[redhat-3.13] pgupgrade: Ensure that old pg deployments are terminated before upgrade job is run (PROJQUAY-8092)

controllers/quay/features.go

@@ -408,7 +408,7 @@ func (r *QuayRegistryReconciler) checkMonitoringAvailable(
 // checkPostgresVersion returns the image name used by the currently deployed postgres version
 func (r *QuayRegistryReconciler) checkNeedsPostgresUpgradeForComponent(
 	ctx context.Context, qctx *quaycontext.QuayRegistryContext, quay *v1.QuayRegistry, component v1.ComponentKind,
-) error {
+) (err error, scaledDown bool) {
 	componentInfo := map[v1.ComponentKind]struct {
 		deploymentSuffix string
 		upgradeField     *bool
@@ -419,7 +419,7 @@ func (r *QuayRegistryReconciler) checkNeedsPostgresUpgradeForComponent(
 
 	info, ok := componentInfo[component]
 	if !ok {
-		return fmt.Errorf("invalid component kind: %s", component)
+		return fmt.Errorf("invalid component kind: %s", component), false
 	}
 
 	deploymentName := fmt.Sprintf("%s-%s", quay.GetName(), info.deploymentSuffix)
@@ -435,7 +435,7 @@ func (r *QuayRegistryReconciler) checkNeedsPostgresUpgradeForComponent(
 		postgresDeployment,
 	); err != nil {
 		r.Log.Info(fmt.Sprintf("%s deployment not found, skipping", component))
-		return nil
+		return nil, true
 	}
 
 	deployedImageName := postgresDeployment.Spec.Template.Spec.Containers[0].Image
@@ -458,9 +458,45 @@ func (r *QuayRegistryReconciler) checkNeedsPostgresUpgradeForComponent(
 		*info.upgradeField = true
 	} else {
 		r.Log.Info(fmt.Sprintf("%s does not need to perform an upgrade", component))
+		return nil, true
 	}
 
-	return nil
+	// at this point we have determined that these postgres deployments need to be upgraded and can set them to 0 replicas
+	// so that the upgrade job can run with no interference
+	r.Log.Info(fmt.Sprintf("scaling down %s deployment", component))
+	postgresDeployment.Spec.Replicas = &[]int32{0}[0]
+	postgresDeployment.Spec.Template.Spec.TerminationGracePeriodSeconds = &[]int64{600}[0]
+	if err := r.Client.Update(ctx, postgresDeployment); err != nil {
+		r.Log.Error(err, "unable to update postgres deployment replicas")
+	}
+	// now we wait to ensure that the deployment has scaled down before we proceed
+
+	terminatingPods := []corev1.Pod{}
+	podList := &corev1.PodList{}
+	labelSelector, err := metav1.LabelSelectorAsSelector(postgresDeployment.Spec.Selector)
+	if err != nil {
+		r.Log.Error(err, "unable to get label selector for postgres deployment")
+	}
+	err = r.Client.List(ctx, podList, &client.ListOptions{
+		LabelSelector: labelSelector,
+	})
+	if err != nil {
+		r.Log.Error(err, "unable to list pods for postgres deployment")
+		return err, false
+	}
+
+	for _, pod := range podList.Items {
+		if pod.Status.Phase == corev1.PodRunning {
+			terminatingPods = append(terminatingPods, pod)
+		}
+	}
+
+	if len(terminatingPods) > 0 {
+		r.Log.Info(fmt.Sprintf("Found %d pods in terminating status", len(terminatingPods)))
+		return nil, false
+	}
+
+	return nil, true
 }
 
 func extractImageName(imageName string) string {

controllers/quay/quayregistry_controller.go

@@ -553,7 +553,7 @@ func (r *QuayRegistryReconciler) Reconcile(ctx context.Context, req ctrl.Request
 
 	// Populate the QuayContext with whether or not the QuayRegistry needs an upgrade
 	if v1.ComponentIsManaged(updatedQuay.Spec.Components, v1.ComponentPostgres) {
-		err := r.checkNeedsPostgresUpgradeForComponent(ctx, quayContext, updatedQuay, v1.ComponentPostgres)
+		err, scaledDown := r.checkNeedsPostgresUpgradeForComponent(ctx, quayContext, updatedQuay, v1.ComponentPostgres)
 		if err != nil {
 			return r.reconcileWithCondition(
 				ctx,
@@ -564,11 +564,14 @@ func (r *QuayRegistryReconciler) Reconcile(ctx context.Context, req ctrl.Request
 				fmt.Sprintf("error checking for pg upgrade: %s", err),
 			)
 		}
+		if !scaledDown {
+			return r.Requeue, nil
+		}
 	}
 
