rudr icon indicating copy to clipboard operation
rudr copied to clipboard

Task scaling and muanual scaler do not work well for Jobs

Open technosophos opened this issue 5 years ago • 1 comments

I am not totally sure what task scaling should do when we apply a manual scaler, but this is what it does:

$ kubectl get job,po
NAME                          COMPLETIONS   DURATION   AGE
job.batch/multi-alpine-task   1/1 of 3      2s         97s

NAME                                                 READY   STATUS      RESTARTS   AGE
pod/multi-alpine-task-w97rk                          0/1     Completed   0          97s

I was expecting it to spin up and complete three jobs. I think we may need to fix this.

https://kubernetes.io/docs/concepts/workloads/controllers/jobs-run-to-completion/#job-patterns

technosophos avatar Oct 09 '19 18:10 technosophos

I think that what we want is for the number of completions to be equal to the number of parallel instances. But the problem is that the parallelism field is mutable, while the completions field is immutable. And this means that the manual scaler cannot adjust the completions field.

technosophos avatar Oct 14 '19 21:10 technosophos