Drop Node event when EC2 instance does not exist

pkg/controllers/tagging/tagging_controller.go

@@ -292,6 +292,17 @@ func (tc *Controller) tagEc2Instance(node *v1.Node) error {
 	err := tc.cloud.TagResource(string(instanceID), tc.tags)
 
 	if err != nil {
+		if awsv1.IsAWSErrorInstanceNotFound(err) {
+			// This can happen for two reasons:
+			// 1. The CreateTags API is eventually consistent. In rare cases, a newly-created instance may not be taggable for a short period.
+			// 2. The event in our workQueue is stale, and the instance no longer exists. Tagging will never succeed.
+			//
+			// In either case, we don't return an error and requeue the current node event.
+			// In the first case, we will attempt tagging on the next node update event (in 5 minutes); which is an acceptable delay in the rare case.
+			// In the second case this would increase the workQueue size (causing higher dequeue latency for valid events).
+			klog.Infof("Skip tagging since EC2 instance %s for node %s does not exist", instanceID, node.GetName())
+			return nil
+		}
 		klog.Errorf("Error in tagging EC2 instance %s for node %s, error: %v", instanceID, node.GetName(), err)
 		return err
 	}

pkg/providers/v1/aws.go

@@ -1707,7 +1707,7 @@ func (c *Cloud) InstanceExistsByProviderID(ctx context.Context, providerID strin
 	instances, err := c.ec2.DescribeInstances(request)
 	if err != nil {
 		// if err is InstanceNotFound, return false with no error
-		if isAWSErrorInstanceNotFound(err) {
+		if IsAWSErrorInstanceNotFound(err) {
 			return false, nil
 		}
 		return false, err
@@ -1946,7 +1946,7 @@ func (c *Cloud) GetZoneByNodeName(ctx context.Context, nodeName types.NodeName)
 
 }
 
-func isAWSErrorInstanceNotFound(err error) bool {
+func IsAWSErrorInstanceNotFound(err error) bool {
 	if err == nil {
 		return false
 	}