CODEWITHSUNDEEP
kubernetescontrolsmonitoringmetricsplane
Sasidhar Sekar
Sasidhar Sekar

Reputation: 113

Documentation/Code base for control plane metrics

I'm trying to capture the impact of enabling Horizontal Pod Autoscaler (HPA) resources on the performance of HPA, in Kubernetes. I have found a few metrics related to the HPA but find that they lack documentation. For example, the metric horizontalpodautoscaler_queue_latency is available but it is not clear what unit it is measured in - microseconds, milliseconds or anything else.

Can anyone point me to any documentation related to control plane metrics? It would be great if you can also point me to the code base of these metrics as well because I could not find any reference to control plan metrics (tried searching for horizontalpodautoscaler_queue_latency) in https://github.com/kubernetes/kubernetes.

Thanks a ton.

Upvotes: 1

Views: 44

Answers (2)

Sasidhar Sekar
Sasidhar Sekar

Reputation: 113

@KoopaKiller You are right on the deprecated comment. I chose the wrong example to illustrate the concern. horizontalpodautoscaler_queue_latency is indeed deprecated. Maybe the reason I'm unable to look for it in the codebase.

I raised this question as a common concern for all control plane metrics, where for so many of these, the unit of measurement (particularly for latency) isn't clear.

Since I raised this question, I have managed to find the metrics codebase for control plane components. Below is a link for the scheduler.
https://github.com/kubernetes/kubernetes/blob/0926c9c47677e944c2dadcd7b39587866331eeba/pkg/scheduler/metrics/metrics.go

I expect the other components to be in https://github.com/kubernetes/kubernetes/blob/0926c9c47677e944c2dadcd7b39587866331eeba/pkg/, although I found that the structure is not standardized. So, you might have to search a bit to find what you're looking for.

Upvotes: 0

Mr.KoopaKiller
Mr.KoopaKiller

Reputation: 3982

There is no much information about this metric, but I found this metrics is mensured in quantile and not time (milisecs, microsecs etc.).

Here you can find the following:

horizontalpodautoscaler_queue_latency_count

(Deprecated) How long an item stays in workqueuehorizontalpodautoscaler before being requested.

Also, seems the metric is deprecated, maybe it is the reason that you cannot found too much information about.

Upvotes: 1

Related Questions