func runLivenessTest(f *framework.Framework, pod *v1.Pod, expectNumRestarts int, timeout time.Duration) {
podClient := f.PodClient()
ns := f.Namespace.Name
Expect(pod.Spec.Containers).NotTo(BeEmpty())
containerName := pod.Spec.Containers[0].Name
// At the end of the test, clean up by removing the pod.
defer func() {
By("deleting the pod")
podClient.Delete(pod.Name, v1.NewDeleteOptions(0))
}()
By(fmt.Sprintf("Creating pod %s in namespace %s", pod.Name, ns))
podClient.Create(pod)
// 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.)
framework.ExpectNoError(framework.WaitForPodNotPending(f.ClientSet, ns, pod.Name, pod.ResourceVersion),
fmt.Sprintf("starting pod %s in namespace %s", pod.Name, ns))
framework.Logf("Started pod %s in namespace %s", pod.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 := podClient.Get(pod.Name, metav1.GetOptions{})
framework.ExpectNoError(err, fmt.Sprintf("getting pod %s in namespace %s", pod.Name, ns))
initialRestartCount := v1.GetExistingContainerStatus(pod.Status.ContainerStatuses, containerName).RestartCount
framework.Logf("Initial restart count of pod %s is %d", pod.Name, initialRestartCount)
// Wait for the restart state to be as desired.
deadline := time.Now().Add(timeout)
lastRestartCount := initialRestartCount
observedRestarts := int32(0)
for start := time.Now(); time.Now().Before(deadline); time.Sleep(2 * time.Second) {
pod, err = podClient.Get(pod.Name, metav1.GetOptions{})
framework.ExpectNoError(err, fmt.Sprintf("getting pod %s", pod.Name))
restartCount := v1.GetExistingContainerStatus(pod.Status.ContainerStatuses, containerName).RestartCount
if restartCount != lastRestartCount {
framework.Logf("Restart count of pod %s/%s is now %d (%v elapsed)",
ns, pod.Name, restartCount, time.Since(start))
if restartCount < lastRestartCount {
framework.Failf("Restart count should increment monotonically: restart cont of pod %s/%s changed from %d to %d",
ns, pod.Name, lastRestartCount, restartCount)
}
}
observedRestarts = restartCount - initialRestartCount
if expectNumRestarts > 0 && int(observedRestarts) >= expectNumRestarts {
// Stop if we have observed more than expectNumRestarts restarts.
break
}
lastRestartCount = restartCount
}
// If we expected 0 restarts, fail if observed any restart.
// If we expected n restarts (n > 0), fail if we observed < n restarts.
if (expectNumRestarts == 0 && observedRestarts > 0) || (expectNumRestarts > 0 &&
int(observedRestarts) < expectNumRestarts) {
framework.Failf("pod %s/%s - expected number of restarts: %d, found restarts: %d",
ns, pod.Name, expectNumRestarts, observedRestarts)
}
}
httpYaml := mkpath("http-liveness.yaml")
nsFlag := fmt.Sprintf("--namespace=%v", ns)
framework.RunKubectlOrDie("create", "-f", filepath.Join(framework.TestContext.OutputDir, execYaml), nsFlag)
framework.RunKubectlOrDie("create", "-f", filepath.Join(framework.TestContext.OutputDir, httpYaml), nsFlag)
// Since both containers start rapidly, we can easily run this test in parallel.
var wg sync.WaitGroup
passed := true
checkRestart := func(podName string, timeout time.Duration) {
err := framework.WaitForPodNameRunningInNamespace(c, podName, ns)
Expect(err).NotTo(HaveOccurred())
for t := time.Now(); time.Since(t) < timeout; time.Sleep(framework.Poll) {
pod, err := c.Core().Pods(ns).Get(podName)
framework.ExpectNoError(err, fmt.Sprintf("getting pod %s", podName))
stat := v1.GetExistingContainerStatus(pod.Status.ContainerStatuses, podName)
framework.Logf("Pod: %s, restart count:%d", stat.Name, stat.RestartCount)
if stat.RestartCount > 0 {
framework.Logf("Saw %v restart, succeeded...", podName)
wg.Done()
return
}
}
framework.Logf("Failed waiting for %v restart! ", podName)
passed = false
wg.Done()
}
By("Check restarts")
// Start the "actual test", and wait for both pods to complete.