db upgrage: upgrade postgres 15 to postgres 16

The Postgres upgrade from 15 to 16 or any later versions can be done by adding the env variable named POSTGRES_UPGRADE = copy in the pg db pod.

More about upgrade process is here: https://www.postgresql.org/docs/current/pgupgrade.html

In our case, we achieve the upgrade by setting this env var in noobaa-db-pg STS.

Once after upgrade is done, we have to remove the POSTGRES_UPGRADE env from the STS. Otherwise pg db pod fails to come up because the db pod check PG directory version against upgrade image version and exits if there is no upgrade required and POSTGRES_UPGRADE is still provided.

To remove this env, we store the upgrade status in noobaa-core CR after the PG upgrade and refer the same status during db reconcile. If the upgrade is already done, we remove this env from desired sts status.

This will help to upgrade the db smoothly

Signed-off-by: Vinayak Hariharmath <[email protected]>

pkg/options/options.go

@@ -72,7 +72,7 @@ var NooBaaImage = ContainerImage
 
 // DBImage is the default db image url
 // it can be overridden for testing or different registry locations.
-var DBImage = "quay.io/sclorg/postgresql-15-c9s"
+var DBImage = "quay.io/sclorg/postgresql-16-c9s"
 
 // Psql12Image is the default postgres12 db image url
 // currently it can not be overridden.

pkg/system/phase2_creating.go

@@ -33,6 +33,7 @@ import (
 	"k8s.io/apimachinery/pkg/runtime"
 	"k8s.io/apimachinery/pkg/util/intstr"
 	"sigs.k8s.io/controller-runtime/pkg/client"
+	"sigs.k8s.io/controller-runtime/pkg/controller/controllerutil"
 )
 
 const (
@@ -120,6 +121,9 @@ func (r *Reconciler) ReconcilePhaseCreatingForMainClusters() error {
 		return err
 	}
 
+	if err := r.UpgradeDb(); err != nil {
+		return err
+	}
 	// create the db only if postgres secret is not given
 	if r.NooBaa.Spec.ExternalPgSecret == nil {
 		if err := r.ReconcileDB(); err != nil {
@@ -152,6 +156,34 @@ func (r *Reconciler) ReconcilePhaseCreatingForMainClusters() error {
 	return nil
 }
 
+func (r *Reconciler) UpgradeDb() error {
+	var result controllerutil.OperationResult
+	var err error
+
+	if r.NooBaa.Status.PostgresUpdatePhase != nbv1.UpgradePhaseFinished {
+		result, err = controllerutil.CreateOrUpdate(
+			r.Ctx, r.Client, r.NooBaaPostgresDB, func() error {
+				r.setPGUpgradeEnvInSTS()
+				r.updateDBImageForUpgrade()
+				return nil
+			},
+		)
+
+		if result == controllerutil.OperationResultUpdated {
+			r.NooBaa.Status.PostgresUpdatePhase = nbv1.UpgradePhaseFinished
+		}
+	} else if r.NooBaa.Status.PostgresUpdatePhase == nbv1.UpgradePhaseFinished {
+		_, err = controllerutil.CreateOrUpdate(
+			r.Ctx, r.Client, r.NooBaaPostgresDB, func() error {
+				r.unSetPGUpgradeEnvInSTS()
+				return nil
+			},
+		)
+	}
+
+	return err
+}
+
 // SetDesiredServiceAccount updates the ServiceAccount as desired for reconciling
 func (r *Reconciler) SetDesiredServiceAccount() error {
 	if r.ServiceAccount.Annotations == nil {
@@ -223,6 +255,68 @@ func (r *Reconciler) SetDesiredServiceDBForPostgres() error {
 	return nil
 }
 
+func (r *Reconciler) updateDBImageForUpgrade() {
+	var podSpec = &r.NooBaaPostgresDB.Spec.Template.Spec
+
+	for i := range podSpec.Containers {
+		c := &podSpec.Containers[i]
+		if c.Name == "db" {
+			// Fetch the DB image from options.go
+			c.Image = options.DBImage
+			if r.NooBaa.Spec.DBResources != nil {
+				c.Resources = *r.NooBaa.Spec.DBResources
+			}
+
+			c.Lifecycle = &corev1.Lifecycle{
+				PreStop: &corev1.LifecycleHandler{
+					Exec: &corev1.ExecAction{
+						Command: []string{"/bin/sh", "-c", "pg_ctl -D /var/lib/pgsql/data/userdata/ -w -t 60 -m fast stop"},
+					},
+				},
+			}
+
+		}
+	}
+}
+
+// Set env "POSTGRESQL_UPGRADE=copy" in PG sts to iniate the upgrade
+func (r *Reconciler) setPGUpgradeEnvInSTS() {
+	for i, container := range r.NooBaaPostgresDB.Spec.Template.Spec.Containers {
+		if container.Name == "db" {
+			for _, env := range container.Env {
+				if env.Name == "POSTGRESQL_UPGRADE" {
+					return
+				}
+			}
+			envVars := container.Env
+			newEnvVar := corev1.EnvVar{
+				Name:  "POSTGRESQL_UPGRADE",
+				Value: "copy",
+			}
+
+			envVars = append(envVars, newEnvVar)
+			r.NooBaaPostgresDB.Spec.Template.Spec.Containers[i].Env = envVars
+			return
+		}
+	}
+}
+
+// unSet env "POSTGRESQL_UPGRADE=copy" in PG sts after upgrade
+func (r *Reconciler) unSetPGUpgradeEnvInSTS() {
+	for i, container := range r.NooBaaPostgresDB.Spec.Template.Spec.Containers {
+		if container.Name == "db" {
+			newEnvVars := []corev1.EnvVar{}
+			for _, env := range container.Env {
+				if env.Name != "POSTGRESQL_UPGRADE" {
+					newEnvVars = append(newEnvVars, env)
+				}
+			}
+			r.NooBaaPostgresDB.Spec.Template.Spec.Containers[i].Env = newEnvVars
+			return
+		}
+	}
+}
+
 // SetDesiredNooBaaDB updates the NooBaaDB as desired for reconciling
 func (r *Reconciler) SetDesiredNooBaaDB() error {
 	var NooBaaDBTemplate *appsv1.StatefulSet = nil
@@ -239,6 +333,14 @@ func (r *Reconciler) SetDesiredNooBaaDB() error {
 	NooBaaDB.Spec.ServiceName = r.ServiceDbPg.Name
 	NooBaaDBTemplate = util.KubeObject(bundle.File_deploy_internal_statefulset_postgres_db_yaml).(*appsv1.StatefulSet)
 
+	// Make sure the POSTGRES_UPGRADE=copy flag is unset before reconcile.
+	// Not setting the flag soon after sts update in Upgrade_db() because
+	// removing the flag while db upgrade still in progress will create issue.
+	// So doing it here
+	if r.NooBaa.Status.PostgresUpdatePhase == nbv1.UpgradePhaseFinished {
+		r.unSetPGUpgradeEnvInSTS()
+	}
+
 	podSpec := &NooBaaDB.Spec.Template.Spec
 	podSpec.ServiceAccountName = "noobaa-db"
 	defaultUID := int64(10001)
@@ -1150,6 +1252,7 @@ func (r *Reconciler) ReconcileDB() error {
 		}
 
 	}
+
 	return nil
 }