fix/assing-no-public (#138)

* try to assign static public IP if node missing ephmeral IP address

* keep daemonset running; avoid CrashLoopBackOff

Author: alexei-led

cmd/main.go

@@ -88,12 +88,22 @@ func assignAddress(c context.Context, log *logrus.Entry, assigner address.Assign
 	defer ticker.Stop()
 
 	for retryCounter := 0; retryCounter <= cfg.RetryAttempts; retryCounter++ {
+		log.WithFields(logrus.Fields{
+			"node":           node.Name,
+			"instance":       node.Instance,
+			"filter":         cfg.Filter,
+			"retry-counter":  retryCounter,
+			"retry-attempts": cfg.RetryAttempts,
+		}).Debug("assigning static public IP address to node")
 		err := assigner.Assign(ctx, node.Instance, node.Zone, cfg.Filter, cfg.OrderBy)
 		if err == nil || errors.Is(err, address.ErrStaticIPAlreadyAssigned) {
 			return nil
 		}
 
-		log.WithError(err).Errorf("failed to assign static public IP address to node %s", node.Name)
+		log.WithError(err).WithFields(logrus.Fields{
+			"node":     node.Name,
+			"instance": node.Instance,
+		}).Error("failed to assign static public IP address to node")
 		log.Infof("retrying after %v", cfg.RetryInterval)
 
 		select {
@@ -148,26 +158,33 @@ func run(c context.Context, log *logrus.Entry, cfg *config.Config) error {
 		}
 	}()
 
-	select {
-	case err = <-errorCh:
-		if err != nil {
-			return errors.Wrap(err, "assigning static public IP address")
-		}
-	case <-ctx.Done():
-		log.Infof("kubeip agent gracefully stopped")
-		if cfg.ReleaseOnExit {
-			log.Infof("releasing static public IP address")
-			releaseCtx, releaseCancel := context.WithTimeout(context.Background(), unassignTimeout) // release the static public IP address within 5 minutes
-			defer releaseCancel()
-			// use a different context for releasing the static public IP address since the main context is canceled
-			if err = assigner.Unassign(releaseCtx, n.Instance, n.Zone); err != nil { //nolint:contextcheck
-				return errors.Wrap(err, "failed to release static public IP address")
+	for {
+		select {
+		case err = <-errorCh:
+			if err != nil {
+				return errors.Wrap(err, "assigning static public IP address")
+			}
+		case <-ctx.Done():
+			log.Infof("kubeip agent gracefully stopped")
+			if cfg.ReleaseOnExit {
+				log.Infof("releasing static public IP address")
+				err = func() error {
+					releaseCtx, releaseCancel := context.WithTimeout(context.Background(), unassignTimeout) // release the static public IP address within 5 minutes
+					defer releaseCancel()
+					// use a different context for releasing the static public IP address since the main context is canceled
+					if err = assigner.Unassign(releaseCtx, n.Instance, n.Zone); err != nil {
+						return errors.Wrap(err, "failed to release static public IP address")
+					}
+					return nil
+				}()
+				if err != nil {
+					return err //nolint:wrapcheck
+				}
+				log.Infof("static public IP address released")
 			}
-			log.Infof("static public IP address released")
+			return nil
 		}
 	}
-
-	return nil
 }
 
 func runCmd(c *cli.Context) error {

internal/address/gcp.go

@@ -28,6 +28,10 @@ const (
 	maxRetries                  = 10 // number of retries for assigning ephemeral public IP address
 )
 
+var (
+	ErrNoPublicIPAssigned = errors.New("no public IP address assigned to the instance")
+)
+
 type internalAssigner interface {
 	CheckAddressAssigned(region, addressName string) (bool, error)
 	AddInstanceAddress(ctx context.Context, instance *compute.Instance, zone string, address *compute.Address) error
@@ -233,7 +237,7 @@ func (a *gcpAssigner) Assign(ctx context.Context, instanceID, zone string, filte
 	a.logger.WithField("addresses", ips).Debugf("found %d available addresses", len(addresses))
 
 	// delete current ephemeral public IP address
-	if err = a.DeleteInstanceAddress(ctx, instance, zone); err != nil {
+	if err = a.DeleteInstanceAddress(ctx, instance, zone); err != nil && !errors.Is(err, ErrNoPublicIPAssigned) {
 		return errors.Wrap(err, "failed to delete current public IP address")
 	}
 
@@ -359,12 +363,12 @@ func (a *gcpAssigner) Unassign(ctx context.Context, instanceID, zone string) err
 func getAccessConfig(networkInterface *compute.NetworkInterface, ipv6 bool) (*compute.AccessConfig, error) {
 	if ipv6 {
 		if len(networkInterface.Ipv6AccessConfigs) == 0 {
-			return nil, errors.New("instance network interface has no IPv6 access configs")
+			return nil, ErrNoPublicIPAssigned
 		}
 		return networkInterface.Ipv6AccessConfigs[0], nil
 	}
 	if len(networkInterface.AccessConfigs) == 0 {
-		return nil, errors.New("instance network interface has no access configs")
+		return nil, ErrNoPublicIPAssigned
 	}
 	return networkInterface.AccessConfigs[0], nil
 }