update node_controller.go

Reproduction steps:
1. Start mysql-operator.
2. Start the mysql instance.
3. Wait for all mysql instances to run successfully.
4. Close mysql-operator (to simulate network failure, node down, etc.).
5. Restart the node containing the mysql instance.
6. The mysql instances on these nodes will start automatically, but since init will set configfure to 0.
7. Wait for all these nodes to stop changing.
8. Start mysql-operator, which will not automatically repair the current cluster because it is listening to the updated pod.
9. and the mysql cluster is in an unhealthy state due to configfure=0.

Author: jiuker

pkg/controller/node/node_controller.go

@@ -120,9 +120,8 @@ func add(mgr manager.Manager, r reconcile.Reconciler) error {
 
 	// Watch for changes to MysqlCluster
 	err = c.Watch(&source.Kind{Type: &corev1.Pod{}}, &handler.EnqueueRequestForObject{}, predicate.Funcs{
-		// no need to init nodes when are created
 		CreateFunc: func(evt event.CreateEvent) bool {
-			return false
+			return isOwnedByMySQL(evt.Object) && isRunning(evt.Object) && !isReady(evt.Object)
 		},
 
 		// trigger node initialization only on pod update, after pod is created for a while