// testHostIP tests that a pod gets a host IP
func testHostIP(c *client.Client, ns string, pod *api.Pod) {
podClient := c.Pods(ns)
By("creating pod")
defer podClient.Delete(pod.Name, nil)
_, err := podClient.Create(pod)
if err != nil {
Failf("Failed to create pod: %v", err)
}
By("ensuring that pod is running and has a hostIP")
// Wait for the pods to enter the running state. Waiting loops until the pods
// are running so non-running pods cause a timeout for this test.
err = waitForPodRunningInNamespace(c, pod.Name, ns)
Expect(err).NotTo(HaveOccurred())
// Try to make sure we get a hostIP for each pod.
hostIPTimeout := 2 * time.Minute
t := time.Now()
for {
p, err := podClient.Get(pod.Name)
Expect(err).NotTo(HaveOccurred())
if p.Status.HostIP != "" {
Logf("Pod %s has hostIP: %s", p.Name, p.Status.HostIP)
break
}
if time.Since(t) >= hostIPTimeout {
Failf("Gave up waiting for hostIP of pod %s after %v seconds",
p.Name, time.Since(t).Seconds())
}
Logf("Retrying to get the hostIP of pod %s", p.Name)
time.Sleep(5 * time.Second)
}
}
func verifyResult(c *client.Client, podName string, ns string, oldNotRunning int) {
allPods, err := c.Pods(api.NamespaceAll).List(labels.Everything(), fields.Everything())
expectNoError(err)
_, notRunningPods := getPodsNumbers(allPods)
schedEvents, err := c.Events(ns).List(
labels.Everything(),
fields.Set{
"involvedObject.kind": "Pod",
"involvedObject.name": podName,
"involvedObject.namespace": ns,
"source": "scheduler",
"reason": "FailedScheduling",
}.AsSelector())
expectNoError(err)
printed := false
printOnce := func(msg string) string {
if !printed {
printed = true
return msg
} else {
return ""
}
}
Expect(notRunningPods).To(Equal(1+oldNotRunning), printOnce(fmt.Sprintf("Pods found in the cluster: %#v", allPods)))
Expect(schedEvents.Items).ToNot(BeEmpty(), printOnce(fmt.Sprintf("Pods found in the cluster: %#v", allPods)))
}
func verifyExpectedRcsExistAndGetExpectedPods(c *client.Client) ([]string, error) {
expectedPods := []string{}
// Iterate over the labels that identify the replication controllers that we
// want to check. The rcLabels contains the value values for the k8s-app key
// that identify the replication controllers that we want to check. Using a label
// rather than an explicit name is preferred because the names will typically have
// a version suffix e.g. heapster-monitoring-v1 and this will change after a rolling
// update e.g. to heapster-monitoring-v2. By using a label query we can check for the
// situaiton when a heapster-monitoring-v1 and heapster-monitoring-v2 replication controller
// is running (which would be an error except during a rolling update).
for _, rcLabel := range rcLabels {
rcList, err := c.ReplicationControllers(api.NamespaceSystem).List(labels.Set{"k8s-app": rcLabel}.AsSelector())
if err != nil {
return nil, err
}
if len(rcList.Items) != 1 {
return nil, fmt.Errorf("expected to find one replica for RC with label %s but got %d",
rcLabel, len(rcList.Items))
}
for _, rc := range rcList.Items {
podList, err := c.Pods(api.NamespaceSystem).List(labels.Set(rc.Spec.Selector).AsSelector(), fields.Everything())
if err != nil {
return nil, err
}
for _, pod := range podList.Items {
expectedPods = append(expectedPods, string(pod.UID))
}
}
}
return expectedPods, nil
}
// StartPods check for numPods in TestNS. If they exist, it no-ops, otherwise it starts up
// a temp rc, scales it to match numPods, then deletes the rc leaving behind the pods.
func StartPods(numPods int, host string, restClient *client.Client) error {
start := time.Now()
defer func() {
glog.Infof("StartPods took %v with numPods %d", time.Since(start), numPods)
}()
hostField := fields.OneTermEqualSelector(client.PodHost, host)
pods, err := restClient.Pods(TestNS).List(labels.Everything(), hostField)
if err != nil || len(pods.Items) == numPods {
return err
}
glog.Infof("Found %d pods that match host %v, require %d", len(pods.Items), hostField, numPods)
// For the sake of simplicity, assume all pods in TestNS have selectors matching TestRCManifest.
