Create multiple control plane loadbalancers concurrently

Author: mdbooth

pkg/cloud/services/elb/loadbalancer.go

@@ -67,60 +67,93 @@ const additionalTargetGroupPrefix = "additional-listener-"
 // cantAttachSGToNLBRegions is a set of regions that do not support Security Groups in NLBs.
 var cantAttachSGToNLBRegions = sets.New("us-iso-east-1", "us-iso-west-1", "us-isob-east-1")
 
+type lbReconciler func() error
+
 // ReconcileLoadbalancers reconciles the load balancers for the given cluster.
 func (s *Service) ReconcileLoadbalancers() error {
 	s.scope.Debug("Reconciling load balancers")
 
 	var errs []error
+	var lbReconcilers []lbReconciler
+
+	// The following splits load balancer reconciliation into 2 phases:
+	// 1. Get or create the load balancer
+	// 2. Reconcile the load balancer
+	// We ensure that we only wait for the load balancer to become available in
+	// the reconcile phase. This is useful when creating multiple load
+	// balancers, as they can take several minutes to become available.
 
 	for _, lbSpec := range s.scope.ControlPlaneLoadBalancers() {
 		if lbSpec == nil {
 			continue
 		}
 		switch lbSpec.LoadBalancerType {
 		case infrav1.LoadBalancerTypeClassic:
-			errs = append(errs, s.reconcileClassicLoadBalancer())
+			reconciler, err := s.getOrCreateClassicLoadBalancer()
+			if err != nil {
+				errs = append(errs, err)
+			} else {
+				lbReconcilers = append(lbReconcilers, reconciler)
+			}
 		case infrav1.LoadBalancerTypeNLB, infrav1.LoadBalancerTypeALB, infrav1.LoadBalancerTypeELB:
-			errs = append(errs, s.reconcileV2LB(lbSpec))
+			reconciler, err := s.getOrCreateV2LB(lbSpec)
+			if err != nil {
+				errs = append(errs, err)
+			} else {
+				lbReconcilers = append(lbReconcilers, reconciler)
+			}
 		default:
 			errs = append(errs, fmt.Errorf("unknown or unsupported load balancer type on primary load balancer: %s", lbSpec.LoadBalancerType))
 		}
 	}
 
+	// Reconcile all load balancers
+	for _, reconciler := range lbReconcilers {
+		if err := reconciler(); err != nil {
+			errs = append(errs, err)
+		}
+	}
+
 	return kerrors.NewAggregate(errs)
 }
 
-// reconcileV2LB creates a load balancer. It also takes care of generating unique names across
+// getOrCreateV2LB creates a load balancer. It also takes care of generating unique names across
 // namespaces by appending the namespace to the name.
-func (s *Service) reconcileV2LB(lbSpec *infrav1.AWSLoadBalancerSpec) error {
+func (s *Service) getOrCreateV2LB(lbSpec *infrav1.AWSLoadBalancerSpec) (lbReconciler, error) {
 	name, err := LBName(s.scope, lbSpec)
 	if err != nil {
-		return errors.Wrap(err, "failed to get control plane load balancer name")
+		return nil, errors.Wrap(err, "failed to get control plane load balancer name")
 	}
 
 	// Get default api server spec.
 	desiredLB, err := s.getAPIServerLBSpec(name, lbSpec)
 	if err != nil {
-		return err
+		return nil, err
 	}
 	lb, err := s.describeLB(name, lbSpec)
 	switch {
 	case IsNotFound(err) && s.scope.ControlPlaneEndpoint().IsValid():
 		// if elb is not found and owner cluster ControlPlaneEndpoint is already populated, then we should not recreate the elb.
-		return errors.Wrapf(err, "no loadbalancer exists for the AWSCluster %s, the cluster has become unrecoverable and should be deleted manually", s.scope.InfraClusterName())
+		return nil, errors.Wrapf(err, "no loadbalancer exists for the AWSCluster %s, the cluster has become unrecoverable and should be deleted manually", s.scope.InfraClusterName())
 	case IsNotFound(err):
 		lb, err = s.createLB(desiredLB, lbSpec)
 		if err != nil {
 			s.scope.Error(err, "failed to create LB")
-			return err
+			return nil, err
 		}
 
 		s.scope.Debug("Created new network load balancer for apiserver", "api-server-lb-name", lb.Name)
 	case err != nil:
 		// Failed to describe the classic ELB
-		return err
+		return nil, err
 	}
 
+	return func() error {
+		return s.reconcileV2LB(lb, desiredLB, lbSpec)
+	}, nil
+}
+
+func (s *Service) reconcileV2LB(lb *infrav1.LoadBalancer, desiredLB *infrav1.LoadBalancer, lbSpec *infrav1.AWSLoadBalancerSpec) error {
 	wReq := &elbv2.DescribeLoadBalancersInput{
 		LoadBalancerArns: aws.StringSlice([]string{lb.ARN}),
 	}
@@ -507,35 +540,61 @@ func (s *Service) describeLB(name string, lbSpec *infrav1.AWSLoadBalancerSpec) (
 	return fromSDKTypeToLB(out.LoadBalancers[0], outAtt.Attributes, tags), nil
 }
 
