[kube-prometheus-stack] allow kube-state-metrics job name to be overridden

[jhandguy]

What this PR does / why we need it

When the nameOverride value of the kube-state-metrics sub-chart is overridden in the values of the kube-prometheus-stack chart, like so:

kube-state-metrics:
  nameOverride: kube-state-metrics-test

Some Prometheus alerting rules are no longer valid, as the job label is directly hard-coded in some queries as "kube-state-metrics", like for instance in the KubePodCrashLooping alerting rule:

- alert: KubePodCrashLooping
   expr: max_over_time(kube_pod_container_status_waiting_reason{reason="CrashLoopBackOff", job="kube-state-metrics", namespace=~"{{ $targetNamespace }}"}[5m]) >= 1

This PR ensures all Prometheus alerting rules using the kube-state-metrics service name as job label also take the nameOverride value of the kube-state-metrics sub-chart into account.

Which issue this PR fixes

• fixes [kube-prometheus-stack] Rules uses hardcoded job names #4110

Special notes for your reviewer

Checklist

• DCO signed
• Chart Version bumped
• Title of the PR starts with chart name (e.g. [prometheus-couchdb-exporter])

[jkroepke]

Hi,

thanks for you contribution. However all files insides rules-1.14 directory are auto-generated by an python script. In case I would merge this, the changes are getting reverted by next person who execution the script.

Please adjust the script https://github.com/prometheus-community/helm-charts/blob/main/charts/kube-prometheus-stack/hack/sync_prometheus_rules.py to include your changes.

[jhandguy]

Hi @jkroepke, thanks for your fast review!
I have adapted the python script and regenerated the alerting rules with it.

[jkroepke]

@jhandguy did you test this on your local system?

If I run

jkr@joe-nb kube-prometheus-stack % helm upgrade -i kube-prometheus-stack . --set kube-state-metrics.fullnameOverride=kube-state-metrics-test

The all metrics still have job=kube-state-metrics. If I'm looking deeper on the kube-state-metrics chart, it seems like that the value of the named templated kube-state-metrics.name will be used by default which only depends on nameOverride and .Chart.Name.

If I only override fullname, the job label will not changed here.

[jhandguy]

@jhandguy did you test this on your local system?

If I run

jkr@joe-nb kube-prometheus-stack % helm upgrade -i kube-prometheus-stack . --set kube-state-metrics.fullnameOverride=kube-state-metrics-test

The all metrics still have job=kube-state-metrics. If I'm looking deeper on the kube-state-metrics chart, it seems like that the value of the named templated kube-state-metrics.name will be used by default which only depends on nameOverride and .Chart.Name.

If I only override fullname, the job label will not changed here.

@jkroepke you are absolutely right, I wrongly assumed that fullnameOverride was being used and the chart I tested this change with (as a sub-chart) had both fullnameOverride and nameOverride set to the same value. Apologies for the confusion, I should have tested this with a kind cluster instead.

The script now uses nameOverride instead of fullnameOverride.

[jkroepke]

Thanks!