controller := RCFromManifest(TestRCManifest)
// Make the rc unique to the given host.
controller.Spec.Replicas = numPods
controller.Spec.Template.Spec.NodeName = host
controller.Name = controller.Name + host
controller.Spec.Selector["host"] = host
controller.Spec.Template.Labels["host"] = host
if rc, err := StartRC(controller, restClient); err != nil {
return err
} else {
// Delete the rc, otherwise when we restart master components for the next benchmark
// the rc controller will race with the pods controller in the rc manager.
return restClient.ReplicationControllers(TestNS).Delete(rc.Name)
}
}
func newPodOnNode(c *client.Client, namespace, podName, nodeName string) error {
pod, err := c.Pods(namespace).Create(podOnNode(podName, nodeName, serveHostnameImage))
if err == nil {
Logf("Created pod %s on node %s", pod.ObjectMeta.Name, nodeName)
} else {
Logf("Failed to create pod %s on node %s: %v", podName, nodeName, err)
}
return err
}
func runSchedulerNoPhantomPodsTest(client *client.Client) {
pod := &api.Pod{
Spec: api.PodSpec{
Containers: []api.Container{
{
Name: "c1",
Image: "kubernetes/pause",
Ports: []api.ContainerPort{
{ContainerPort: 1234, HostPort: 9999},
},
ImagePullPolicy: api.PullIfNotPresent,
},
},
},
}
// Assuming we only have two kublets, the third pod here won't schedule
// if the scheduler doesn't correctly handle the delete for the second
// pod.
pod.ObjectMeta.Name = "phantom.foo"
foo, err := client.Pods(api.NamespaceDefault).Create(pod)
if err != nil {
glog.Fatalf("Failed to create pod: %v, %v", pod, err)
}
if err := wait.Poll(time.Second, time.Second*30, podRunning(client, foo.Namespace, foo.Name)); err != nil {
glog.Fatalf("FAILED: pod never started running %v", err)
}
pod.ObjectMeta.Name = "phantom.bar"
bar, err := client.Pods(api.NamespaceDefault).Create(pod)
if err != nil {
glog.Fatalf("Failed to create pod: %v, %v", pod, err)
}
if err := wait.Poll(time.Second, time.Second*30, podRunning(client, bar.Namespace, bar.Name)); err != nil {
glog.Fatalf("FAILED: pod never started running %v", err)
}
// Delete a pod to free up room.
glog.Infof("Deleting pod %v", bar.Name)
err = client.Pods(api.NamespaceDefault).Delete(bar.Name, nil)
if err != nil {
glog.Fatalf("FAILED: couldn't delete pod %q: %v", bar.Name, err)
}
pod.ObjectMeta.Name = "phantom.baz"
baz, err := client.Pods(api.NamespaceDefault).Create(pod)
if err != nil {
glog.Fatalf("Failed to create pod: %v, %v", pod, err)
}
if err := wait.Poll(time.Second, time.Second*60, podRunning(client, baz.Namespace, baz.Name)); err != nil {
glog.Fatalf("FAILED: (Scheduler probably didn't process deletion of 'phantom.bar') Pod never started running: %v", err)
}
glog.Info("Scheduler doesn't make phantom pods: test passed.")
