reconcileClusterMachineTermination should return nil if explicit enqueue is done and reconcileClusterMachineClass() should avoid redundant processing of machines (#1056)

Author: r4mek

pkg/util/provider/machinecontroller/machine.go

@@ -277,7 +277,7 @@ func (c *controller) reconcileClusterMachineTermination(key string) error {
 	if err != nil {
 		klog.Errorf("cannot reconcile machine %q: %s", machine.Name, err)
 		c.enqueueMachineTerminationAfter(machine, time.Duration(retry), err.Error())
-		return err
+		return nil
 	}
 
 	// Process a delete event
@@ -292,15 +292,14 @@ func (c *controller) reconcileClusterMachineTermination(key string) error {
 
 	if err != nil {
 		c.enqueueMachineTerminationAfter(machine, time.Duration(retryPeriod), err.Error())
-		return err
 	} else {
 		// If the informer loses connection to the API server it may need to resync.
 		// If a resource is deleted while the watch is down, the informer won’t get
 		// delete event because the object is already gone. To avoid this edge-case,
 		// a requeue is scheduled post machine deletion as well.
 		c.enqueueMachineTerminationAfter(machine, time.Duration(retryPeriod), "post-deletion reconcile")
-		return nil
 	}
+	return nil
 }
 
 /*

pkg/util/provider/machinecontroller/machineclass.go

@@ -88,7 +88,7 @@ func (c *controller) reconcileClusterMachineClassKey(key string) error {
 		c.enqueueMachineClassAfter(class, time.Duration(machineutils.ShortRetry))
 	} else {
 		// Re-enqueue periodically to avoid missing of events
-		// TODO: Get ride of this logic
+		// TODO: Get rid of this logic
 		c.enqueueMachineClassAfter(class, time.Duration(machineutils.LongRetry))
 	}
 
@@ -131,19 +131,18 @@ func (c *controller) reconcileClusterMachineClass(ctx context.Context, class *v1
 		}
 
 		for _, m := range machines {
-			c.enqueueMachine(m, reason)
+			// Enqueue only the non-terminating machines to support changes to virtual capacity to nodes.
+			// Changes to virtual capacity need not be reflected on terminating machines as they are being deleted.
+			if m.DeletionTimestamp == nil {
+				c.enqueueMachine(m, reason)
+			}
 		}
 		return nil
 	}
 
 	if len(machines) > 0 {
 		// Machines are still referring the machine class, please wait before deletion
 		klog.V(3).Infof("Cannot remove finalizer on %s because still (%d) machines are referencing it", class.Name, len(machines))
-
-		for _, machine := range machines {
-			c.addMachine(machine)
-		}
-
 		return fmt.Errorf("Retry as machine objects are still referring the machineclass")
 	}