Do we consider the readiness time of the pod while scaling up?

Question

Do we consider the readiness time of the pod while scaling up?

jotish opened this issue 5 years ago · 5 comments

Answer 1 · 2020-06-19T08:24:04.000Z

@jotish can you please elaborate more in the description. Not very clear. Lost the context.

Answer 2 · 2020-07-12T20:56:27.000Z

I think I have also experienced this.
when a pod taking a lot of time until ready, the auto-scaler "thinks" that the pod already consuming jobs, although its not the case.
this causes wrong scaling problems.
I've reported an issue here:
#101

Answer 3 · 2020-07-16T14:41:19.000Z

Currently the scaling happens based on available replicas. f00ef3f in #105 changes this behavior so scaling happens based on spec replicas which includes both available and not ready replicas so that it doesn't scale up beyond the target.

Answer 4 · 2020-07-17T11:15:35.000Z

Closed #102 as #105 fixes this problem

Answer 5 · 2020-07-22T09:54:07.000Z

Pods no longer scale up on every tick(default 20seconds). It all depends on how many workers are needed to consumer the backlog(available)+reserved(inprocess) messages and the queue RPM. Closing this issue, as it solved now. @jotish