}
// 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 *client.Client, ns string, labelSelector labels.Selector) (int, []string) {
pods, err := c.Pods(ns).List(labelSelector, fields.Everything())
expectNoError(err)
failedContainers := 0
containerRestartNodes := util.NewStringSet()
for _, p := range pods.Items {
for _, v := range FailedContainers(&p) {
failedContainers = failedContainers + v.restarts
containerRestartNodes.Insert(p.Spec.NodeName)
}
}
return failedContainers, containerRestartNodes.List()
}
func runBuildRunningPodDeleteTest(t *testing.T, clusterAdminClient *client.Client, clusterAdminKubeClient *kclient.Client) {
buildWatch, err := clusterAdminClient.Builds(testutil.Namespace()).Watch(labels.Everything(), fields.Everything(), "0")
if err != nil {
t.Fatalf("Couldn't subscribe to Builds %v", err)
}
defer buildWatch.Stop()
created, err := clusterAdminClient.Builds(testutil.Namespace()).Create(mockBuild())
if err != nil {
t.Fatalf("Couldn't create Build: %v", err)
}
podWatch, err := clusterAdminKubeClient.Pods(testutil.Namespace()).Watch(labels.Everything(), fields.Everything(), created.ResourceVersion)
if err != nil {
t.Fatalf("Couldn't subscribe to Pods %v", err)
}
defer podWatch.Stop()
// wait for initial build event from the creation of the imagerepo with tag latest
event := waitForWatch(t, "initial build added", buildWatch)
if e, a := watchapi.Added, event.Type; e != a {
t.Fatalf("expected watch event type %s, got %s", e, a)
}
newBuild := event.Object.(*buildapi.Build)
// initial pod creation for build
event = waitForWatch(t, "build pod created", podWatch)
if e, a := watchapi.Added, event.Type; e != a {
t.Fatalf("expected watch event type %s, got %s", e, a)
}
event = waitForWatch(t, "build updated to pending", buildWatch)
if e, a := watchapi.Modified, event.Type; e != a {
t.Fatalf("expected watch event type %s, got %s", e, a)
}
newBuild = event.Object.(*buildapi.Build)
if newBuild.Status.Phase != buildapi.BuildPhasePending {
t.Fatalf("expected build status to be marked pending, but was marked %s", newBuild.Status.Phase)
}
clusterAdminKubeClient.Pods(testutil.Namespace()).Delete(buildutil.GetBuildPodName(newBuild), kapi.NewDeleteOptions(0))
event = waitForWatch(t, "build updated to error", buildWatch)
if e, a := watchapi.Modified, event.Type; e != a {
t.Fatalf("expected watch event type %s, got %s", e, a)
}
newBuild = event.Object.(*buildapi.Build)
if newBuild.Status.Phase != buildapi.BuildPhaseError {
t.Fatalf("expected build status to be marked error, but was marked %s", newBuild.Status.Phase)
}
}
func translatePodNameToIpOrFail(c *client.Client, ns string, expectedEndpoints map[string][]int) map[string][]int {
portsByIp := make(map[string][]int)
for name, portList := range expectedEndpoints {
pod, err := c.Pods(ns).Get(name)
if err != nil {
Failf("failed to get pod %s, that's pretty weird. validation failed: %s", name, err)
}
portsByIp[pod.Status.PodIP] = portList
By(fmt.Sprintf(""))
}
By(fmt.Sprintf("successfully translated pod names to ips: %v -> %v on namespace %s", expectedEndpoints, portsByIp, ns))
return portsByIp
}
func doServiceAccountAPIRequests(t *testing.T, c *client.Client, ns string, authenticated bool, canRead bool, canWrite bool) {
testSecret := &api.Secret{
ObjectMeta: api.ObjectMeta{Name: "testSecret"},
Data: map[string][]byte{"test": []byte("data")},
}
readOps := []testOperation{
func() error { _, err := c.Secrets(ns).List(labels.Everything(), fields.Everything()); return err },
func() error { _, err := c.Pods(ns).List(labels.Everything(), fields.Everything()); return err },
}
writeOps := []testOperation{
func() error { _, err := c.Secrets(ns).Create(testSecret); return err },
func() error { return c.Secrets(ns).Delete(testSecret.Name) },
}
for _, op := range readOps {
err := op()
unauthorizedError := errors.IsUnauthorized(err)
forbiddenError := errors.IsForbidden(err)
switch {
case !authenticated && !unauthorizedError:
t.Fatalf("expected unauthorized error, got %v", err)
case authenticated && unauthorizedError:
t.Fatalf("unexpected unauthorized error: %v", err)
case authenticated && canRead && forbiddenError:
t.Fatalf("unexpected forbidden error: %v", err)
case authenticated && !canRead && !forbiddenError:
t.Fatalf("expected forbidden error, got: %v", err)
}
}
for _, op := range writeOps {
err := op()
unauthorizedError := errors.IsUnauthorized(err)
forbiddenError := errors.IsForbidden(err)
switch {
case !authenticated && !unauthorizedError:
t.Fatalf("expected unauthorized error, got %v", err)
case authenticated && unauthorizedError:
t.Fatalf("unexpected unauthorized error: %v", err)
case authenticated && canWrite && forbiddenError:
t.Fatalf("unexpected forbidden error: %v", err)
case authenticated && !canWrite && !forbiddenError:
t.Fatalf("expected forbidden error, got: %v", err)
}
}
}
func handleAttachReplicationController(c *client.Client, controller *api.ReplicationController, opts *AttachOptions) error {
var pods *api.PodList
for pods == nil || len(pods.Items) == 0 {
var err error
if pods, err = c.Pods(controller.Namespace).List(labels.SelectorFromSet(controller.Spec.Selector), fields.Everything()); err != nil {
return err
}
if len(pods.Items) == 0 {
fmt.Fprint(opts.Out, "Waiting for pod to be scheduled\n")
time.Sleep(2 * time.Second)
}
}
pod := &pods.Items[0]
return handleAttachPod(c, pod, opts)
}
func podScheduled(c *client.Client, podNamespace, podName string) wait.ConditionFunc {
return func() (bool, error) {
pod, err := c.Pods(podNamespace).Get(podName)
if errors.IsNotFound(err) {
return false, nil
}
if err != nil {
// This could be a connection error so we want to retry.
