How to best collect available metrics in platform

[OlleLarsson]

What should be investigated. As a part of being able to define what metrics we offer in the platform we first need a good way of collecting them.

I see two ways of doing this

• Collect from documentation, e.g.
  • etcd
  • opensearch
  • cAdvisor
  • kubernetes
• Collect from running applications, e.g. with kubectl port-forward + curl, ssh + curl for Kubelet.

A problem with collecting from documentation is that, I could be wrong, not all applications expose this in documentation.

A problem with collecting from running applications is that we then need to have a live environment and we need to account for different features sets, e.g. exposed metrics might be different depending on what features are enabled.

What artifacts should this produce.

• At least an investigation document on the two approaches or if you think that another approach might work, and which one is recommend,
• but preferably also an implementation of collecting metrics, if no implementation a task for implementing it should be created!