Merge pull request https://github.com/kubernetes/contrib/pull/986 from mwielgus/more_debug

Cluster autoscaler: more debug in estimates and pod conditions

cluster-autoscaler/cluster_autoscaler.go

@@ -168,7 +168,9 @@ func main() {
 						continue
 					}
 					node := nodeInfo.Node()
-					estimate := bestOption.estimator.Estimate(node)
+					estimate, report := bestOption.estimator.Estimate(node)
+					glog.V(1).Info(bestOption.estimator.GetDebug())
+					glog.V(1).Info(report)
 					glog.V(1).Infof("Estimated %d nodes needed in %s", estimate, bestOption.migConfig.Url())
 
 					currentSize, err := gceManager.GetMigSize(bestOption.migConfig)

cluster-autoscaler/estimator/estimator.go

@@ -17,6 +17,8 @@ limitations under the License.
 package estimator
 
 import (
+	"bytes"
+	"fmt"
 	"math"
 
 	kube_api "k8s.io/kubernetes/pkg/api"
@@ -78,23 +80,41 @@ func maxInt(a, b int) int {
 	return b
 }
 
+// GetDebug returns debug information about the current state of BasicNodeEstimator
+func (basicEstimator *BasicNodeEstimator) GetDebug() string {
+	var buffer bytes.Buffer
+	buffer.WriteString("Resources needed:\n")
+	buffer.WriteString(fmt.Sprintf("CPU: %s\n", basicEstimator.cpuSum.String()))
+	buffer.WriteString(fmt.Sprintf("Mem: %s\n", basicEstimator.memorySum.String()))
+	for port, count := range basicEstimator.portSum {
+		buffer.WriteString(fmt.Sprintf("Port %d: %d\n", port, count))
+	}
+	return buffer.String()
+}
+
 // Estimate estimates the number needed of nodes of the given shape.
-func (basicEstimator *BasicNodeEstimator) Estimate(node *kube_api.Node) int {
+func (basicEstimator *BasicNodeEstimator) Estimate(node *kube_api.Node) (int, string) {
+	var buffer bytes.Buffer
+	buffer.WriteString("Needed nodes according to:\n")
 	result := 0
 	if cpuCapcaity, ok := node.Status.Capacity[kube_api.ResourceCPU]; ok {
-		result = maxInt(result,
-			int(math.Ceil(float64(basicEstimator.cpuSum.MilliValue())/float64(cpuCapcaity.MilliValue()))))
+		prop := int(math.Ceil(float64(basicEstimator.cpuSum.MilliValue()) / float64(cpuCapcaity.MilliValue())))
+		buffer.WriteString(fmt.Sprintf("CPU: %d\n", prop))
+		result = maxInt(result, prop)
 	}
 	if memCapcaity, ok := node.Status.Capacity[kube_api.ResourceMemory]; ok {
-		result = maxInt(result,
-			int(math.Ceil(float64(basicEstimator.memorySum.Value())/float64(memCapcaity.Value()))))
+		prop := int(math.Ceil(float64(basicEstimator.memorySum.Value()) / float64(memCapcaity.Value())))
+		buffer.WriteString(fmt.Sprintf("Mem: %d\n", prop))
+		result = maxInt(result, prop)
 	}
 	if podCapcaity, ok := node.Status.Capacity[kube_api.ResourcePods]; ok {
-		result = maxInt(result,
-			int(math.Ceil(float64(basicEstimator.count)/float64(podCapcaity.Value()))))
+		prop := int(math.Ceil(float64(basicEstimator.count) / float64(podCapcaity.Value())))
+		buffer.WriteString(fmt.Sprintf("Pods: %d\n", prop))
+		result = maxInt(result, prop)
 	}
-	for _, count := range basicEstimator.portSum {
+	for port, count := range basicEstimator.portSum {
+		buffer.WriteString(fmt.Sprintf("Port %d: %d\n", port, count))
 		result = maxInt(result, count)
 	}
-	return result
+	return result, buffer.String()
 }

cluster-autoscaler/estimator/estimator_test.go

@@ -63,8 +63,10 @@ func TestEstimate(t *testing.T) {
 			},
 		},
 	}
-
-	assert.Equal(t, 3, estimator.Estimate(node))
+	estimate, report := estimator.Estimate(node)
+	assert.Contains(t, estimator.GetDebug(), "CPU")
+	assert.Contains(t, report, "CPU")
+	assert.Equal(t, 3, estimate)
 }
 
 func TestEstimateWithPorts(t *testing.T) {
@@ -105,5 +107,8 @@ func TestEstimateWithPorts(t *testing.T) {
 		},
 	}
 
-	assert.Equal(t, 5, estimator.Estimate(node))
+	estimate, report := estimator.Estimate(node)
+	assert.Contains(t, estimator.GetDebug(), "CPU")
+	assert.Contains(t, report, "CPU")
+	assert.Equal(t, 5, estimate)
 }

cluster-autoscaler/utils.go

@@ -169,6 +169,8 @@ func resetOldPods(kubeClient *kube_client.Client, pods []*kube_api.Pod, threshol
 	for _, pod := range pods {
 		_, condition := kube_api.GetPodCondition(&pod.Status, kube_api.PodScheduled)
 		if condition != nil && !condition.LastTransitionTime.After(threshold) {
+			glog.V(4).Infof("Reseting pod condition for %s/%s, last transition: %s",
+				pod.Namespace, pod.Name, condition.LastTransitionTime.Time.String())
 			if err := resetPodScheduledCondition(kubeClient, pod); err != nil {
 				glog.Errorf("Error during reseting pod condition for %s/%s: %v", pod.Namespace, pod.Name, err)
 			}