return false, nil
}
if pod.Spec.NodeName == "" {
return false, nil
}
return true, nil
}
}
func podsCreated(c *client.Client, ns, name string, replicas int) (*api.PodList, error) {
timeout := time.Minute
// List the pods, making sure we observe all the replicas.
label := labels.SelectorFromSet(labels.Set(map[string]string{"name": name}))
for start := time.Now(); time.Since(start) < timeout; time.Sleep(5 * time.Second) {
pods, err := c.Pods(ns).List(label, fields.Everything())
if err != nil {
return nil, err
}
Logf("Pod name %s: Found %d pods out of %d", name, len(pods.Items), replicas)
if len(pods.Items) == replicas {
return pods, nil
}
}
return nil, fmt.Errorf("Pod name %s: Gave up waiting %v for %d pods to come up", name, timeout, replicas)
}
func podRunning(c *client.Client, podNamespace string, podName string) wait.ConditionFunc {
return func() (bool, error) {
pod, err := c.Pods(podNamespace).Get(podName)
if apierrors.IsNotFound(err) {
return false, nil
}
if err != nil {
// This could be a connection error so we want to retry, but log the error.
glog.Errorf("Error when reading pod %q: %v", podName, err)
return false, nil
}
if pod.Status.Phase != api.PodRunning {
return false, nil
}
return true, nil
}
}
func runBuildDeleteTest(t *testing.T, clusterAdminClient *client.Client, clusterAdminKubeClient *kclient.Client) {
buildWatch, err := clusterAdminClient.Builds(testutil.Namespace()).Watch(labels.Everything(), fields.Everything(), "0")
if err != nil {
t.Fatalf("Couldn't subscribe to Builds %v", err)
}
defer buildWatch.Stop()
created, err := clusterAdminClient.Builds(testutil.Namespace()).Create(mockBuild())
if err != nil {
t.Fatalf("Couldn't create Build: %v", err)
}
podWatch, err := clusterAdminKubeClient.Pods(testutil.Namespace()).Watch(labels.Everything(), fields.Everything(), created.ResourceVersion)
if err != nil {
t.Fatalf("Couldn't subscribe to Pods %v", err)
}
defer podWatch.Stop()
// wait for initial build event from the creation of the imagerepo with tag latest
event := waitForWatch(t, "initial build added", buildWatch)
if e, a := watchapi.Added, event.Type; e != a {
t.Fatalf("expected watch event type %s, got %s", e, a)
}
newBuild := event.Object.(*buildapi.Build)
// initial pod creation for build
event = waitForWatch(t, "build pod created", podWatch)
if e, a := watchapi.Added, event.Type; e != a {
t.Fatalf("expected watch event type %s, got %s", e, a)
}
clusterAdminClient.Builds(testutil.Namespace()).Delete(newBuild.Name)
event = waitForWatch(t, "pod deleted due to build deleted", podWatch)
if e, a := watchapi.Deleted, event.Type; e != a {
t.Fatalf("expected watch event type %s, got %s", e, a)
}
pod := event.Object.(*kapi.Pod)
if expected := buildutil.GetBuildPodName(newBuild); pod.Name != expected {
t.Fatalf("Expected pod %s to be deleted, but pod %s was deleted", expected, pod.Name)
}
}
func forEachPod(c *client.Client, ns, selectorKey, selectorValue string, fn func(api.Pod)) {
var pods *api.PodList
var err error
for t := time.Now(); time.Since(t) < podListTimeout; time.Sleep(poll) {
pods, err = c.Pods(ns).List(labels.SelectorFromSet(labels.Set(map[string]string{selectorKey: selectorValue})), fields.Everything())
