// Kill all containers in a pod. Returns the number of containers deleted and an error if one occurs.
func (kl *Kubelet) killContainersInPod(pod *api.BoundPod, dockerContainers dockertools.DockerContainers) (int, error) {
podFullName := GetPodFullName(pod)
count := 0
errs := make(chan error, len(pod.Spec.Containers))
wg := sync.WaitGroup{}
for _, container := range pod.Spec.Containers {
// TODO: Consider being more aggressive: kill all containers with this pod UID, period.
if dockerContainer, found, _ := dockerContainers.FindPodContainer(podFullName, pod.UID, container.Name); found {
count++
wg.Add(1)
go func() {
err := kl.killContainer(dockerContainer)
if err != nil {
glog.Errorf("Failed to delete container: %v; Skipping pod %q", err, podFullName)
errs <- err
}
wg.Done()
}()
}
}
wg.Wait()
close(errs)
if len(errs) > 0 {
errList := []error{}
for err := range errs {
errList = append(errList, err)
}
return -1, fmt.Errorf("failed to delete containers (%v)", errList)
}
return count, nil
}
// Delete all containers in a pod (except the network container) returns the number of containers deleted
// and an error if one occurs.
func (kl *Kubelet) deleteAllContainers(pod *Pod, podFullName string, dockerContainers dockertools.DockerContainers) (int, error) {
count := 0
errs := make(chan error, len(pod.Manifest.Containers))
wg := sync.WaitGroup{}
for _, container := range pod.Manifest.Containers {
if dockerContainer, found, _ := dockerContainers.FindPodContainer(podFullName, pod.Manifest.UUID, container.Name); found {
count++
wg.Add(1)
go func() {
err := kl.killContainer(dockerContainer)
if err != nil {
glog.Errorf("Failed to delete container. (%v) Skipping pod %s", err, podFullName)
errs <- err
}
wg.Done()
}()
}
}
wg.Wait()
close(errs)
if len(errs) > 0 {
errList := []error{}
for err := range errs {
errList = append(errList, err)
}
return -1, fmt.Errorf("failed to delete containers (%v)", errList)
}
return count, nil
}
// isPodRunning returns true if all containers of a manifest are running.
func (kl *Kubelet) isPodRunning(pod api.BoundPod, dockerContainers dockertools.DockerContainers) (bool, error) {
for _, container := range pod.Spec.Containers {
dockerContainer, found, _ := dockerContainers.FindPodContainer(GetPodFullName(&pod), pod.UID, container.Name)
if !found {
glog.Infof("container %q not found", container.Name)
return false, nil
}
inspectResult, err := kl.dockerClient.InspectContainer(dockerContainer.ID)
if err != nil {
glog.Infof("failed to inspect container %q: %v", container.Name, err)
return false, err
}
if !inspectResult.State.Running {
glog.Infof("container %q not running: %#v", container.Name, inspectResult.State)
return false, nil
}
}
return true, nil
}
func (kl *Kubelet) syncPod(pod *api.BoundPod, dockerContainers dockertools.DockerContainers) error {
podFullName := GetPodFullName(pod)
uuid := pod.UID
containersToKeep := make(map[dockertools.DockerID]empty)
killedContainers := make(map[dockertools.DockerID]empty)
glog.V(4).Infof("Syncing Pod, podFullName: %q, uuid: %q", podFullName, uuid)
// Make sure we have a network container
var netID dockertools.DockerID
if netDockerContainer, found, _ := dockerContainers.FindPodContainer(podFullName, uuid, networkContainerName); found {
netID = dockertools.DockerID(netDockerContainer.ID)
} else {
glog.V(2).Infof("Network container doesn't exist for pod %q, killing and re-creating the pod", podFullName)
count, err := kl.killContainersInPod(pod, dockerContainers)
if err != nil {
return err
}
netID, err = kl.createNetworkContainer(pod)
if err != nil {
glog.Errorf("Failed to introspect network container: %v; Skipping pod %q", err, podFullName)
return err
}
if count > 0 {
// Re-list everything, otherwise we'll think we're ok.
