Allow S3 Endpoint for WAL Backup to Be Provided as a Secret in PostgresCluster

[chkpwd]

Hey guys 👋 — first off, loving the product!

I had a question that evolved into a potential feature request:
Is it currently possible to provide the S3 endpoint for pgBackRest WAL backups via a Kubernetes Secret (e.g., as a secretRef) rather than specifying it directly in the PostgresCluster manifest?

After some helpful discussion with @andrewlecuyer, it became clear that:

• Removing the s3 section from the PostgresCluster spec and instead defining S3 settings in a secret (e.g. repo1-s3-endpoint, etc.) is partially supported.
• However, if the s3 section is removed, the configMap that would normally be created for pgBackRest doesn't get generated, causing related pods to hang waiting for a missing volume mount.

So creating the configMap manually isn't an option.

It would be great if we could keep the S3 endpoint private, especially when it contains sensitive identifiers like a Cloudflare account ID. While it’s not a critical security leak, it’s still information we’d prefer not to expose in plaintext manifests.

See here:

Ideally, patching the schema to allow a secretRef to provide those values would be the solution.

Would it be possible to enhance PostgresCluster to support this workflow officially?

Thanks again 🙏

Overview

Environment

Please provide the following details:

• Platform: (Kubernetes)
• Platform Version: (v1.32.2)
• PGO Image Tag: (ubi8-15.10-2-v0.3.0)
• Postgres Version (15)
• Storage: (rook-ceph)

Steps to Reproduce

Create a PostgresCluster with minimal without the s3 configuration and provide it via a secretRef.

apiVersion: postgres-operator.crunchydata.com/v1beta1
kind: PostgresCluster
metadata:
  name: ${APP}
spec:
  postgresVersion: 15
  instances:
  [...]
  backups:
    pgbackrest:
      configuration:
        - secret:
            name: "crunchy"
      global:
        repo1-block: "y"
        [...]
      repos:
        - name: repo1

EXPECTED

1. Deployment to consume the secret and create the configMap

ACTUAL

1. Pod hangs because configMap is not created.

[andrewlecuyer]

Thanks for submitting this @chkpwd!