Expect(err).NotTo(HaveOccurred())
if len(pods.Items) > 0 {
break
}
}
if pods == nil || len(pods.Items) == 0 {
Failf("No pods found")
}
for _, pod := range pods.Items {
err = waitForPodRunningInNamespace(c, pod.Name, ns)
Expect(err).NotTo(HaveOccurred())
fn(pod)
}
}
func addEndpointPodOrFail(c *client.Client, ns, name string, labels map[string]string, containerPorts []api.ContainerPort) {
By(fmt.Sprintf("Adding pod %v in namespace %v", name, ns))
pod := &api.Pod{
ObjectMeta: api.ObjectMeta{
Name: name,
Labels: labels,
},
Spec: api.PodSpec{
Containers: []api.Container{
{
Name: "test",
Image: "gcr.io/google_containers/pause",
Ports: containerPorts,
},
},
},
}
_, err := c.Pods(ns).Create(pod)
Expect(err).NotTo(HaveOccurred())
}
func waitForPodCondition(c *client.Client, ns, podName, desc string, timeout time.Duration, condition podCondition) error {
e2e.Logf("Waiting up to %[1]v for pod %-[2]*[3]s status to be %[4]s", timeout, podPrintWidth, podName, desc)
for start := time.Now(); time.Since(start) < timeout; time.Sleep(poll) {
pod, err := c.Pods(ns).Get(podName)
if err != nil {
// Aligning this text makes it much more readable
e2e.Logf("Get pod %-[1]*[2]s in namespace '%[3]s' failed, ignoring for %[4]v. Error: %[5]v",
podPrintWidth, podName, ns, poll, err)
continue
}
done, err := condition(pod)
if done {
return err
}
e2e.Logf("Waiting for pod %-[1]*[2]s in namespace '%[3]s' status to be '%[4]s'"+
"(found phase: %[5]q, readiness: %[6]t) (%[7]v elapsed)",
podPrintWidth, podName, ns, desc, pod.Status.Phase, podReady(pod), time.Since(start))
}
return fmt.Errorf("gave up waiting for pod '%s' to be '%s' after %v", podName, desc, timeout)
}
// Clean both server and client pods.
func volumeTestCleanup(client *client.Client, config VolumeTestConfig) {
By(fmt.Sprint("cleaning the environment after ", config.prefix))
defer GinkgoRecover()
podClient := client.Pods(config.namespace)
// ignore all errors, the pods may not be even created
podClient.Delete(config.prefix+"-client", nil)
podClient.Delete(config.prefix+"-server", nil)
}
func verifyResult(c *client.Client, podName string, ns string) {
allPods, err := c.Pods(api.NamespaceAll).List(labels.Everything(), fields.Everything())
expectNoError(err)
runningPods := 0
notRunningPods := make([]api.Pod, 0)
for _, pod := range allPods.Items {
if pod.Status.Phase == api.PodRunning {
runningPods += 1
} else {
notRunningPods = append(notRunningPods, pod)
}
}
schedEvents, err := c.Events(ns).List(
labels.Everything(),
fields.Set{
"involvedObject.kind": "Pod",
"involvedObject.name": podName,
"involvedObject.namespace": ns,
"source": "scheduler",
"reason": "failedScheduling",
}.AsSelector())
expectNoError(err)
printed := false
printOnce := func(msg string) string {
if !printed {
printed = true
return msg
} else {
return ""
}
}
Expect(len(notRunningPods)).To(Equal(1), printOnce(fmt.Sprintf("Pods found in the cluster: %#v", allPods)))
Expect(schedEvents.Items).ToNot(BeEmpty(), printOnce(fmt.Sprintf("Pods found in the cluster: %#v", allPods)))
Expect(notRunningPods[0].Name).To(Equal(podName), printOnce(fmt.Sprintf("Pods found in the cluster: %#v", allPods)))
}
// podsOnMinions returns true when all of the selected pods exist on a minion.
