Supporting HPA for Zalando postgres

[kannanvr]

Hi,
We can eaily support the HPA by adding few parameters on CRD.

subresources:
  status: {}
  scale:
    labelSelectorPath: .status.selector
    specReplicasPath: .spec.numberOfInstances
    statusReplicasPath: .status.replicas

• Which image of the operator are you using? - registry.opensource.zalan.do/acid/postgres-operator:v1.7.1
• Where do you run it - cloud or metal? Kubernetes or OpenShift? K8s
• Are you running Postgres Operator in production? yes
• Type of issue? feature request

we can scale Up the postgres CRD based on the number of Read Request and so on...

Thanks,
Kannan V

[CyberDem0n]

Stateful services are not for horizontal scaling.

If you start doing so please be prepared to deal with the following problems:

1. When you scale out, the new replica might need to bootstrap. Currently it is mostly done by taking a basebackup from the leader. What is the database size is a few TB? How long would it take? What if scaled out so eagerly that many replicas are pulling backups from the leader?
2. When you scale in, volumes are not removed. But, when you scale them later out, old volumes will be reused, but data would be stale. After that there will be two possibilities. These replicas will slowly catch up by replaying WAL files from the archive, or they will remain stale until reinitialized (see the previous point).

[AnisTigrini]

Hi there @CyberDem0n, I am working to integrate HPA in the CRD of Zalando operator.
From what I see your second argument would be taken care of from Kubernetes 1.23 onwards (alpha). Here is the documentation: https://kubernetes.io/blog/2021/12/16/kubernetes-1-23-statefulset-pvc-auto-deletion/

[baznikin]

Auto-managing read replicas is indeed great feature. I am thinking about work-arounds on my side, but it be great if operator could manage it somehow. HPA is ok solution
https://docs.aws.amazon.com/AmazonRDS/latest/AuroraUserGuide/Aurora.Integrating.AutoScaling.html