func logPodStartupStatus(c *client.Client, expectedPods int, ns string, observedLabels map[string]string, period time.Duration, stopCh chan struct{}) {
label := labels.SelectorFromSet(labels.Set(observedLabels))
podStore := testutils.NewPodStore(c, ns, label, fields.Everything())
defer podStore.Stop()
ticker := time.NewTicker(period)
defer ticker.Stop()
for {
select {
case <-ticker.C:
pods := podStore.List()
startupStatus := testutils.ComputeRCStartupStatus(pods, expectedPods)
framework.Logf(startupStatus.String("Density"))
case <-stopCh:
pods := podStore.List()
startupStatus := testutils.ComputeRCStartupStatus(pods, expectedPods)
framework.Logf(startupStatus.String("Density"))
return
}
}
}
// rebootNode takes node name on provider through the following steps using c:
// - ensures the node is ready
// - ensures all pods on the node are running and ready
// - reboots the node (by executing rebootCmd over ssh)
// - ensures the node reaches some non-ready state
// - ensures the node becomes ready again
// - ensures all pods on the node become running and ready again
//
// It returns true through result only if all of the steps pass; at the first
// failed step, it will return false through result and not run the rest.
func rebootNode(c clientset.Interface, provider, name, rebootCmd string) bool {
// Setup
ns := api.NamespaceSystem
ps := testutils.NewPodStore(c, ns, labels.Everything(), fields.OneTermEqualSelector(api.PodHostField, name))
defer ps.Stop()
// Get the node initially.
framework.Logf("Getting %s", name)
node, err := c.Core().Nodes().Get(name, metav1.GetOptions{})
if err != nil {
framework.Logf("Couldn't get node %s", name)
return false
}
// Node sanity check: ensure it is "ready".
if !framework.WaitForNodeToBeReady(c, name, framework.NodeReadyInitialTimeout) {
return false
}
// Get all the pods on the node that don't have liveness probe set.
// Liveness probe may cause restart of a pod during node reboot, and the pod may not be running.
pods := ps.List()
podNames := []string{}
for _, p := range pods {
probe := false
for _, c := range p.Spec.Containers {
if c.LivenessProbe != nil {
probe = true
break
}
}
if !probe {
podNames = append(podNames, p.ObjectMeta.Name)
}
}
framework.Logf("Node %s has %d assigned pods with no liveness probes: %v", name, len(podNames), podNames)
// For each pod, we do a sanity check to ensure it's running / healthy
// or succeeded now, as that's what we'll be checking later.
if !framework.CheckPodsRunningReadyOrSucceeded(c, ns, podNames, framework.PodReadyBeforeTimeout) {
printStatusAndLogsForNotReadyPods(c, ns, podNames, pods)
return false
}
// Reboot the node.
if err = framework.IssueSSHCommand(rebootCmd, provider, node); err != nil {
framework.Logf("Error while issuing ssh command: %v", err)
return false
}
// Wait for some kind of "not ready" status.
if !framework.WaitForNodeToBeNotReady(c, name, rebootNodeNotReadyTimeout) {
return false
}
// Wait for some kind of "ready" status.
if !framework.WaitForNodeToBeReady(c, name, rebootNodeReadyAgainTimeout) {
return false
}
// Ensure all of the pods that we found on this node before the reboot are
// running / healthy, or succeeded.
if !framework.CheckPodsRunningReadyOrSucceeded(c, ns, podNames, rebootPodReadyAgainTimeout) {
newPods := ps.List()
printStatusAndLogsForNotReadyPods(c, ns, podNames, newPods)
return false
}
framework.Logf("Reboot successful on node %s", name)
return true
}
}
results = append(results, p)
}
return results
}
var _ = framework.KubeDescribe("Restart [Disruptive]", func() {
f := framework.NewDefaultFramework("restart")
var ps *testutils.PodStore
BeforeEach(func() {
// This test requires the ability to restart all nodes, so the provider
// check must be identical to that call.
framework.SkipUnlessProviderIs("gce", "gke")
ps = testutils.NewPodStore(f.Client, api.NamespaceSystem, labels.Everything(), fields.Everything())
})
AfterEach(func() {
if ps != nil {
ps.Stop()
}
})
It("should restart all nodes and ensure all nodes and pods recover", func() {
nn := framework.TestContext.CloudConfig.NumNodes
By("ensuring all nodes are ready")
nodeNamesBefore, err := framework.CheckNodesReady(f.Client, framework.NodeReadyInitialTimeout, nn)
Expect(err).NotTo(HaveOccurred())
framework.Logf("Got the following nodes before restart: %v", nodeNamesBefore)