func podsOnMinions(c *client.Client, podNamespace string, labelSelector labels.Selector) wait.ConditionFunc {
// Wait until all pods are running on the node.
return func() (bool, error) {
pods, err := c.Pods(podNamespace).List(labelSelector, fields.Everything())
if err != nil {
glog.Infof("Unable to get pods to list: %v", err)
return false, nil
}
for i := range pods.Items {
pod := pods.Items[i]
podString := fmt.Sprintf("%q/%q", pod.Namespace, pod.Name)
glog.Infof("Check whether pod %q exists on node %q", podString, pod.Spec.NodeName)
if len(pod.Spec.NodeName) == 0 {
glog.Infof("Pod %q is not bound to a host yet", podString)
return false, nil
}
if pod.Status.Phase != api.PodRunning {
return false, nil
}
}
return true, nil
}
}
func waitForPodRunning(c *client.Client, pod *api.Pod, out io.Writer) error {
for {
pod, err := c.Pods(pod.Namespace).Get(pod.Name)
if err != nil {
return err
}
if pod.Status.Phase == api.PodRunning {
ready := true
for _, status := range pod.Status.ContainerStatuses {
if !status.Ready {
ready = false
break
}
}
if ready {
return nil
}
}
fmt.Fprintf(out, "Waiting for pod %s/%s to be running\n", pod.Namespace, pod.Name)
time.Sleep(2 * time.Second)
continue
}
}
func runLivenessTest(c *client.Client, ns string, podDescr *api.Pod, expectRestart bool) {
By(fmt.Sprintf("Creating pod %s in namespace %s", podDescr.Name, ns))
_, err := c.Pods(ns).Create(podDescr)
expectNoError(err, fmt.Sprintf("creating pod %s", podDescr.Name))
// At the end of the test, clean up by removing the pod.
defer func() {
By("deleting the pod")
c.Pods(ns).Delete(podDescr.Name, nil)
}()
// Wait until the pod is not pending. (Here we need to check for something other than
// 'Pending' other than checking for 'Running', since when failures occur, we go to
// 'Terminated' which can cause indefinite blocking.)
expectNoError(waitForPodNotPending(c, ns, podDescr.Name),
fmt.Sprintf("starting pod %s in namespace %s", podDescr.Name, ns))
By(fmt.Sprintf("Started pod %s in namespace %s", podDescr.Name, ns))
// Check the pod's current state and verify that restartCount is present.
By("checking the pod's current state and verifying that restartCount is present")
pod, err := c.Pods(ns).Get(podDescr.Name)
expectNoError(err, fmt.Sprintf("getting pod %s in namespace %s", podDescr.Name, ns))
initialRestartCount := api.GetExistingContainerStatus(pod.Status.ContainerStatuses, "liveness").RestartCount
By(fmt.Sprintf("Initial restart count of pod %s is %d", podDescr.Name, initialRestartCount))
// Wait for the restart state to be as desired.
restarts, deadline := false, time.Now().Add(2*time.Minute)
for start := time.Now(); time.Now().Before(deadline); time.Sleep(2 * time.Second) {
pod, err = c.Pods(ns).Get(podDescr.Name)
expectNoError(err, fmt.Sprintf("getting pod %s", podDescr.Name))
restartCount := api.GetExistingContainerStatus(pod.Status.ContainerStatuses, "liveness").RestartCount
By(fmt.Sprintf("Restart count of pod %s/%s is now %d (%v elapsed)",
ns, podDescr.Name, restartCount, time.Since(start)))
if restartCount > initialRestartCount {
By(fmt.Sprintf("Restart count of pod %s/%s changed from %d to %d",
ns, podDescr.Name, initialRestartCount, restartCount))
restarts = true
break
}
}
if restarts != expectRestart {
Failf("pod %s/%s - expected restarts: %t, found restarts: %t",
ns, podDescr.Name, expectRestart, restarts)
}
}
func createRunningPod(wg *sync.WaitGroup, c *client.Client, name, ns, image string, labels map[string]string) {
defer GinkgoRecover()
defer wg.Done()
pod := &api.Pod{
TypeMeta: api.TypeMeta{
Kind: "Pod",
},
ObjectMeta: api.ObjectMeta{
Name: name,
Labels: labels,
},
Spec: api.PodSpec{
Containers: []api.Container{
{
Name: name,
Image: image,
},
},
},
}
_, err := c.Pods(ns).Create(pod)
expectNoError(err)
expectNoError(waitForPodRunningInNamespace(c, name, ns))
}
// Simplified version of RunRC, that does not create RC, but creates plain Pods and
// requires passing whole Pod definition, which is needed to test various Scheduler predicates.
