fix(flux): make the database health check actually apply (#614)

kubernetes/components/apps/yucca-database.yaml

@@ -10,12 +10,11 @@ spec:
     - name: cnpg-operator
     - name: cnpg-plugin-barman
     - name: openebs
-  # Gate dependents on the database SERVING, not on Helm succeeding. CNPG holds
-  # the Cluster's Ready condition False while any instance is short (e.g. a lost
-  # node strands an instance on its node-local PV), which times out the Helm
-  # release and — through dependsOn — freezes the image rollout of every app
-  # behind it, even though the primary is up and taking queries. readyInstances
-  # is the field that tracks "someone can serve"; the Ready condition is not.
+  # Gate on serving, not on Helm succeeding: CNPG holds Ready False whenever an
+  # instance is short, which froze every dependent app's images for 18h.
+  # Threshold 2 — one stranded instance is survivable, a lone primary is not.
+  # wait MUST stay false: it health-checks the whole inventory and ignores
+  # healthChecks, which puts the stuck HelmRelease back in the way.
   healthChecks:
     - apiVersion: postgresql.cnpg.io/v1
       kind: Cluster
@@ -24,13 +23,13 @@ spec:
   healthCheckExprs:
     - apiVersion: postgresql.cnpg.io/v1
       kind: Cluster
-      current: status.readyInstances >= 1
+      current: status.readyInstances >= 2
   interval: 1h
   retryInterval: 2m
   timeout: 10m
   path: ./kubernetes/apps/base/yucca-database
   prune: true
-  wait: true
+  wait: false
   sourceRef:
     kind: GitRepository
     name: ${MANIFEST_SOURCE:=flux-system}