func (cli *HeapsterMetricsClient) GetNodeMetrics(nodeName string, selector labels.Selector) ([]metricsapi.NodeMetrics, error) {
params := map[string]string{"labelSelector": selector.String()}
path, err := nodeMetricsUrl(nodeName)
if err != nil {
return []metricsapi.NodeMetrics{}, err
}
resultRaw, err := GetHeapsterMetrics(cli, path, params)
if err != nil {
return []metricsapi.NodeMetrics{}, err
}
metrics := make([]metricsapi.NodeMetrics, 0)
if len(nodeName) == 0 {
metricsList := metricsapi.NodeMetricsList{}
err = json.Unmarshal(resultRaw, &metricsList)
if err != nil {
return []metricsapi.NodeMetrics{}, fmt.Errorf("failed to unmarshall heapster response: %v", err)
}
metrics = append(metrics, metricsList.Items...)
} else {
var singleMetric metricsapi.NodeMetrics
err = json.Unmarshal(resultRaw, &singleMetric)
if err != nil {
return []metricsapi.NodeMetrics{}, fmt.Errorf("failed to unmarshall heapster response: %v", err)
}
metrics = append(metrics, singleMetric)
}
return metrics, nil
}
func (cli *HeapsterMetricsClient) GetPodMetrics(namespace string, podName string, allNamespaces bool, selector labels.Selector) ([]metricsapi.PodMetrics, error) {
if allNamespaces {
namespace = api.NamespaceAll
}
path, err := podMetricsUrl(namespace, podName)
if err != nil {
return []metricsapi.PodMetrics{}, err
}
params := map[string]string{"labelSelector": selector.String()}
allMetrics := make([]metricsapi.PodMetrics, 0)
resultRaw, err := GetHeapsterMetrics(cli, path, params)
if err != nil {
return []metricsapi.PodMetrics{}, err
}
if len(podName) == 0 {
metrics := metricsapi.PodMetricsList{}
err = json.Unmarshal(resultRaw, &metrics)
if err != nil {
return []metricsapi.PodMetrics{}, fmt.Errorf("failed to unmarshall heapster response: %v", err)
}
allMetrics = append(allMetrics, metrics.Items...)
} else {
var singleMetric metricsapi.PodMetrics
err = json.Unmarshal(resultRaw, &singleMetric)
if err != nil {
return []metricsapi.PodMetrics{}, fmt.Errorf("failed to unmarshall heapster response: %v", err)
}
allMetrics = append(allMetrics, singleMetric)
}
return allMetrics, nil
}
// LabelsSelectorParam adds the given selector as a query parameter
func (r *Request) LabelsSelectorParam(s labels.Selector) *Request {
if r.err != nil {
return r
}
if s == nil {
return r
}
if s.Empty() {
return r
}
return r.setParam(metav1.LabelSelectorQueryParam(r.content.GroupVersion.String()), s.String())
}
// getContainerRestarts returns the count of container restarts across all pods matching the given labelSelector,
// and a list of nodenames across which these containers restarted.
func getContainerRestarts(c clientset.Interface, ns string, labelSelector labels.Selector) (int, []string) {
options := v1.ListOptions{LabelSelector: labelSelector.String()}
pods, err := c.Core().Pods(ns).List(options)
framework.ExpectNoError(err)
failedContainers := 0
containerRestartNodes := sets.NewString()
for _, p := range pods.Items {
for _, v := range testutils.FailedContainers(&p) {
failedContainers = failedContainers + v.Restarts
containerRestartNodes.Insert(p.Spec.NodeName)
}
}
return failedContainers, containerRestartNodes.List()
}
// filterLabels returns a list of pods which have labels.
func filterLabels(selectors map[string]string, cli clientset.Interface, ns string) (*v1.PodList, error) {
var err error
var selector labels.Selector
var pl *v1.PodList
// List pods based on selectors. This might be a tiny optimization rather then filtering
// everything manually.
if len(selectors) > 0 {
selector = labels.SelectorFromSet(labels.Set(selectors))
options := v1.ListOptions{LabelSelector: selector.String()}
pl, err = cli.Core().Pods(ns).List(options)
} else {
pl, err = cli.Core().Pods(ns).List(v1.ListOptions{})
}
return pl, err
}
func (h *HeapsterMetricsClient) GetResourceMetric(resource v1.ResourceName, namespace string, selector labels.Selector) (PodResourceInfo, time.Time, error) {
metricPath := fmt.Sprintf("/apis/metrics/v1alpha1/namespaces/%s/pods", namespace)
params := map[string]string{"labelSelector": selector.String()}
resultRaw, err := h.services.
ProxyGet(h.heapsterScheme, h.heapsterService, h.heapsterPort, metricPath, params).
