Merge pull request #1990 from arm64b/no-valid-nodes-handling

Return error when no valid nodes available to run osd

pkg/operator/ceph/cluster/osd/osd.go

@@ -160,6 +160,11 @@ func (c *Cluster) Start() error {
 		logger.Debugf("storage nodes: %+v", c.Storage.Nodes)
 	}
 	validNodes := k8sutil.GetValidNodes(c.Storage.Nodes, c.context.Clientset, c.placement)
+	// no valid node is ready to run an osd
+	if len(validNodes) == 0 {
+		logger.Warningf("no valid node available to run an osd in namespace %s", c.Namespace)
+		return nil
+	}
 	logger.Infof("%d of the %d storage nodes are valid", len(validNodes), len(c.Storage.Nodes))
 	c.Storage.Nodes = validNodes
 	// orchestrate individual nodes, starting with any that are still ongoing (in the case that we

pkg/operator/ceph/cluster/osd/status_test.go

@@ -60,6 +60,10 @@ func TestOrchestrationStatus(t *testing.T) {
 }
 
 func mockNodeOrchestrationCompletion(c *Cluster, nodeName string, statusMapWatcher *watch.FakeWatcher) {
+	// if no valid osd node, don't need to check its status, return immediately
+	if len(c.Storage.Nodes) == 0 {
+		return
+	}
 	for {
 		// wait for the node's orchestration status to change to "starting"
 		cmName := fmt.Sprintf(orchestrationStatusMapName, nodeName)