runtime, and the relevant code:

https://github.com/golang/go/blob/master/src/runtime/signal_unix.go#L1068-L1074

https://github.com/golang/go/blob/master/src/runtime/signal_unix.go#L869-L875

Here's a test workflow I've been using for the past weekend, available for your use.

apiVersion: argoproj.io/v1alpha1
kind: CronWorkflow
metadata:
   generateName: etcd-code-artifact-
spec:
   schedule: "*/1 * * * *"
   concurrencyPolicy: "Forbid"
   startingDeadlineSeconds: 0
   workflowSpec:
    entrypoint: run
    imagePullSecrets:
    - name: ccr-secret
    templates:
      - name: run
        inputs:
            artifacts:
            - name: repo-artifact
              path: /src
              git:
                repo: "https://github.com/etcd-io/etcd.git"
        container:
          image: "python:3.9-buster"
          command: [ sh, -c ]
          args: [ "git status && ls && pwd" ]
          workingDir: /src
        outputs:
          artifacts:
            - name: repo-artifact
              path: /src

[alexec]

Closing as I've created #10523 which will push images for testing.