Merge pull request #12952 from sp98/hostpath-to-pvs

mon: failover mons from hostpath to persistent volumes

pkg/operator/ceph/cluster/mon/health.go

@@ -366,9 +366,32 @@ func (c *Cluster) checkHealth(ctx context.Context) error {
 		}
 	}
 
+	// failover mons running on host path to use persistent volumes if VolumeClaimTemplate is set and vice versa
+	for _, mon := range c.ClusterInfo.Monitors {
+		if c.HasMonPathChanged(mon.Name) {
+			logger.Infof("fail over mon %q due to change in mon path", mon.Name)
+			c.failMon(len(c.ClusterInfo.Monitors), desiredMonCount, mon.Name)
+			return nil
+		}
+	}
+
 	return nil
 }
 
+// HasMonPathChanged checks if the mon storage path has changed from host path to persistent volume or vice versa
+func (c *Cluster) HasMonPathChanged(mon string) bool {
+	var monPathChanged bool
+	if c.mapping.Schedule[mon] != nil && c.spec.Mon.VolumeClaimTemplate != nil {
+		logger.Infof("mon %q path has changed from host path to persistent volumes", mon)
+		monPathChanged = true
+	} else if c.mapping.Schedule[mon] == nil && c.spec.Mon.VolumeClaimTemplate == nil {
+		logger.Infof("mon %q path has changed from persistent volumes to host path", mon)
+		monPathChanged = true
+	}
+
+	return monPathChanged
+}
+
 func (c *Cluster) trackMonInOrOutOfQuorum(monName string, inQuorum bool) (bool, error) {
 	updateNeeded := false
 	var monsOutOfQuorum []string

pkg/operator/ceph/cluster/mon/health_test.go

@@ -645,3 +645,37 @@ func TestUpdateMonInterval(t *testing.T) {
 		assert.Equal(t, time.Minute, h.interval)
 	})
 }
+
+func TestHasMonPathChanged(t *testing.T) {
+	t.Run("mon path changed from pv to hostpath", func(t *testing.T) {
+		c := New(context.TODO(), &clusterd.Context{}, "ns", cephv1.ClusterSpec{}, nil)
+		c.mapping.Schedule["a"] = nil
+		result := c.HasMonPathChanged("a")
+		assert.True(t, result)
+	})
+
+	t.Run("mon path has not changed from pv to hostpath", func(t *testing.T) {
+		c := New(context.TODO(), &clusterd.Context{}, "ns", cephv1.ClusterSpec{}, nil)
+		c.spec.Mon.VolumeClaimTemplate = &v1.PersistentVolumeClaim{Spec: v1.PersistentVolumeClaimSpec{}}
+		c.mapping.Schedule["b"] = nil
+		result := c.HasMonPathChanged("b")
+		c.spec.Mon.VolumeClaimTemplate = nil
+		assert.False(t, result)
+	})
+
+	t.Run("mon path changed from hostpath to pv", func(t *testing.T) {
+		c := New(context.TODO(), &clusterd.Context{}, "ns", cephv1.ClusterSpec{}, nil)
+		c.mapping.Schedule["c"] = &opcontroller.MonScheduleInfo{}
+		c.spec.Mon.VolumeClaimTemplate = &v1.PersistentVolumeClaim{Spec: v1.PersistentVolumeClaimSpec{}}
+		result := c.HasMonPathChanged("c")
+		assert.True(t, result)
+	})
+
+	t.Run("mon path has not changed from host path to pv", func(t *testing.T) {
+		c := New(context.TODO(), &clusterd.Context{}, "ns", cephv1.ClusterSpec{}, nil)
+		c.mapping.Schedule["d"] = &opcontroller.MonScheduleInfo{}
+		result := c.HasMonPathChanged("d")
+		c.spec.Mon.VolumeClaimTemplate = nil
+		assert.False(t, result)
+	})
+}

pkg/operator/ceph/cluster/mon/mon.go

@@ -1229,6 +1229,12 @@ func (c *Cluster) commitMaxMonIDRequireIncrementing(desiredMaxMonID int, require
 var updateDeploymentAndWait = UpdateCephDeploymentAndWait
 
 func (c *Cluster) updateMon(m *monConfig, d *apps.Deployment) error {
+
+	if c.HasMonPathChanged(m.DaemonName) {
+		logger.Infof("path has changed for mon %q. Skip updating mon deployment %q in order to failover the mon", m.DaemonName, d.Name)
+		return nil
+	}
+
 	// Expand mon PVC if storage request for mon has increased in cephcluster crd
 	if c.monVolumeClaimTemplate(m) != nil {
 		desiredPvc, err := c.makeDeploymentPVC(m, false)