Merge pull request #2502 from mwielgus/ca-starting-fix

Automatic merge from submit-queue

Cluster-autoscaler: Fix isNodeStarting

Fix for: kubernetes/kubernetes#43709

cc: @MaciekPytel @fgrzadkowski

cluster-autoscaler/Makefile

@@ -21,7 +21,7 @@ ifndef REGISTRY
 	$(ERR)
 endif
 	docker build --pull -t ${REGISTRY}/cluster-autoscaler:${TAG} .
-	gcloud docker push ${REGISTRY}/cluster-autoscaler:${TAG}
+	gcloud docker -- push ${REGISTRY}/cluster-autoscaler:${TAG}
 
 clean:
 	rm -f cluster-autoscaler

cluster-autoscaler/clusterstate/clusterstate.go

@@ -40,7 +40,7 @@ const (
 
 	// MaxStatusSettingDelayAfterCreation is the maximum time for node to set its initial status after the
 	// node is registered.
-	MaxStatusSettingDelayAfterCreation = time.Minute
+	MaxStatusSettingDelayAfterCreation = 2 * time.Minute
 )
 
 // ScaleUpRequest contains information about the requested node group scale up.
@@ -615,6 +615,16 @@ func isNodeNotStarted(node *apiv1.Node) bool {
 			condition.LastTransitionTime.Time.Sub(node.CreationTimestamp.Time) < MaxStatusSettingDelayAfterCreation {
 			return true
 		}
+		if condition.Type == apiv1.NodeOutOfDisk &&
+			condition.Status == apiv1.ConditionTrue &&
+			condition.LastTransitionTime.Time.Sub(node.CreationTimestamp.Time) < MaxStatusSettingDelayAfterCreation {
+			return true
+		}
+		if condition.Type == apiv1.NodeNetworkUnavailable &&
+			condition.Status == apiv1.ConditionTrue &&
+			condition.LastTransitionTime.Time.Sub(node.CreationTimestamp.Time) < MaxStatusSettingDelayAfterCreation {
+			return true
+		}
 	}
 	return false
 }

cluster-autoscaler/core/scale_up.go

@@ -59,6 +59,7 @@ func ScaleUp(context *AutoscalingContext, unschedulablePods []*apiv1.Pod, nodes
 			upcomingNodes = append(upcomingNodes, nodeTemplate)
 		}
 	}
+	glog.V(4).Infof("Upcoming %d nodes", len(upcomingNodes))
 
 	podsRemainUnschedulable := make(map[*apiv1.Pod]bool)
 	expansionOptions := make([]expander.Option, 0)
@@ -119,12 +120,16 @@ func ScaleUp(context *AutoscalingContext, unschedulablePods []*apiv1.Pod, nodes
 			}
 			if option.NodeCount > 0 {
 				expansionOptions = append(expansionOptions, option)
+			} else {
+				glog.V(2).Infof("No need for any nodes in %s", nodeGroup.Id())
 			}
+		} else {
+			glog.V(4).Info("No pod can fit to %s", nodeGroup.Id())
 		}
 	}
 
 	if len(expansionOptions) == 0 {
-		glog.V(1).Info("No node group can help with pending pods.")
+		glog.V(1).Info("No expansion options")
 		for pod, unschedulable := range podsRemainUnschedulable {
 			if unschedulable {
 				context.Recorder.Event(pod, apiv1.EventTypeNormal, "NotTriggerScaleUp",

cluster-autoscaler/core/static_autoscaler.go

@@ -187,12 +187,16 @@ func (a *StaticAutoscaler) RunOnce(currentTime time.Time) {
 	// in the describe situation.
 	schedulablePodsPresent := false
 	if a.VerifyUnschedulablePods {
+
+		glog.V(4).Infof("Filtering out schedulables")
 		newUnschedulablePodsToHelp := FilterOutSchedulable(unschedulablePodsToHelp, readyNodes, allScheduled,
 			a.PredicateChecker)
 
 		if len(newUnschedulablePodsToHelp) != len(unschedulablePodsToHelp) {
 			glog.V(2).Info("Schedulable pods present")
 			schedulablePodsPresent = true
+		} else {
+			glog.V(4).Info("No schedulable pods")
 		}
 		unschedulablePodsToHelp = newUnschedulablePodsToHelp
 	}