Fail suspended steps after deadline

[markterm]

Suspended steps would never time out, regardless of the setting of activeDeadlineSeconds.

This change fixes that.

[xianlubird]

Should we add a example or tests file?

[markterm]

There's already an example with activeDeadlineSeconds, but I've added a test.

[sarabala1979]

looks like it is only working on activeDeadlineSeconds: 0 you need to requeue the wf to keep checking the deadlinseconds. Try the below example.

apiVersion: argoproj.io/v1alpha1
kind: Workflow
metadata:
  generateName: suspend-template1
spec:
  entrypoint: suspend
  activeDeadlineSeconds: 5
  templates:
    - name: suspend
      suspend: {}

[markterm]

Is this not an existing problem with activeDeadlineSeconds rather than relating to this commit? Eg if you have a pod that doesn't have a status change rather than a suspend step then I also think the wf wouldn't be requeued ...

…

On Tue, 5 Nov 2019 at 01:25, Saravanan Balasubramanian < ***@***.***> wrote: looks like it is only working on activeDeadlineSeconds: 0 you need to requeue the wf to keep checking the deadlinseconds. Try the below example. apiVersion: argoproj.io/v1alpha1 kind: Workflow metadata: generateName: suspend-template1 spec: entrypoint: suspend activeDeadlineSeconds: 5 templates: - name: suspend suspend: {} — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#1704>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AABO7LSJSNTDX7ULCOTQYD3QSDDSDANCNFSM4JDTHAFA> .

[sarabala1979]

Yes, you are right, it is an existing issue. I can see every 10 mins fresh all WF refresh. We need to come up with some implementation like

1. Keep requeue the WF if it suspended with duration or WF has activedeadlinseconds
2. Implement scheduler to requeue the WF with a specified time duration (suspend->duration or activedeadlineseconds).

if we have the above implementation then the controller will behave the expected behavior.

[sarabala1979]

Workflow code is already using c.workqueue.AddAfter(key, addAfter) Can you use this to requeue the suspended wf after activedeadlineseconds?

[markterm]

@sarabala1979 I have applied this to requeue suspended steps.