New Canceled status for Stopped/Terminated Workflows

[jmeridth]

Summary

new Canceled status

NOTE: debates can ensue regarding one l (American English) or two l (British English). Looks like CircleCI and TravisCI use one. They could be doing i18n and because I'm in the USA it renders as one. 🤷

Use Cases

Since many consumers of argo-workflows are using it for Continuous Integration (CI), when the following sequence occurs, a Cancelled status would be nice:

• pull request created
• workflow starts CI run
• new commit(s) pushed to pull request
• if the previous workflow is still running, it should be terminated and status set to Canceled (currently when terminated its status is set to Failed)
• new workflow starts CI run, including new commits

This will allow for the analytics to be accurate on which workflows actually Failed 😄

---

Message from the maintainers:

Love this enhancement proposal? Give it a 👍. We prioritise the proposals with the most 👍.

[jmeridth]

We currently determine previous workflows by having passed GitHub org, repo and pr number as labels and before we kick off a new workflow (via argo events to a custom service) we call the argo-workflows API to find any previously running workflows with those labels and terminate them (resulting in a ❌ Failed status with message Stopped with strategy 'Terminate'

[agilgur5]

Yea I thought about this recently too, it is a bit confusing that there is no custom status for a shutdown and instead just a message. Had a question about this on Slack recently too.

This would be a breaking change and require a sizeable refactor (with a pretty large chance for bugs as there is quite a lot of code dependent on statuses), but I do think it would be a good to distinguish if possible.

There's also a question here of if we should add a Cancelled NodeStatus as well. I believe that currently the running node is marked as Failed and all later nodes are marked as Skipped, both of which require a bit of interpretation right now as they are not the same as regular failures or regular skips. This popped up in a code review too

NOTE: debates can ensue regarding one l (American English) or two l (British English).

Wait a minute, I'm American and thought this was always two l, i.e. "Cancelled" 😵‍💫

[terrytangyuan]

Some comments in #11992

[agilgur5]

There's also a question here of if we should add a Cancelled NodeStatus as well. I believe that currently the running node is marked as Failed and all later nodes are marked as Skipped, both of which require a bit of interpretation right now as they are not the same as regular failures or regular skips. This popped up in a code review too

This was changed in #13214 to Failed, but Cancelled would still be easier to understand/interpret