func startPods(c *client.Client, replicas int, ns string, podNamePrefix string, pod api.Pod) {
pods, err := c.Pods(api.NamespaceAll).List(labels.Everything(), fields.Everything())
expectNoError(err)
podsRunningBefore := len(pods.Items)
for i := 0; i < replicas; i++ {
podName := fmt.Sprintf("%v-%v", podNamePrefix, i)
pod.ObjectMeta.Name = podName
pod.ObjectMeta.Labels["name"] = podName
pod.Spec.Containers[0].Name = podName
_, err = c.Pods(ns).Create(&pod)
expectNoError(err)
}
// Wait for pods to start running. Note: this is a functional
// test, not a performance test, so the timeout needs to be
// sufficiently long that it's only triggered if things are
// completely broken vs. running slowly.
timeout := 10 * time.Minute
startTime := time.Now()
currentlyRunningPods := 0
for podsRunningBefore+replicas != currentlyRunningPods {
allPods, err := c.Pods(api.NamespaceAll).List(labels.Everything(), fields.Everything())
expectNoError(err)
runningPods := 0
for _, pod := range allPods.Items {
if pod.Status.Phase == api.PodRunning {
runningPods += 1
}
}
currentlyRunningPods = runningPods
Logf("%v pods running", currentlyRunningPods)
if startTime.Add(timeout).Before(time.Now()) {
Logf("Timed out after %v waiting for pods to start running.", timeout)
break
}
time.Sleep(5 * time.Second)
}
Expect(currentlyRunningPods).To(Equal(podsRunningBefore + replicas))
}
It("liveness pods should be automatically restarted", func() {
mkpath := func(file string) string {
return filepath.Join(testContext.RepoRoot, "docs", "user-guide", "liveness", file)
}
execYaml := mkpath("exec-liveness.yaml")
httpYaml := mkpath("http-liveness.yaml")
nsFlag := fmt.Sprintf("--namespace=%v", ns)
runKubectl("create", "-f", execYaml, nsFlag)
runKubectl("create", "-f", httpYaml, nsFlag)
checkRestart := func(podName string, timeout time.Duration) {
err := waitForPodRunningInNamespace(c, podName, ns)
Expect(err).NotTo(HaveOccurred())
for t := time.Now(); time.Since(t) < timeout; time.Sleep(poll) {
pod, err := c.Pods(ns).Get(podName)
expectNoError(err, fmt.Sprintf("getting pod %s", podName))
restartCount := api.GetExistingContainerStatus(pod.Status.ContainerStatuses, "liveness").RestartCount
Logf("Pod: %s restart count:%d", podName, restartCount)
if restartCount > 0 {
return
}
}
Failf("Pod %s was not restarted", podName)
}
By("Check restarts")
checkRestart("liveness-exec", time.Minute)
checkRestart("liveness-http", time.Minute)
})
})
Selector: labels,
Ports: []api.ServicePort{{
Port: 80,
TargetPort: util.NewIntOrStringFromInt(80),
}},
},
}
_, err := c.Services(ns).Create(service)
Expect(err).NotTo(HaveOccurred())
validateEndpointsOrFail(c, ns, serviceName, map[string][]int{})
var names []string
defer func() {
for _, name := range names {
err := c.Pods(ns).Delete(name, nil)
Expect(err).NotTo(HaveOccurred())
}
}()
name1 := "test1"
addEndpointPodOrFail(c, ns, name1, labels, []api.ContainerPort{{ContainerPort: 80}})
names = append(names, name1)
validateEndpointsOrFail(c, ns, serviceName, map[string][]int{name1: {80}})
name2 := "test2"
addEndpointPodOrFail(c, ns, name2, labels, []api.ContainerPort{{ContainerPort: 80}})
names = append(names, name2)
validateEndpointsOrFail(c, ns, serviceName, map[string][]int{name1: {80}, name2: {80}})
})
// This test verifies that max-pods flag works as advertised. It assumes that cluster add-on pods stay stable
// and cannot be run in parallel with any other test that touches Nodes or Pods. It is so because to check
// if max-pods is working we need to fully saturate the cluster and keep it in this state for few seconds.