-func (s *Service) reconcileClassicLoadBalancer() error {
+func (s *Service) getOrCreateClassicLoadBalancer() (lbReconciler, error) {
 	// Generate a default control plane load balancer name. The load balancer name cannot be
 	// generated by the defaulting webhook, because it is derived from the cluster name, and that
 	// name is undefined at defaulting time when generateName is used.
 	name, err := ELBName(s.scope)
 	if err != nil {
-		return errors.Wrap(err, "failed to get control plane load balancer name")
+		return nil, errors.Wrap(err, "failed to get control plane load balancer name")
 	}
 
 	// Get default api server spec.
 	spec, err := s.getAPIServerClassicELBSpec(name)
 	if err != nil {
-		return err
+		return nil, err
 	}
 
 	apiELB, err := s.describeClassicELB(spec.Name)
 	switch {
 	case IsNotFound(err) && s.scope.ControlPlaneEndpoint().IsValid():
 		// if elb is not found and owner cluster ControlPlaneEndpoint is already populated, then we should not recreate the elb.
-		return errors.Wrapf(err, "no loadbalancer exists for the AWSCluster %s, the cluster has become unrecoverable and should be deleted manually", s.scope.InfraClusterName())
+		return nil, errors.Wrapf(err, "no loadbalancer exists for the AWSCluster %s, the cluster has become unrecoverable and should be deleted manually", s.scope.InfraClusterName())
 	case IsNotFound(err):
 		apiELB, err = s.createClassicELB(spec)
 		if err != nil {
-			return err
+			return nil, err
 		}
 		s.scope.Debug("Created new classic load balancer for apiserver", "api-server-elb-name", apiELB.Name)
 	case err != nil:
 		// Failed to describe the classic ELB
-		return err
+		return nil, err
+	}
+
+	return func() error {
+		return s.reconcileClassicLoadBalancer(apiELB, spec)
+	}, nil
+}
+
+func (s *Service) reconcileClassicLoadBalancer(apiELB *infrav1.LoadBalancer, spec *infrav1.LoadBalancer) error {
+	if spec.HealthCheck != nil {
+		if err := wait.WaitForWithRetryable(wait.NewBackoff(), func() (bool, error) {
+			if _, err := s.ELBClient.ConfigureHealthCheck(&elb.ConfigureHealthCheckInput{
+				LoadBalancerName: aws.String(spec.Name),
+				HealthCheck: &elb.HealthCheck{
+					Target:             aws.String(spec.HealthCheck.Target),
+					Interval:           aws.Int64(int64(spec.HealthCheck.Interval.Seconds())),
+					Timeout:            aws.Int64(int64(spec.HealthCheck.Timeout.Seconds())),
+					HealthyThreshold:   aws.Int64(spec.HealthCheck.HealthyThreshold),
+					UnhealthyThreshold: aws.Int64(spec.HealthCheck.UnhealthyThreshold),
+				},
+			}); err != nil {
+				return false, err
+			}
+			return true, nil
+		}, awserrors.LoadBalancerNotFound); err != nil {
+			return errors.Wrapf(err, "failed to configure health check for classic load balancer: %v", spec)
+		}
 	}
 
 	if apiELB.IsManaged(s.scope.Name()) {
@@ -1193,26 +1252,6 @@ func (s *Service) createClassicELB(spec *infrav1.LoadBalancer) (*infrav1.LoadBal
 		return nil, errors.Wrapf(err, "failed to create classic load balancer: %v", spec)
 	}
 
-	if spec.HealthCheck != nil {
-		if err := wait.WaitForWithRetryable(wait.NewBackoff(), func() (bool, error) {
-			if _, err := s.ELBClient.ConfigureHealthCheck(&elb.ConfigureHealthCheckInput{
-				LoadBalancerName: aws.String(spec.Name),
-				HealthCheck: &elb.HealthCheck{
-					Target:             aws.String(spec.HealthCheck.Target),
-					Interval:           aws.Int64(int64(spec.HealthCheck.Interval.Seconds())),
-					Timeout:            aws.Int64(int64(spec.HealthCheck.Timeout.Seconds())),
-					HealthyThreshold:   aws.Int64(spec.HealthCheck.HealthyThreshold),
-					UnhealthyThreshold: aws.Int64(spec.HealthCheck.UnhealthyThreshold),
-				},
-			}); err != nil {
-				return false, err
-			}
-			return true, nil
-		}, awserrors.LoadBalancerNotFound); err != nil {
-			return nil, errors.Wrapf(err, "failed to configure health check for classic load balancer: %v", spec)
-		}
-	}
-
 	s.scope.Info("Created classic load balancer", "dns-name", *out.DNSName)
 
 	res := spec.DeepCopy()

pkg/cloud/services/elb/loadbalancer_test.go

@@ -2546,7 +2546,10 @@ func TestReconcileV2LB(t *testing.T) {
 				scope:       clusterScope,
 				ELBV2Client: elbV2APIMocks,
 			}
-			err = s.reconcileV2LB(clusterScope.ControlPlaneLoadBalancer())
+			reconciler, err := s.getOrCreateV2LB(clusterScope.ControlPlaneLoadBalancer())
+			if err == nil {
+				err = reconciler()
+			}
 			lb := s.scope.Network().APIServerELB
 
 			tc.check(t, &lb, err)