func (self *podAndContainerCollector) Collect(ch chan<- prometheus.Metric) {
runningContainers, err := self.containerCache.RunningContainers()
if err != nil {
glog.Warning("Failed to get running container information while collecting metrics: %v", err)
return
}
// Get a set of running pods.
runningPods := make(map[types.UID]struct{})
for _, cont := range runningContainers {
_, uid, _, _, err := dockertools.ParseDockerName(cont.Names[0])
if err != nil {
continue
}
runningPods[uid] = struct{}{}
}
ch <- prometheus.MustNewConstMetric(
runningPodCountDesc,
prometheus.GaugeValue,
float64(len(runningPods)))
ch <- prometheus.MustNewConstMetric(
runningContainerCountDesc,
prometheus.GaugeValue,
float64(len(runningContainers)))
}
// SyncPods synchronizes the configured list of pods (desired state) with the host current state.
func (kl *Kubelet) SyncPods(pods []api.BoundPod) error {
glog.V(4).Infof("Desired: %#v", pods)
var err error
desiredContainers := make(map[podContainer]empty)
desiredPods := make(map[string]empty)
dockerContainers, err := dockertools.GetKubeletDockerContainers(kl.dockerClient, false)
if err != nil {
glog.Errorf("Error listing containers: %#v", dockerContainers)
return err
}
// Check for any containers that need starting
for ix := range pods {
pod := &pods[ix]
podFullName := GetPodFullName(pod)
uuid := pod.UID
desiredPods[uuid] = empty{}
// Add all containers (including net) to the map.
desiredContainers[podContainer{podFullName, uuid, networkContainerName}] = empty{}
for _, cont := range pod.Spec.Containers {
desiredContainers[podContainer{podFullName, uuid, cont.Name}] = empty{}
}
// Run the sync in an async manifest worker.
kl.podWorkers.Run(podFullName, func() {
err := kl.syncPod(pod, dockerContainers)
if err != nil {
glog.Errorf("Error syncing pod, skipping: %s", err)
}
})
}
// Kill any containers we don't need.
for _, container := range dockerContainers {
// Don't kill containers that are in the desired pods.
podFullName, uuid, containerName, _ := dockertools.ParseDockerName(container.Names[0])
if _, found := desiredPods[uuid]; found {
// syncPod() will handle this one.
continue
}
pc := podContainer{podFullName, uuid, containerName}
if _, ok := desiredContainers[pc]; !ok {
glog.V(1).Infof("Killing unwanted container %+v", pc)
err = kl.killContainer(container)
if err != nil {
glog.Errorf("Error killing container %+v: %s", pc, err)
}
}
}
// Remove any orphaned volumes.
kl.reconcileVolumes(pods)
return err
}
// SyncPods synchronizes the configured list of pods (desired state) with the host current state.
func (kl *Kubelet) SyncPods(pods []Pod) error {
glog.Infof("Desired [%s]: %+v", kl.hostname, pods)
var err error
desiredContainers := make(map[podContainer]empty)
dockerContainers, err := dockertools.GetKubeletDockerContainers(kl.dockerClient)
if err != nil {
glog.Errorf("Error listing containers %#v", dockerContainers)
return err
}
// Check for any containers that need starting
for i := range pods {
pod := &pods[i]
podFullName := GetPodFullName(pod)
uuid := pod.Manifest.UUID
// Add all containers (including net) to the map.
desiredContainers[podContainer{podFullName, uuid, networkContainerName}] = empty{}
for _, cont := range pod.Manifest.Containers {
desiredContainers[podContainer{podFullName, uuid, cont.Name}] = empty{}
}
// Run the sync in an async manifest worker.
kl.podWorkers.Run(podFullName, func() {
err := kl.syncPod(pod, dockerContainers)
if err != nil {
glog.Errorf("Error syncing pod: %v skipping.", err)
}
})
}
// Kill any containers we don't need
existingContainers, err := dockertools.GetKubeletDockerContainers(kl.dockerClient)
if err != nil {
glog.Errorf("Error listing containers: %v", err)
return err
}
for _, container := range existingContainers {
// Don't kill containers that are in the desired pods.
podFullName, uuid, containerName, _ := dockertools.ParseDockerName(container.Names[0])
if _, ok := desiredContainers[podContainer{podFullName, uuid, containerName}]; !ok {
err = kl.killContainer(container)
if err != nil {
glog.Errorf("Error killing container: %v", err)
}
}
}
// Remove any orphaned volumes.
kl.reconcileVolumes(pods)
return err
}
// Get all containers that are evictable. Evictable containers are: not running
// and created more than MinAge ago.
func (cgc *realContainerGC) evictableContainers() (containersByEvictUnit, []containerGCInfo, error) {
containers, err := dockertools.GetKubeletDockerContainers(cgc.dockerClient, true)
if err != nil {
return containersByEvictUnit{}, []containerGCInfo{}, err
}
unidentifiedContainers := make([]containerGCInfo, 0)
evictUnits := make(containersByEvictUnit)
newestGCTime := time.Now().Add(-cgc.policy.MinAge)
for _, container := range containers {
// Prune out running containers.
