0

I have following yaml-configuration for my horizontal pod autoscaler in GCP.

//...
  metrics:
    - type: Pods
      pods:
        metric:
          name: "my-custom-metric"
        target:
          type: AverageValue
          averageValue: 60

//..

I have added the metric as a custom metric in metric explorer in Cloud Monitoring. It is a log based metric. I have even added it to a dashboard and it looks fine.

For HPA the metric is not found. I get "Unable to read all metrics".

I'm not sure why? Do I need to add specific permissions in my serviceaccount? Or do I need to use a Custom Metrics Adapter? Does the HPA not have access without a adapter?

I have even tried to add the full name path for the metric name, but I get an error that "/" is not allowed in the name.

I would appreciate any help!

user11122042
  • 11
  • 3

1 Answers1

1

Indeed, you need to deploy a metric server, please refer to this link. And bear in mind that it works for cpu and memory.

  1. CPU and Memory HPA

Notice:

Kubelet certificate needs to be signed by cluster Certificate Authority (or disable certificate validation by passing --kubelet-insecure-tls to Metrics Server) So add under args (you will get some headaches and TLS can be added later):

- --kubelet-insecure-tls
  1. Custom metrics

Notice: Before reading below check this out.

This will require some custom resource definition as well as adapter, metrics api. This video might help A LOT :)

https://www.youtube.com/watch?v=iodq-4srXA8&t=550s

Bonus track: do not hesitate to use siege to stress the pods, also i would try scale up and scale down :)

jmvcollaborator
  • 2,141
  • 1
  • 6
  • 17
  • Do I always need the adapter? Is it not possible that the hpa uses custom metrics (integers) without an additional adapter? I would have thought that hpa can access Google Cloud Monitoring without additional needs. – user11122042 Aug 10 '23 at 09:27