DoRaw()
if err != nil {
return nil, time.Time{}, fmt.Errorf("failed to get heapster service: %v", err)
}
glog.V(4).Infof("Heapster metrics result: %s", string(resultRaw))
metrics := metricsapi.PodMetricsList{}
err = json.Unmarshal(resultRaw, &metrics)
if err != nil {
return nil, time.Time{}, fmt.Errorf("failed to unmarshal heapster response: %v", err)
}
if len(metrics.Items) == 0 {
return nil, time.Time{}, fmt.Errorf("no metrics returned from heapster")
}
res := make(PodResourceInfo, len(metrics.Items))
for _, m := range metrics.Items {
podSum := int64(0)
missing := len(m.Containers) == 0
for _, c := range m.Containers {
resValue, found := c.Usage[v1.ResourceName(resource)]
if !found {
missing = true
glog.V(2).Infof("missing resource metric %v for container %s in pod %s/%s", resource, c.Name, namespace, m.Name)
continue
}
podSum += resValue.MilliValue()
}
if !missing {
res[m.Name] = int64(podSum)
}
}
timestamp := metrics.Items[0].Timestamp.Time
return res, timestamp, nil
}
func NewPodStore(c clientset.Interface, namespace string, label labels.Selector, field fields.Selector) *PodStore {
lw := &cache.ListWatch{
ListFunc: func(options v1.ListOptions) (runtime.Object, error) {
options.LabelSelector = label.String()
options.FieldSelector = field.String()
obj, err := c.Core().Pods(namespace).List(options)
return runtime.Object(obj), err
},
WatchFunc: func(options v1.ListOptions) (watch.Interface, error) {
options.LabelSelector = label.String()
options.FieldSelector = field.String()
return c.Core().Pods(namespace).Watch(options)
},
}
store := cache.NewStore(cache.MetaNamespaceKeyFunc)
stopCh := make(chan struct{})
reflector := cache.NewReflector(lw, &v1.Pod{}, store, 0)
reflector.RunUntil(stopCh)
return &PodStore{Store: store, stopCh: stopCh, Reflector: reflector}
}
// Wait up to 10 minutes for all matching pods to become Running and at least one
// matching pod exists.
func WaitForPodsWithLabelRunning(c clientset.Interface, ns string, label labels.Selector) error {
running := false
PodStore := NewPodStore(c, ns, label, fields.Everything())
defer PodStore.Stop()
waitLoop:
for start := time.Now(); time.Since(start) < 10*time.Minute; time.Sleep(5 * time.Second) {
pods := PodStore.List()
if len(pods) == 0 {
continue waitLoop
}
for _, p := range pods {
if p.Status.Phase != v1.PodRunning {
continue waitLoop
}
}
running = true
break
}
if !running {
return fmt.Errorf("Timeout while waiting for pods with labels %q to be running", label.String())
}
return nil
}
func (h *HeapsterMetricsClient) GetRawMetric(metricName string, namespace string, selector labels.Selector) (PodMetricsInfo, time.Time, error) {
podList, err := h.podsGetter.Pods(namespace).List(v1.ListOptions{LabelSelector: selector.String()})
if err != nil {
return nil, time.Time{}, fmt.Errorf("failed to get pod list while fetching metrics: %v", err)
}
if len(podList.Items) == 0 {
return nil, time.Time{}, fmt.Errorf("no pods matched the provided selector")
}
podNames := make([]string, len(podList.Items))
for i, pod := range podList.Items {
podNames[i] = pod.Name
}
now := time.Now()
startTime := now.Add(heapsterQueryStart)
metricPath := fmt.Sprintf("/api/v1/model/namespaces/%s/pod-list/%s/metrics/%s",
namespace,
strings.Join(podNames, ","),
metricName)
resultRaw, err := h.services.
ProxyGet(h.heapsterScheme, h.heapsterService, h.heapsterPort, metricPath, map[string]string{"start": startTime.Format(time.RFC3339)}).
