Expose metrics for MachineDeployment status

[embik]

We need metrics to track MachineDeployments managed by this machine-controller over time. Primarily, the MachineDeployments' status information wrapped in metrics would be helpful, so something like:

• machine_deployment_available_replicas
• machine_deployment_ready_replicas
• machine_deployment_replicas
• machine_deployment_updated_replicas

With these metrics, alerts could be defined that e.g. alert on machine_deployment_replicas > machine_deployment_updated_replicas being true for more than 30 minutes (random example here).

[adberger]

For this use case https://github.com/kubernetes/kube-state-metrics/blob/main/docs/metrics/extend/customresourcestate-metrics.md could be enough?

[embik]

For this use case https://github.com/kubernetes/kube-state-metrics/blob/main/docs/metrics/extend/customresourcestate-metrics.md could be enough?

That's definitely a way to implement metrics for these values if you need them, thank you for bringing up this option. At best, we expose those natively in future releases, but if you (as in you stumbled over this issue because you are missing those metrics) need this asap, the CRD solution from kube-state-metrics should help.

[rajaSahil]

@embik Are there any plans to take this up?
I suppose we have to go ahead and write a simple exporter for this, or are there any other way to do it?

[embik]

@rajaSahil machine-controller already has a metrics endpoint (see https://github.com/kubermatic/machine-controller/blob/main/cmd/machine-controller/main.go#L257 for the "entry point"). So we should just add more metrics there instead of writing a dedicated exporter.

[rajaSahil]

Okay.
@embik Can you assign this to me, I can take a look at it.

[embik]

You can comment /assign to get it assigned to yourself, try it out! 😄

[rajaSahil]

Sure, thank you!

[rajaSahil]

/assign