// runPod runs a single pod and wait until all containers are running.
func (kl *Qinglet) runPod(pod *api.Pod, retryDelay time.Duration) error {
delay := retryDelay
retry := 0
for {
pods, err := kl.containerRuntime.GetPods(false)
if err != nil {
return fmt.Errorf("failed to get qinglet pods: %v", err)
}
p := container.Pods(pods).FindPodByID(pod.UID)
running, err := kl.isPodRunning(pod, p)
if err != nil {
return fmt.Errorf("failed to check pod status: %v", err)
}
if running {
glog.Infof("pod %q containers running", pod.Name)
return nil
}
glog.Infof("pod %q containers not running: syncing", pod.Name)
// We don't create mirror pods in this mode; pass a dummy boolean value
// to sycnPod.
if err = kl.syncPod(pod, nil, p, SyncPodUpdate); err != nil {
return fmt.Errorf("error syncing pod: %v", err)
}
if retry >= RunOnceMaxRetries {
return fmt.Errorf("timeout error: pod %q containers not running after %d retries", pod.Name, RunOnceMaxRetries)
}
// TODO(proppy): health checking would be better than waiting + checking the state at the next iteration.
glog.Infof("pod %q containers synced, waiting for %v", pod.Name, delay)
time.Sleep(delay)
retry++
delay *= RunOnceRetryDelayBackoff
}
}
func (p *podWorkers) managePodLoop(podUpdates <-chan workUpdate) {
var minRuntimeCacheTime time.Time
for newWork := range podUpdates {
func() {
defer p.checkForUpdates(newWork.pod.UID, newWork.updateCompleteFn)
// We would like to have the state of the containers from at least
// the moment when we finished the previous processing of that pod.
if err := p.runtimeCache.ForceUpdateIfOlder(minRuntimeCacheTime); err != nil {
glog.Errorf("Error updating the container runtime cache: %v", err)
return
}
pods, err := p.runtimeCache.GetPods()
if err != nil {
glog.Errorf("Error getting pods while syncing pod: %v", err)
return
}
err = p.syncPodFn(newWork.pod, newWork.mirrorPod,
qingcontainer.Pods(pods).FindPodByID(newWork.pod.UID), newWork.updateType)
if err != nil {
glog.Errorf("Error syncing pod %s, skipping: %v", newWork.pod.UID, err)
p.recorder.Eventf(newWork.pod, "failedSync", "Error syncing pod, skipping: %v", err)
return
}
minRuntimeCacheTime = time.Now()
newWork.updateCompleteFn()
}()
}
}
func (f *fakePodWorkers) UpdatePod(pod *api.Pod, mirrorPod *api.Pod, updateComplete func()) {
pods, err := f.runtimeCache.GetPods()
if err != nil {
f.t.Errorf("Unexpected error: %v", err)
}
if err := f.syncPodFn(pod, mirrorPod, qingcontainer.Pods(pods).FindPodByID(pod.UID), SyncPodUpdate); err != nil {
f.t.Errorf("Unexpected error: %v", err)
}
}
// GetPodStatus currently invokes GetPods() to return the status.
// TODO(yifan): Split the get status logic from GetPods().
func (r *runtime) GetPodStatus(pod *api.Pod) (*api.PodStatus, error) {
pods, err := r.GetPods(true)
if err != nil {
return nil, err
}
p := qingcontainer.Pods(pods).FindPodByID(pod.UID)
if len(p.Containers) == 0 {
return nil, fmt.Errorf("cannot find status for pod: %q", qingcontainer.BuildPodFullName(pod.Name, pod.Namespace))
}
return &p.Status, nil
}