controller/volume: avoid to wait all replica healthy

    The case we met is that we have three replicas, one of which is not
    ready. Then the engine state could not be changed to running. Seems
    we have enough replica for the engine running on the above case.

    So we could continue next replica checking if it is not running
    instead skip this round. A healthy should be OK for the engine
    to run.

Signed-off-by: Vicente Cheng <vicente.cheng@suse.com>

controller/volume_controller.go

@@ -1690,6 +1690,7 @@ func (c *VolumeController) openVolumeDependentResources(v *longhorn.Volume, e *l
 	}
 
 	replicaAddressMap := map[string]string{}
+	anyRunningReplica := false
 	for _, r := range rs {
 		// Ignore unscheduled replicas
 		if r.Spec.NodeID == "" {
@@ -1707,10 +1708,12 @@ func (c *VolumeController) openVolumeDependentResources(v *longhorn.Volume, e *l
 		if r.Status.CurrentState == longhorn.InstanceStateError {
 			continue
 		}
-		// wait for all potentially healthy replicas become running
-		if r.Status.CurrentState != longhorn.InstanceStateRunning {
-			log.WithField("replica", r.Name).Debug("Replica is not running yet, just skip this round reconcile.")
-			return nil
+		// wait at least one replica is running
+		if r.Status.CurrentState == longhorn.InstanceStateRunning {
+			anyRunningReplica = true
+		} else {
+			log.WithField("replica", r.Name).Debug("Replica is not running yet, just continue to check next.")
+			continue
 		}
 		if r.Status.IP == "" {
 			log.WithField("replica", r.Name).Warn("Replica is running but IP is empty")
@@ -1726,6 +1729,9 @@ func (c *VolumeController) openVolumeDependentResources(v *longhorn.Volume, e *l
 		}
 		replicaAddressMap[r.Name] = imutil.GetURL(r.Status.StorageIP, r.Status.Port)
 	}
+	if !anyRunningReplica {
+		return nil
+	}
 	if len(replicaAddressMap) == 0 {
 		return fmt.Errorf("no healthy or scheduled replica for starting")
 	}