dockerContainers, err = dockertools.GetKubeletDockerContainers(kl.dockerClient, false)
if err != nil {
glog.Errorf("Error listing containers %#v", dockerContainers)
return err
}
}
}
containersToKeep[netID] = empty{}
podVolumes, err := kl.mountExternalVolumes(pod)
if err != nil {
glog.Errorf("Unable to mount volumes for pod %q: %v; skipping pod", podFullName, err)
return err
}
podStatus := api.PodStatus{}
info, err := kl.GetPodInfo(podFullName, uuid)
if err != nil {
glog.Errorf("Unable to get pod with name %q and uuid %q info, health checks may be invalid", podFullName, uuid)
}
netInfo, found := info[networkContainerName]
if found {
podStatus.PodIP = netInfo.PodIP
}
for _, container := range pod.Spec.Containers {
expectedHash := dockertools.HashContainer(&container)
if dockerContainer, found, hash := dockerContainers.FindPodContainer(podFullName, uuid, container.Name); found {
containerID := dockertools.DockerID(dockerContainer.ID)
glog.V(3).Infof("pod %q container %q exists as %v", podFullName, container.Name, containerID)
// look for changes in the container.
if hash == 0 || hash == expectedHash {
// TODO: This should probably be separated out into a separate goroutine.
healthy, err := kl.healthy(podFullName, uuid, podStatus, container, dockerContainer)
if err != nil {
glog.V(1).Infof("health check errored: %v", err)
containersToKeep[containerID] = empty{}
continue
}
if healthy == health.Healthy {
containersToKeep[containerID] = empty{}
continue
}
glog.V(1).Infof("pod %q container %q is unhealthy. Container will be killed and re-created.", podFullName, container.Name, healthy)
} else {
glog.V(1).Infof("pod %q container %q hash changed (%d vs %d). Container will be killed and re-created.", podFullName, container.Name, hash, expectedHash)
}
if err := kl.killContainer(dockerContainer); err != nil {
glog.V(1).Infof("Failed to kill container %q: %v", dockerContainer.ID, err)
continue
}
killedContainers[containerID] = empty{}
// Also kill associated network container
if netContainer, found, _ := dockerContainers.FindPodContainer(podFullName, uuid, networkContainerName); found {
if err := kl.killContainer(netContainer); err != nil {
glog.V(1).Infof("Failed to kill network container %q: %v", netContainer.ID, err)
continue
}
}
}
// Check RestartPolicy for container
recentContainers, err := dockertools.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.Spec.RestartPolicy.Always == nil {
if pod.Spec.RestartPolicy.Never != nil {
glog.V(3).Infof("Already ran container with name %s--%s--%s, do nothing",
podFullName, uuid, container.Name)
continue
}
if pod.Spec.RestartPolicy.OnFailure != nil {
// Check the exit code of last run
if recentContainers[0].State.ExitCode == 0 {
glog.V(3).Infof("Already successfully ran container with name %s--%s--%s, do nothing",
podFullName, uuid, container.Name)
continue
}
}
}
glog.V(3).Infof("Container with name %s--%s--%s doesn't exist, creating %#v", podFullName, uuid, container.Name, container)
ref, err := containerRef(pod, &container)
if err != nil {
glog.Errorf("Couldn't make a ref to pod %v, container %v: '%v'", pod.Name, container.Name, err)
}
if !api.IsPullNever(container.ImagePullPolicy) {
present, err := kl.dockerPuller.IsImagePresent(container.Image)
latest := dockertools.RequireLatestImage(container.Image)
if err != nil {
if ref != nil {
record.Eventf(ref, "failed", "failed", "Failed to inspect image %q", container.Image)
}
glog.Errorf("Failed to inspect image %q: %v; skipping pod %q container %q", container.Image, err, podFullName, container.Name)
continue
}
if api.IsPullAlways(container.ImagePullPolicy) ||
(api.IsPullIfNotPresent(container.ImagePullPolicy) && (!present || latest)) {
if err := kl.pullImage(container.Image, ref); err != nil {
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?
glog.Errorf("Error running pod %q container %q: %v", podFullName, container.Name, err)
continue
}
containersToKeep[containerID] = empty{}
}
// Kill any containers in this pod which were not identified above (guards against duplicates).
for id, container := range dockerContainers {
curPodFullName, curUUID, _, _ := dockertools.ParseDockerName(container.Names[0])
if curPodFullName == podFullName && curUUID == uuid {
// Don't kill containers we want to keep or those we already killed.