data, err := cgc.dockerClient.InspectContainer(container.ID)
if err != nil {
// Container may have been removed already, skip.
continue
} else if data.State.Running {
continue
} else if newestGCTime.Before(data.Created) {
continue
}
containerInfo := containerGCInfo{
id: container.ID,
name: container.Names[0],
createTime: data.Created,
}
containerName, _, err := dockertools.ParseDockerName(container.Names[0])
if err != nil {
unidentifiedContainers = append(unidentifiedContainers, containerInfo)
} else {
key := evictUnit{
uid: containerName.PodUID,
name: containerName.ContainerName,
}
containerInfo.podNameWithNamespace = containerName.PodFullName
containerInfo.containerName = containerName.ContainerName
evictUnits[key] = append(evictUnits[key], containerInfo)
}
}
// Sort the containers by age.
for uid := range evictUnits {
sort.Sort(byCreated(evictUnits[uid]))
}
return evictUnits, unidentifiedContainers, nil
}
func (kl *Kubelet) killContainerByID(ID, name string) error {
glog.Infof("Killing: %s", ID)
err := kl.dockerClient.StopContainer(ID, 10)
if len(name) == 0 {
return err
}
podFullName, uuid, containerName, _ := dockertools.ParseDockerName(name)
kl.LogEvent(&api.Event{
Event: "STOP",
Manifest: &api.ContainerManifest{
//TODO: This should be reported using either the apiserver schema or the kubelet schema
ID: podFullName,
UUID: uuid,
},
Container: &api.Container{
Name: containerName,
},
})
return err
}
// TODO: Also enforce a maximum total number of containers.
func (kl *Kubelet) GarbageCollectContainers() error {
if kl.maxContainerCount == 0 {
return nil
}
containers, err := dockertools.GetKubeletDockerContainers(kl.dockerClient, true)
if err != nil {
return err
}
uuidToIDMap := map[string][]string{}
for _, container := range containers {
_, uuid, name, _ := dockertools.ParseDockerName(container.ID)
uuidName := uuid + "." + name
uuidToIDMap[uuidName] = append(uuidToIDMap[uuidName], container.ID)
}
for _, list := range uuidToIDMap {
if len(list) <= kl.maxContainerCount {
continue
}
if err := kl.purgeOldest(list); err != nil {
return err
}
}
return 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
}
// SyncPods synchronizes the configured list of pods (desired state) with the host current state.
func (kl *Kubelet) SyncPods(pods []api.BoundPod) error {
glog.V(4).Infof("Desired: %#v", pods)
var err error
desiredContainers := make(map[podContainer]empty)
desiredPods := make(map[types.UID]empty)
dockerContainers, err := dockertools.GetKubeletDockerContainers(kl.dockerClient, false)
if err != nil {
glog.Errorf("Error listing containers: %#v", dockerContainers)
return err
}
// Check for any containers that need starting
for ix := range pods {
pod := &pods[ix]
podFullName := GetPodFullName(pod)
uid := pod.UID
desiredPods[uid] = empty{}
// Add all containers (including net) to the map.
desiredContainers[podContainer{podFullName, uid, networkContainerName}] = empty{}
for _, cont := range pod.Spec.Containers {
desiredContainers[podContainer{podFullName, uid, cont.Name}] = empty{}
}
// Run the sync in an async manifest worker.
kl.podWorkers.Run(podFullName, func() {
if err := kl.syncPod(pod, dockerContainers); err != nil {
glog.Errorf("Error syncing pod, skipping: %v", err)
record.Eventf(pod, "failedSync", "Error syncing pod, skipping: %v", err)
}
})
}
// Kill any containers we don't need.
for _, container := range dockerContainers {
// Don't kill containers that are in the desired pods.
podFullName, uid, containerName, _ := dockertools.ParseDockerName(container.Names[0])
if _, found := desiredPods[uid]; found {
// syncPod() will handle this one.
continue
}
_, _, podAnnotations := ParsePodFullName(podFullName)
if source := podAnnotations[ConfigSourceAnnotationKey]; !kl.sourceReady(source) {
// If the source for this container is not ready, skip deletion, so that we don't accidentally
// delete containers for sources that haven't reported yet.
glog.V(4).Infof("Skipping delete of container (%q), source (%s) aren't ready yet.", podFullName, source)
continue
}
pc := podContainer{podFullName, uid, containerName}
if _, ok := desiredContainers[pc]; !ok {
glog.V(1).Infof("Killing unwanted container %+v", pc)
err = kl.killContainer(container)
if err != nil {
glog.Errorf("Error killing container %+v: %v", pc, err)
}
}
}
// Remove any orphaned pods.
err = kl.cleanupOrphanedPods(pods)
if err != nil {
return err
}
// Remove any orphaned volumes.
err = kl.cleanupOrphanedVolumes(pods)
if err != nil {
return err
}
return err
}