DoRaw()
if err != nil {
return nil, time.Time{}, fmt.Errorf("failed to get heapster service: %v", err)
}
var metrics heapster.MetricResultList
err = json.Unmarshal(resultRaw, &metrics)
if err != nil {
return nil, time.Time{}, fmt.Errorf("failed to unmarshal heapster response: %v", err)
}
glog.V(4).Infof("Heapster metrics result: %s", string(resultRaw))
if len(metrics.Items) != len(podNames) {
// if we get too many metrics or two few metrics, we have no way of knowing which metric goes to which pod
// (note that Heapster returns *empty* metric items when a pod does not exist or have that metric, so this
// does not cover the "missing metric entry" case)
return nil, time.Time{}, fmt.Errorf("requested metrics for %v pods, got metrics for %v", len(podNames), len(metrics.Items))
}
var timestamp *time.Time
res := make(PodMetricsInfo, len(metrics.Items))
for i, podMetrics := range metrics.Items {
val, podTimestamp, hadMetrics := collapseTimeSamples(podMetrics, time.Minute)
if hadMetrics {
res[podNames[i]] = val
if timestamp == nil || podTimestamp.Before(*timestamp) {
timestamp = &podTimestamp
}
}
}
if timestamp == nil {
timestamp = &time.Time{}
}
return res, *timestamp, nil
}
// GetResourceReplicas calculates the desired replica count based on a target resource utilization percentage
// of the given resource for pods matching the given selector in the given namespace, and the current replica count
func (c *ReplicaCalculator) GetResourceReplicas(currentReplicas int32, targetUtilization int32, resource v1.ResourceName, namespace string, selector labels.Selector) (replicaCount int32, utilization int32, timestamp time.Time, err error) {
metrics, timestamp, err := c.metricsClient.GetResourceMetric(resource, namespace, selector)
if err != nil {
return 0, 0, time.Time{}, fmt.Errorf("unable to get metrics for resource %s: %v", resource, err)
}
podList, err := c.podsGetter.Pods(namespace).List(v1.ListOptions{LabelSelector: selector.String()})
if err != nil {
return 0, 0, time.Time{}, fmt.Errorf("unable to get pods while calculating replica count: %v", err)
}
if len(podList.Items) == 0 {
return 0, 0, time.Time{}, fmt.Errorf("no pods returned by selector while calculating replica count")
}
requests := make(map[string]int64, len(podList.Items))
readyPodCount := 0
unreadyPods := sets.NewString()
missingPods := sets.NewString()
for _, pod := range podList.Items {
podSum := int64(0)
for _, container := range pod.Spec.Containers {
if containerRequest, ok := container.Resources.Requests[resource]; ok {
podSum += containerRequest.MilliValue()
} else {
return 0, 0, time.Time{}, fmt.Errorf("missing request for %s on container %s in pod %s/%s", resource, container.Name, namespace, pod.Name)
}
}
requests[pod.Name] = podSum
if pod.Status.Phase != v1.PodRunning || !v1.IsPodReady(&pod) {
// save this pod name for later, but pretend it doesn't exist for now
unreadyPods.Insert(pod.Name)
delete(metrics, pod.Name)
continue
}
if _, found := metrics[pod.Name]; !found {
// save this pod name for later, but pretend it doesn't exist for now
missingPods.Insert(pod.Name)
continue
}
readyPodCount++
}
if len(metrics) == 0 {
return 0, 0, time.Time{}, fmt.Errorf("did not receive metrics for any ready pods")
}
usageRatio, utilization, err := metricsclient.GetResourceUtilizationRatio(metrics, requests, targetUtilization)
if err != nil {
return 0, 0, time.Time{}, err
}
rebalanceUnready := len(unreadyPods) > 0 && usageRatio > 1.0
if !rebalanceUnready && len(missingPods) == 0 {
if math.Abs(1.0-usageRatio) <= tolerance {
// return the current replicas if the change would be too small
return currentReplicas, utilization, timestamp, nil
}
// if we don't have any unready or missing pods, we can calculate the new replica count now
return int32(math.Ceil(usageRatio * float64(readyPodCount))), utilization, timestamp, nil
}
if len(missingPods) > 0 {
if usageRatio < 1.0 {
// on a scale-down, treat missing pods as using 100% of the resource request
for podName := range missingPods {
metrics[podName] = requests[podName]
}
} else if usageRatio > 1.0 {
// on a scale-up, treat missing pods as using 0% of the resource request
for podName := range missingPods {
metrics[podName] = 0
}
}
}
if rebalanceUnready {
// on a scale-up, treat unready pods as using 0% of the resource request
for podName := range unreadyPods {
metrics[podName] = 0
}
}
// re-run the utilization calculation with our new numbers
newUsageRatio, _, err := metricsclient.GetResourceUtilizationRatio(metrics, requests, targetUtilization)
if err != nil {
return 0, utilization, time.Time{}, err
}
if math.Abs(1.0-newUsageRatio) <= tolerance || (usageRatio < 1.0 && newUsageRatio > 1.0) || (usageRatio > 1.0 && newUsageRatio < 1.0) {
// return the current replicas if the change would be too small,
// or if the new usage ratio would cause a change in scale direction
return currentReplicas, utilization, timestamp, nil
}
// return the result, where the number of replicas considered is
// however many replicas factored into our calculation
return int32(math.Ceil(newUsageRatio * float64(len(metrics)))), utilization, timestamp, nil
}
