2

I have a 3-node kubernetes cluster with 3 n1-standard-1 :1 vCPU, 3,75 Go RAM, in a gcloud projet. Recently, I ran a test that created quite a few jobs (about 150). Half the jobs were running an hello world script, the other half were failing due to configuration error, and those jobs kept retrying by restarting new pods.

In my console I observed the following graph:

is the unit % or 1.0

It seems that my cluster is capped at 3% of maximum utilization, if I read the units correctly. What bothers me is that in my test, the amount of pods/utilisation should not be constant, so I was not expecting the utilization to be a flat line like this.

Also, I noticed about about 30 pods were in unschedulable state, so my initial thought was that I reached maximum utilization of the instance.

EDIT

It is clear now that my test is too poorly designed, so I cannot conclude about maximum utilization so far.

  • 1
    Could you add the content of the script? – 030 Dec 26 '17 at 11:20
  • 30 pods were in unschedulable state Could you add the output of kubectl get services? – 030 Dec 26 '17 at 11:21
4

It seems that my cluster is capped at 3% of maximum utilization, if I read the units correctly. This makes no sense at all, why isn't it capable of reaching 100% ?

Why do you think it does make no sense at all?

I ran a few tests that created quite a few pods

What is a few tests?

(each pod running basically an hello world script)

What does this script do? Print helloworld? If that is the case then it could be possible that the cluster will not use 100%.

By the way, how large is the cluster? How many agents? How many memory? So if the cluster possesses 1024GB of memory then it could be the case it is using 3%.

In summary, to indicate whether something makes no sense at all a number of facts is required:

  • content of the helloworld script
  • characteristics of the cluster, e.g. totol cpu, memory, type of disks, storage capacity, number of agents, masters
  • definition of few tests and few pods. How many tests were run, frequency, e.g. 10/s, how many pods

If you would really like to loadtest a k8s cluster, kubernetes has some documentation about this, e.g. this document explains how to load test a k8s cluster and in this case force autoscaling.

  • thank you for the answer, I realize now my question needs a lot of clarification. Let me edit. – Overdrivr Dec 26 '17 at 11:12

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.