It("validates MaxPods limit number of pods that are allowed to run.", func() {
totalPodCapacity = 0
for _, node := range nodeList.Items {
podCapacity, found := node.Status.Capacity["pods"]
Expect(found).To(Equal(true))
totalPodCapacity += podCapacity.Value()
Logf("Node: %v", node)
}
pods, err := c.Pods(api.NamespaceAll).List(labels.Everything(), fields.Everything())
expectNoError(err)
currentlyRunningPods := len(pods.Items)
podsNeededForSaturation := int(totalPodCapacity) - currentlyRunningPods
By(fmt.Sprintf("Starting additional %v Pods to fully saturate the cluster max pods and trying to start another one", podsNeededForSaturation))
startPods(c, podsNeededForSaturation, ns, "maxp", api.Pod{
TypeMeta: api.TypeMeta{
Kind: "Pod",
},
ObjectMeta: api.ObjectMeta{
Name: "",
Labels: map[string]string{"name": ""},
},
Spec: api.PodSpec{
AfterEach(func() {
// Remove any remaining pods from this test if the
// replication controller still exists and the replica count
// isn't 0. This means the controller wasn't cleaned up
// during the test so clean it up here
rc, err := c.ReplicationControllers(ns).Get(RCName)
if err == nil && rc.Spec.Replicas != 0 {
By("Cleaning up the replication controller")
err := DeleteRC(c, ns, RCName)
expectNoError(err)
}
By("Removing additional pods if any")
for i := 1; i <= minionCount; i++ {
name := additionalPodsPrefix + "-" + strconv.Itoa(i)
c.Pods(ns).Delete(name, nil)
}
By(fmt.Sprintf("Destroying namespace for this suite %v", ns))
if err := c.Namespaces().Delete(ns); err != nil {
Failf("Couldn't delete ns %s", err)
}
expectNoError(writePerfData(c, fmt.Sprintf(testContext.OutputDir+"/%s", uuid), "after"))
// Verify latency metrics
// TODO: We should reset metrics before the test. Currently previous tests influence latency metrics.
highLatencyRequests, err := HighLatencyRequests(c, 3*time.Second, util.NewStringSet("events"))
expectNoError(err)
Expect(highLatencyRequests).NotTo(BeNumerically(">", 0), "There should be no high-latency requests")
})
// Factor out the cases into two separate tests.
It("[replication controller] recreates pods scheduled on the unreachable minion node "+
"AND allows scheduling of pods on a minion after it rejoins the cluster", func() {
// Create a replication controller for a service that serves its hostname.
// The source for the Docker container kubernetes/serve_hostname is in contrib/for-demos/serve_hostname
name := "my-hostname-net"
newSVCByName(c, ns, name)
replicas := testContext.CloudConfig.NumNodes
newRCByName(c, ns, name, replicas)
err := verifyPods(c, ns, name, true, replicas)
Expect(err).NotTo(HaveOccurred(), "Each pod should start running and responding")
By("choose a node with at least one pod - we will block some network traffic on this node")
label := labels.SelectorFromSet(labels.Set(map[string]string{"name": name}))
pods, err := c.Pods(ns).List(label, fields.Everything()) // list pods after all have been scheduled
Expect(err).NotTo(HaveOccurred())
nodeName := pods.Items[0].Spec.NodeName
node, err := c.Nodes().Get(nodeName)
Expect(err).NotTo(HaveOccurred())
By(fmt.Sprintf("block network traffic from node %s", node.Name))
performTemporaryNetworkFailure(c, ns, name, replicas, pods.Items[0].Name, node)
Logf("Waiting %v for node %s to be ready once temporary network failure ends", resizeNodeReadyTimeout, node.Name)
if !waitForNodeToBe(c, node.Name, true, resizeNodeReadyTimeout) {
Failf("Node %s did not become ready within %v", node.Name, resizeNodeReadyTimeout)
}
By("verify whether new pods can be created on the re-attached node")
// increasing the RC size is not a valid way to test this