 	// Populate the QuayContext with whether or not the QuayRegistry needs an upgrade
 	if v1.ComponentIsManaged(updatedQuay.Spec.Components, v1.ComponentClairPostgres) {
-		err := r.checkNeedsPostgresUpgradeForComponent(ctx, quayContext, updatedQuay, v1.ComponentClairPostgres)
+		err, scaledDown := r.checkNeedsPostgresUpgradeForComponent(ctx, quayContext, updatedQuay, v1.ComponentClairPostgres)
 		if err != nil {
 			return r.reconcileWithCondition(
 				ctx,
@@ -579,6 +582,9 @@ func (r *QuayRegistryReconciler) Reconcile(ctx context.Context, req ctrl.Request
 				fmt.Sprintf("error checking for pg upgrade: %s", err),
 			)
 		}
+		if !scaledDown {
+			return r.Requeue, nil
+		}
 	}
 
 	if err := r.checkBuildManagerAvailable(quayContext, cbundle); err != nil {

kustomize/components/clairpgupgrade/base/clair-pg-old.deployment.yaml

@@ -18,7 +18,7 @@ spec:
       labels:
         quay-component: clair-postgres-old
     spec:
-      terminationGracePeriodSeconds: 180
+      terminationGraceperiodSeconds: 600
       serviceAccountName: clair-postgres
       volumes:
         - name: clair-postgres-conf-sample
@@ -27,15 +27,6 @@ spec:
         - name: postgres-data
           persistentVolumeClaim:
             claimName: clair-postgres-13
-      initContainers:
-        - name: check-postgres-scale-down
-          image: quay.io/sclorg/postgresql-13-c9s:latest
-          command:
-            - /bin/sh
-            - -c
-            - |
-              echo "Waiting for 30 seconds before starting the main container..."
-              sleep 30
       containers:
         - name: postgres
           image: quay.io/sclorg/postgresql-13-c9s:latest

kustomize/components/clairpgupgrade/base/clair-pg-upgrade.job.yaml

@@ -10,7 +10,7 @@ spec:
   template:
     spec:
       restartPolicy: OnFailure
-      terminationGracePeriodSeconds: 180
+      terminationGraceperiodSeconds: 600
       serviceAccountName: clair-postgres
       volumes:
         - name: clair-postgres-conf-sample

kustomize/components/clairpostgres/postgres.deployment.yaml

@@ -18,7 +18,7 @@ spec:
       labels:
         quay-component: clair-postgres
     spec:
-      terminationGracePeriodSeconds: 180
+      terminationGraceperiodSeconds: 600
       serviceAccountName: clair-postgres
       volumes:
         - name: clair-postgres-conf-sample

kustomize/components/pgupgrade/quay-pg-old.deployment.yaml

@@ -18,7 +18,7 @@ spec:
       labels:
         quay-component: postgres
     spec:
-      terminationGracePeriodSeconds: 180
+      terminationGraceperiodSeconds: 600
       serviceAccountName: quay-database
       volumes:
         - name: postgres-conf-sample

kustomize/components/pgupgrade/quay-pg-upgrade.job.yaml

@@ -10,7 +10,7 @@ spec:
   template:
     spec:
       restartPolicy: OnFailure
-      terminationGracePeriodSeconds: 180
+      terminationGraceperiodSeconds: 600
       serviceAccountName: quay-database
       volumes:
         - name: postgres-conf-sample
@@ -55,10 +55,9 @@ spec:
               cpu: 500m
               memory: 2Gi
           command:
-          - "/bin/sh"
-          - "-c"
+            - "/bin/sh"
+            - "-c"
           args:
-          - >
-            run-postgresql --version || (echo "postgres migration command failed, cleaning up..." && rm -rf /var/lib/pgsql/data/* && exit 1)
+            - >
+              run-postgresql --version || (echo "postgres migration command failed, cleaning up..." && rm -rf /var/lib/pgsql/data/* && exit 1)
   backoffLimit: 50
-

kustomize/components/postgres/postgres.deployment.yaml

@@ -18,7 +18,7 @@ spec:
       labels:
         quay-component: postgres
     spec:
-      terminationGracePeriodSeconds: 180
+      terminationGraceperiodSeconds: 600
       serviceAccountName: quay-database
       volumes:
         - name: postgres-conf-sample