Fix panic on failure to get loadbalancer status (kubernetes#2512)

After loadbalancer creation, the controller calls a wait function that
polls the Octavia API until its state turns ACTIVE, or timeout is
reached.

Before this patch, a failure to GET the loadbalancer in the wait
function would result in a nil loadbalancer to be returned to the
caller, resulting in an immediate panic when accessing its members.

With this patch:
* GET failures are logged and don't break the polling;
* if the wait timeout is reached while GET is failing, the behaviour
  exactly matches what happens when reaching the timeout.

Author: pierreprinetti

pkg/openstack/loadbalancer.go

@@ -302,7 +302,7 @@ func (lbaas *LbaasV2) createOctaviaLoadBalancer(name, clusterName string, servic
 	}
 
 	if loadbalancer, err = openstackutil.WaitActiveAndGetLoadBalancer(lbaas.lb, loadbalancer.ID); err != nil {
-		if loadbalancer.ProvisioningStatus == errorStatus {
+		if loadbalancer != nil && loadbalancer.ProvisioningStatus == errorStatus {
 			// If LB landed in ERROR state we should delete it and retry the creation later.
 			if err = lbaas.deleteLoadBalancer(loadbalancer, service, svcConf, true); err != nil {
 				return nil, fmt.Errorf("loadbalancer %s is in ERROR state and there was an error when removing it: %v", loadbalancer.ID, err)
@@ -1896,7 +1896,7 @@ func (lbaas *LbaasV2) deleteFIPIfCreatedByProvider(fip *floatingips.FloatingIP,
 	return true, nil
 }
 
-// deleteLoadBalancer removes the LB and it's children either by using Octavia cascade deletion or manually
+// deleteLoadBalancer removes the LB and its children either by using Octavia cascade deletion or manually
 func (lbaas *LbaasV2) deleteLoadBalancer(loadbalancer *loadbalancers.LoadBalancer, service *corev1.Service, svcConf *serviceConfig, needDeleteLB bool) error {
 	if needDeleteLB && lbaas.opts.CascadeDelete {
 		klog.InfoS("Deleting load balancer", "lbID", loadbalancer.ID, "service", klog.KObj(service))

pkg/util/openstack/loadbalancer.go

@@ -177,7 +177,8 @@ func WaitActiveAndGetLoadBalancer(client *gophercloud.ServiceClient, loadbalance
 		var err error
 		loadbalancer, err = loadbalancers.Get(client, loadbalancerID).Extract()
 		if mc.ObserveRequest(err) != nil {
-			return false, err
+			klog.Warningf("Failed to fetch loadbalancer status from OpenStack (lbID %q): %s", loadbalancerID, err)
+			return false, nil
 		}
 		if loadbalancer.ProvisioningStatus == activeStatus {
 			klog.InfoS("Load balancer ACTIVE", "lbID", loadbalancerID)