Merge pull request #1147 from dchen1107/restart

Introduce the simplest RestartPolicy and handling.

pkg/kubelet/config/config_test.go

@@ -52,7 +52,8 @@ func CreateValidPod(name, namespace string) kubelet.Pod {
 		Name:      name,
 		Namespace: namespace,
 		Manifest: api.ContainerManifest{
-			Version: "v1beta1",
+			Version:       "v1beta1",
+			RestartPolicy: api.RestartPolicy{Always: &api.RestartPolicyAlways{}},
 		},
 	}
 }

pkg/kubelet/config/http_test.go

@@ -105,8 +105,12 @@ func TestExtractFromHTTP(t *testing.T) {
 			manifests: api.ContainerManifest{Version: "v1beta1", ID: "foo"},
 			expected: CreatePodUpdate(kubelet.SET,
 				kubelet.Pod{
-					Name:     "foo",
-					Manifest: api.ContainerManifest{Version: "v1beta1", ID: "foo"},
+					Name: "foo",
+					Manifest: api.ContainerManifest{
+						Version:       "v1beta1",
+						ID:            "foo",
+						RestartPolicy: api.RestartPolicy{Always: &api.RestartPolicyAlways{}},
+					},
 				}),
 		},
 		{

pkg/kubelet/docker.go

@@ -153,6 +153,33 @@ func getKubeletDockerContainers(client DockerInterface) (DockerContainers, error
 	return result, nil
 }
 
+// getRecentDockerContainersWithName returns a list of dead docker containers which matches the name
+// and uuid given.
+func getRecentDockerContainersWithNameAndUUID(client DockerInterface, podFullName, uuid, containerName string) ([]*docker.Container, error) {
+	var result []*docker.Container
+	containers, err := client.ListContainers(docker.ListContainersOptions{All: true})
+	if err != nil {
+		return nil, err
+	}
+	for _, dockerContainer := range containers {
+		dockerPodName, dockerUUID, dockerContainerName, _ := parseDockerName(dockerContainer.Names[0])
+		if dockerPodName != podFullName {
+			continue
+		}
+		if uuid != "" && dockerUUID != uuid {
+			continue
+		}
+		if dockerContainerName != containerName {
+			continue
+		}
+		inspectResult, _ := client.InspectContainer(dockerContainer.ID)
+		if inspectResult != nil && !inspectResult.State.Running && !inspectResult.State.Paused {
+			result = append(result, inspectResult)
+		}
+	}
+	return result, nil
+}
+
 // ErrNoContainersInPod is returned when there are no running containers for a given pod
 var ErrNoContainersInPod = errors.New("no containers exist for this pod")
 

pkg/kubelet/kubelet.go

@@ -516,11 +516,35 @@ func (kl *Kubelet) syncPod(pod *Pod, dockerContainers DockerContainers) error {
 			killedContainers[containerID] = empty{}
 		}
 
-		glog.Infof("Container doesn't exist, creating %#v", container)
+		// Check RestartPolicy for container
+		recentContainers, err := getRecentDockerContainersWithNameAndUUID(kl.dockerClient, podFullName, uuid, container.Name)
+		if err != nil {
+			glog.Errorf("Error listing recent containers with name and uuid:%s--%s--%s", podFullName, uuid, container.Name)
+			// TODO(dawnchen): error handling here?
+		}
+
+		if len(recentContainers) > 0 && pod.Manifest.RestartPolicy.Always == nil {
+			if pod.Manifest.RestartPolicy.Never != nil {
+				glog.Infof("Already ran container with name %s--%s--%s, do nothing",
+					podFullName, uuid, container.Name)
+				continue
+			}
+			if pod.Manifest.RestartPolicy.OnFailure != nil {
+				// Check the exit code of last run
+				if recentContainers[0].State.ExitCode == 0 {
+					glog.Infof("Already successfully ran container with name %s--%s--%s, do nothing",
+						podFullName, uuid, container.Name)
+					continue
+				}
+			}
+		}
+
+		glog.Infof("Container with name %s--%s--%s doesn't exist, creating %#v", podFullName, uuid, container.Name, container)
 		if err := kl.dockerPuller.Pull(container.Image); err != nil {
 			glog.Errorf("Failed to pull image %s: %v skipping pod %s container %s.", container.Image, err, podFullName, container.Name)
 			continue
 		}
+		// TODO(dawnchen): Check RestartPolicy.DelaySeconds before restart a container
 		containerID, err := kl.runContainer(pod, &container, podVolumes, "container:"+string(netID))
 		if err != nil {
 			// TODO(bburns) : Perhaps blacklist a container after N failures?

