// Blocks outgoing network traffic on 'node'. Then verifies that 'podNameToDisappear',
// that belongs to replication controller 'rcName', really disappeared.
// Finally, it checks that the replication controller recreates the
// pods on another node and that now the number of replicas is equal 'replicas'.
// At the end (even in case of errors), the network traffic is brought back to normal.
// This function executes commands on a node so it will work only for some
// environments.
func performTemporaryNetworkFailure(c *client.Client, ns, rcName string, replicas int32, podNameToDisappear string, node *api.Node) {
host := getNodeExternalIP(node)
master := getMaster(c)
By(fmt.Sprintf("block network traffic from node %s to the master", node.Name))
defer func() {
// This code will execute even if setting the iptables rule failed.
// It is on purpose because we may have an error even if the new rule
// had been inserted. (yes, we could look at the error code and ssh error
// separately, but I prefer to stay on the safe side).
By(fmt.Sprintf("Unblock network traffic from node %s to the master", node.Name))
framework.UnblockNetwork(host, master)
}()
framework.Logf("Waiting %v to ensure node %s is ready before beginning test...", resizeNodeReadyTimeout, node.Name)
if !framework.WaitForNodeToBe(c, node.Name, api.NodeReady, true, resizeNodeReadyTimeout) {
framework.Failf("Node %s did not become ready within %v", node.Name, resizeNodeReadyTimeout)
}
framework.BlockNetwork(host, master)
framework.Logf("Waiting %v for node %s to be not ready after simulated network failure", resizeNodeNotReadyTimeout, node.Name)
if !framework.WaitForNodeToBe(c, node.Name, api.NodeReady, false, resizeNodeNotReadyTimeout) {
framework.Failf("Node %s did not become not-ready within %v", node.Name, resizeNodeNotReadyTimeout)
}
framework.Logf("Waiting for pod %s to be removed", podNameToDisappear)
err := framework.WaitForRCPodToDisappear(c, ns, rcName, podNameToDisappear)
Expect(err).NotTo(HaveOccurred())
By("verifying whether the pod from the unreachable node is recreated")
err = framework.VerifyPods(c, ns, rcName, true, replicas)
Expect(err).NotTo(HaveOccurred())
// network traffic is unblocked in a deferred function
}
pods, err := c.Core().Pods(ns).List(options) // list pods after all have been scheduled
Expect(err).NotTo(HaveOccurred())
nodeName := pods.Items[0].Spec.NodeName
node, err := c.Core().Nodes().Get(nodeName)
Expect(err).NotTo(HaveOccurred())
// This creates a temporary network partition, verifies that 'podNameToDisappear',
// that belongs to replication controller 'rcName', really disappeared (because its
// grace period is set to 0).
// Finally, it checks that the replication controller recreates the
// pods on another node and that now the number of replicas is equal 'replicas'.
By(fmt.Sprintf("blocking network traffic from node %s", node.Name))
testUnderTemporaryNetworkFailure(c, ns, node, func() {
framework.Logf("Waiting for pod %s to be removed", pods.Items[0].Name)
err := framework.WaitForRCPodToDisappear(c, ns, name, pods.Items[0].Name)
Expect(err).NotTo(HaveOccurred())
By("verifying whether the pod from the unreachable node is recreated")
err = framework.VerifyPods(c, ns, name, true, replicas)
Expect(err).NotTo(HaveOccurred())
})
framework.Logf("Waiting %v for node %s to be ready once temporary network failure ends", resizeNodeReadyTimeout, node.Name)
if !framework.WaitForNodeToBeReady(c, node.Name, resizeNodeReadyTimeout) {
framework.Failf("Node %s did not become ready within %v", node.Name, resizeNodeReadyTimeout)
}
// sleep a bit, to allow Watch in NodeController to catch up.
time.Sleep(5 * time.Second)
selectorKey, selectorValue = sentinelRC, "true"
label = labels.SelectorFromSet(labels.Set(map[string]string{selectorKey: selectorValue}))
err = framework.WaitForPodsWithLabelRunning(c, ns, label)
Expect(err).NotTo(HaveOccurred())
forEachPod(selectorKey, selectorValue, func(pod api.Pod) {
if pod.Name != bootstrapPodName {
_, err := framework.LookForStringInLog(ns, pod.Name, "sentinel", expectedOnSentinel, serverStartTimeout)
Expect(err).NotTo(HaveOccurred())
}
})
}
checkAllLogs()
By("turning down bootstrap")
framework.RunKubectlOrDie("delete", "-f", bootstrapYaml, nsFlag)
err = framework.WaitForRCPodToDisappear(c, ns, redisRC, bootstrapPodName)
Expect(err).NotTo(HaveOccurred())
By("waiting for the new master election")
checkAllLogs()
})
})
framework.KubeDescribe("Spark", func() {
It("should start spark master, driver and workers", func() {
mkpath := func(file string) string {
return filepath.Join(framework.TestContext.RepoRoot, "examples/spark", file)
}
// TODO: Add Zepplin and Web UI to this example.
serviceYaml := mkpath("spark-master-service.yaml")
masterYaml := mkpath("spark-master-controller.yaml")