_, keep := containersToKeep[id]
_, killed := killedContainers[id]
if !keep && !killed {
glog.V(1).Infof("Killing unwanted container in pod %q: %+v", curUUID, container)
err = kl.killContainer(container)
if err != nil {
glog.Errorf("Error killing container: %v", err)
}
}
}
}
return nil
}
func (kl *Kubelet) syncPod(pod *Pod, dockerContainers dockertools.DockerContainers) error {
podFullName := GetPodFullName(pod)
uuid := pod.Manifest.UUID
containersToKeep := make(map[dockertools.DockerID]empty)
killedContainers := make(map[dockertools.DockerID]empty)
// Make sure we have a network container
var netID dockertools.DockerID
if networkDockerContainer, found, _ := dockerContainers.FindPodContainer(podFullName, uuid, networkContainerName); found {
netID = dockertools.DockerID(networkDockerContainer.ID)
} else {
glog.Infof("Network container doesn't exist, creating")
count, err := kl.deleteAllContainers(pod, podFullName, dockerContainers)
if err != nil {
return err
}
dockerNetworkID, err := kl.createNetworkContainer(pod)
if err != nil {
glog.Errorf("Failed to introspect network container. (%v) Skipping pod %s", err, podFullName)
return err
}
netID = dockerNetworkID
if count > 0 {
// relist everything, otherwise we'll think we're ok
dockerContainers, err = dockertools.GetKubeletDockerContainers(kl.dockerClient)
if err != nil {
glog.Errorf("Error listing containers %#v", dockerContainers)
return err
}
}
}
containersToKeep[netID] = empty{}
podVolumes, err := kl.mountExternalVolumes(&pod.Manifest)
if err != nil {
glog.Errorf("Unable to mount volumes for pod %s: (%v) Skipping pod.", podFullName, err)
return err
}
podState := api.PodState{Manifest: api.ContainerManifest{UUID: uuid}}
info, err := kl.GetPodInfo(podFullName, uuid)
if err != nil {
glog.Errorf("Unable to get pod with name %s and uuid %s info, health checks may be invalid.",
podFullName, uuid)
}
netInfo, found := info[networkContainerName]
if found && netInfo.NetworkSettings != nil {
podState.PodIP = netInfo.NetworkSettings.IPAddress
}
for _, container := range pod.Manifest.Containers {
expectedHash := dockertools.HashContainer(&container)
if dockerContainer, found, hash := dockerContainers.FindPodContainer(podFullName, uuid, container.Name); found {
containerID := dockertools.DockerID(dockerContainer.ID)
glog.V(1).Infof("pod %s container %s exists as %v", podFullName, container.Name, containerID)
// look for changes in the container.
if hash == 0 || hash == expectedHash {
// TODO: This should probably be separated out into a separate goroutine.
healthy, err := kl.healthy(podFullName, podState, container, dockerContainer)
if err != nil {
glog.V(1).Infof("health check errored: %v", err)
containersToKeep[containerID] = empty{}
continue
}
if healthy == health.Healthy {
containersToKeep[containerID] = empty{}
continue
}
glog.V(1).Infof("pod %s container %s is unhealthy.", podFullName, container.Name, healthy)
} else {
glog.V(1).Infof("container hash changed %d vs %d.", hash, expectedHash)
}
if err := kl.killContainer(dockerContainer); err != nil {
glog.V(1).Infof("Failed to kill container %s: %v", dockerContainer.ID, err)
continue
}
killedContainers[containerID] = empty{}
}
// Check RestartPolicy for container
recentContainers, err := dockertools.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?
glog.Errorf("Error running pod %s container %s: %v", podFullName, container.Name, err)
continue
}
containersToKeep[containerID] = empty{}
}
// Kill any containers in this pod which were not identified above (guards against duplicates).
for id, container := range dockerContainers {
curPodFullName, curUUID, _, _ := dockertools.ParseDockerName(container.Names[0])
if curPodFullName == podFullName && curUUID == uuid {
// Don't kill containers we want to keep or those we already killed.
_, keep := containersToKeep[id]
_, killed := killedContainers[id]
if !keep && !killed {
err = kl.killContainer(container)
if err != nil {
glog.Errorf("Error killing container: %v", err)
}
}
}
}
return nil
}