pkg/kubelet/kubelet_test.go

@@ -300,7 +300,7 @@ func TestSyncPodsCreatesNetAndContainer(t *testing.T) {
 	kubelet.drainWorkers()
 
 	verifyCalls(t, fakeDocker, []string{
-		"list", "list", "create", "start", "list", "inspect", "create", "start"})
+		"list", "list", "create", "start", "list", "inspect", "list", "create", "start"})
 
 	fakeDocker.lock.Lock()
 	if len(fakeDocker.Created) != 2 ||
@@ -338,7 +338,7 @@ func TestSyncPodsWithNetCreatesContainer(t *testing.T) {
 	kubelet.drainWorkers()
 
 	verifyCalls(t, fakeDocker, []string{
-		"list", "list", "list", "inspect", "create", "start"})
+		"list", "list", "list", "inspect", "list", "create", "start"})
 
 	fakeDocker.lock.Lock()
 	if len(fakeDocker.Created) != 1 ||
@@ -388,7 +388,7 @@ func TestSyncPodsWithNetCreatesContainerCallsHandler(t *testing.T) {
 	kubelet.drainWorkers()
 
 	verifyCalls(t, fakeDocker, []string{
-		"list", "list", "list", "inspect", "create", "start"})
+		"list", "list", "list", "inspect", "list", "create", "start"})
 
 	fakeDocker.lock.Lock()
 	if len(fakeDocker.Created) != 1 ||
@@ -428,7 +428,7 @@ func TestSyncPodsDeletesWithNoNetContainer(t *testing.T) {
 	kubelet.drainWorkers()
 
 	verifyCalls(t, fakeDocker, []string{
-		"list", "list", "stop", "create", "start", "list", "list", "inspect", "create", "start"})
+		"list", "list", "stop", "create", "start", "list", "list", "inspect", "list", "create", "start"})
 
 	// A map iteration is used to delete containers, so must not depend on
 	// order here.
@@ -561,7 +561,7 @@ func TestSyncPodBadHash(t *testing.T) {
 		t.Errorf("unexpected error: %v", err)
 	}
 
-	verifyCalls(t, fakeDocker, []string{"list", "stop", "create", "start"})
+	verifyCalls(t, fakeDocker, []string{"list", "stop", "list", "create", "start"})
 
 	// A map interation is used to delete containers, so must not depend on
 	// order here.
@@ -608,7 +608,7 @@ func TestSyncPodUnhealthy(t *testing.T) {
 		t.Errorf("unexpected error: %v", err)
 	}
 
-	verifyCalls(t, fakeDocker, []string{"list", "stop", "create", "start"})
+	verifyCalls(t, fakeDocker, []string{"list", "stop", "list", "create", "start"})
 
 	// A map interation is used to delete containers, so must not depend on
 	// order here.
@@ -1329,7 +1329,7 @@ func TestSyncPodEventHandlerFails(t *testing.T) {
 		t.Errorf("unexpected error: %v", err)
 	}
 
-	verifyCalls(t, fakeDocker, []string{"list", "create", "start", "stop"})
+	verifyCalls(t, fakeDocker, []string{"list", "list", "create", "start", "stop"})
 
 	if len(fakeDocker.stopped) != 1 {
 		t.Errorf("Wrong containers were stopped: %v", fakeDocker.stopped)