allow empty resources when defaults are empty

[FxKu]

closes #2057

This PR removes the hard-coded internal defaults for the (default) resource config options for Spilo and pooler pods allowing for not setting resources at all while relying on the K8s environment to handle it.

This means deployments lacking a resource section and a config map with no defaults defined are likely to fail because memory/cpu request (and memory limits) are missing.

The PR also adds a safeguard function matchLimitsWithRequestsIfSmaller for situations where limits are smaller than the actual requests, e.g. if no limit is specified in the manifest and the operator falls back on a default value which is still lower than the request.

[idanovinda]

👍

[FxKu]

👍

[antverpp]

This is not working. I updated my operator to 1.11.0. I remove setting the cpu limits, but still receive “1” on limits. The same if I set “0” - it still makes “1”

[ktalo]

This is not working. I updated my operator to 1.11.0. I remove setting the cpu limits, but still receive “1” on limits. The same if I set “0” - it still makes “1”

If you have installed using the helm chart, the defaults are still set in

postgres-operator/charts/postgres-operator/values.yaml

Line 233 in 7c7aa96

# configure resource requests for the Postgres pods

and thus end up in your operator's config. You need to update your operator using a values that unset those (i.e.. set to null).

[marcofunstage]

Hi @ktalo,
I set the following configuration, but doesn't work; setting empty value, restarting postgres or the operator pods didn't help either

apiVersion: helm.toolkit.fluxcd.io/v2beta1
kind: HelmRelease
metadata:
  name: postgres-operator
  namespace: postgres-operator
spec:
  interval: 5m
  chart:
    spec:
      chart: postgres-operator
      version: '1.12.0'
      sourceRef:
        kind: HelmRepository
        name: zalando-postgres-operator
        namespace: flux-system
      interval: 60m
  values:
    configKubernetes:
      enable_cross_namespace_secret: true
    configMajorVersionUpgrade:
      major_version_upgrade_mode: "manual"
    configPostgresPodResources:
      default_cpu_limit: "0"

[ktalo]

If you still have old CRDs present, the defaults are there, too. The CRDs won't get updated with helm upgrade. The default limits are in CRD operatorconfigurations.acid.zalan.do and that needs to be updated too.

EDIT: and for operator config, I had success with the following helm values:

configPostgresPodResources:
  default_cpu_limit: null
  default_cpu_request: null
  default_memory_limit: null
  default_memory_request: null
  min_cpu_limit: null
  min_memory_limit: null