Merge pull request #4465 from leseb/various-fixes

ceph: various controller and crash fixes

pkg/operator/ceph/cluster/cluster.go

@@ -122,21 +122,21 @@ func (c *cluster) detectCephVersion(rookImage, cephImage string, timeout time.Du
 	if err != nil {
 		return nil, errors.Wrapf(err, "failed to extract ceph version")
 	}
-	logger.Infof("Detected ceph image version: %s", version)
+	logger.Infof("Detected ceph image version: %q", version)
 	return version, nil
 }
 
 func (c *cluster) validateCephVersion(version *cephver.CephVersion) error {
 	if !c.Spec.External.Enable {
 		if !version.IsAtLeast(cephver.Minimum) {
-			return errors.Errorf("the version does not meet the minimum version: %s", cephver.Minimum.String())
+			return errors.Errorf("the version does not meet the minimum version: %q", cephver.Minimum.String())
 		}
 
 		if !version.Supported() {
-			logger.Warningf("unsupported ceph version detected: %s.", version)
 			if !c.Spec.CephVersion.AllowUnsupported {
 				return errors.Errorf("allowUnsupported must be set to true to run with this version: %+v", version)
 			}
+			logger.Warningf("unsupported ceph version detected: %q, pursuing", version)
 		}
 	}
 

pkg/operator/ceph/cluster/controller.go

@@ -594,6 +594,11 @@ func (c *ClusterController) onUpdate(oldObj, newObj interface{}) {
 		c.osdChecker.Update(newClust.Spec.RemoveOSDsIfOutAndSafeToRemove)
 	}
 
+	logger.Debugf("old cluster: %+v", oldClust.Spec)
+	logger.Debugf("new cluster: %+v", newClust.Spec)
+
+	cluster.Spec = &newClust.Spec
+
 	// if the image changed, we need to detect the new image version
 	versionChanged := false
 	if oldClust.Spec.CephVersion.Image != newClust.Spec.CephVersion.Image {
@@ -607,11 +612,6 @@ func (c *ClusterController) onUpdate(oldObj, newObj interface{}) {
 		cluster.Info.CephVersion = *version
 	}
 
-	logger.Debugf("old cluster: %+v", oldClust.Spec)
-	logger.Debugf("new cluster: %+v", newClust.Spec)
-
-	cluster.Spec = &newClust.Spec
-
 	// Get cluster running versions
 	versions, err := client.GetAllCephDaemonVersions(c.context, cluster.Namespace)
 	if err != nil {

pkg/operator/ceph/cluster/crash/crash.go

@@ -76,13 +76,19 @@ func (r *ReconcileNode) createOrUpdateCephCrash(node corev1.Node, tolerations []
 		deploymentLabels[string(config.CrashType)] = "crash"
 		deploymentLabels["ceph_daemon_id"] = "crash"
 
+		selectorLabels := map[string]string{
+			corev1.LabelHostname: nodeHostnameLabel,
+			k8sutil.AppAttr:      AppName,
+			NodeNameLabel:        node.GetName(),
+		}
+
 		nodeSelector := map[string]string{corev1.LabelHostname: nodeHostnameLabel}
 
 		// Deployment selector is immutable so we set this value only if
 		// a new object is going to be created
 		if deploy.ObjectMeta.CreationTimestamp.IsZero() {
 			deploy.Spec.Selector = &metav1.LabelSelector{
-				MatchLabels: deploymentLabels,
+				MatchLabels: selectorLabels,
 			}
 		}
 

pkg/operator/ceph/cluster/crash/reconcile.go

@@ -178,9 +178,10 @@ func getImageVersion(image string) (*version.CephVersion, bool) {
 	for i := 0; i < getVersionMaxRetries; i++ {
 		cephVersion, ok := version.GetImageVersion(image)
 		if ok {
+			logger.Debugf("ceph version found %+v", cephVersion)
 			return cephVersion, true
 		}
-		<-time.After(getVersionRetryInterval)
+		<-time.After(time.Second * getVersionRetryInterval)
 	}
 	return nil, false
 }

pkg/operator/ceph/csi/secrets.go

@@ -107,7 +107,7 @@ func cephCSIKeyringCephFSProvisionerCaps() []string {
 	}
 }
 
-func createOrUpdateCSISecret(namespace, csiRBDProvisionerSecretKey, csiRBDNodeSecretKey, csiCephFSProvisionerSecretKey, csiCephFSNodeSecretKey string, k *keyring.SecretStore) error {
+func createOrUpdateCSISecret(namespace, csiRBDProvisionerSecretKey, csiRBDNodeSecretKey, csiCephFSProvisionerSecretKey, csiCephFSNodeSecretKey string, k *keyring.SecretStore, ownerRef *metav1.OwnerReference) error {
 	csiRBDProvisionerSecrets := map[string][]byte{
 		// userID is expected for the rbd provisioner driver
 		"userID":  []byte("csi-rbd-provisioner"),
@@ -147,6 +147,7 @@ func createOrUpdateCSISecret(namespace, csiRBDProvisionerSecretKey, csiRBDNodeSe
 			Data: secret,
 			Type: k8sutil.RookType,
 		}
+		k8sutil.SetOwnerRef(&s.ObjectMeta, ownerRef)
 
 		// Create Kubernetes Secret
 		err := k.CreateSecret(s)
@@ -189,7 +190,7 @@ func CreateCSISecrets(context *clusterd.Context, clusterName string, ownerRef *m
 	}
 
 	// Create or update Kubernetes CSI secret
-	if err := createOrUpdateCSISecret(clusterName, csiRBDProvisionerSecretKey, csiRBDNodeSecretKey, csiCephFSProvisionerSecretKey, csiCephFSNodeSecretKey, k); err != nil {
+	if err := createOrUpdateCSISecret(clusterName, csiRBDProvisionerSecretKey, csiRBDNodeSecretKey, csiCephFSProvisionerSecretKey, csiCephFSNodeSecretKey, k, ownerRef); err != nil {
 		return errors.Wrapf(err, "failed to create kubernetes csi secret")
 	}
 

pkg/operator/ceph/spec/label.go

@@ -38,8 +38,8 @@ const (
 func addCephVersionLabel(cephVersion version.CephVersion, labels map[string]string) {
 	// cephVersion.String() returns a string with a space in it, and labels in k8s are limited to
 	// alphanum characters plus '-', '_', '.'
-	labels[CephVersionLabelKey] = fmt.Sprintf("%d.%d.%d",
-		cephVersion.Major, cephVersion.Minor, cephVersion.Extra)
+	labels[CephVersionLabelKey] = fmt.Sprintf("%d.%d.%d-%d",
+		cephVersion.Major, cephVersion.Minor, cephVersion.Extra, cephVersion.Build)
 }
 
 // AddCephVersionLabelToDeployment adds a label reporting the Ceph version which Rook has detected is