Merge pull request #20671 from vishh/avoid-moving-all-dockers

pkg/kubelet/cm/container_manager_linux.go

@@ -300,6 +300,19 @@ func (cm *containerManagerImpl) SystemContainersLimit() api.ResourceList {
 	}
 }
 
+func isProcessRunningInHost(pid int) (bool, error) {
+	// Get init mount namespace. Mount namespace is unique for all containers.
+	initMntNs, err := os.Readlink("/proc/1/ns/mnt")
+	if err != nil {
+		return false, fmt.Errorf("failed to find mount namespace of init process")
+	}
+	processMntNs, err := os.Readlink(fmt.Sprintf("/proc/%d/ns/mnt", pid))
+	if err != nil {
+		return false, fmt.Errorf("failed to find mount namespace of process %q", pid)
+	}
+	return initMntNs == processMntNs, nil
+}
+
 // Ensures that the Docker daemon is in the desired container.
 func ensureDockerInContainer(cadvisor cadvisor.Interface, oomScoreAdj int, manager *fs.Manager) error {
 	// What container is Docker in?
@@ -322,6 +335,15 @@ func ensureDockerInContainer(cadvisor cadvisor.Interface, oomScoreAdj int, manag
 	// Move if the pid is not already in the desired container.
 	errs := []error{}
 	for _, pid := range pids {
+		if runningInHost, err := isProcessRunningInHost(pid); err != nil {
+			errs = append(errs, err)
+			// Err on the side of caution. Avoid moving the docker daemon unless we are able to identify its context.
+			continue
+		} else if !runningInHost {
+			// Docker daemon is running inside a container. Don't touch that.
+			continue
+		}
+
 		cont, err := getContainer(pid)
 		if err != nil {
 			errs = append(errs, fmt.Errorf("failed to find container of PID %d: %v", pid, err))
@@ -365,10 +387,6 @@ func getContainer(pid int) (string, error) {
 // The reason of leaving kernel threads at root cgroup is that we don't want to tie the
 // execution of these threads with to-be defined /system quota and create priority inversions.
 //
-// The reason of leaving process 1 at root cgroup is that libcontainer hardcoded on
-// the base cgroup path based on process 1. Please see:
-// https://github.com/kubernetes/kubernetes/issues/12789#issuecomment-132384126
-// for detail explanation.
 func ensureSystemContainer(rootContainer *fs.Manager, manager *fs.Manager) error {
 	// Move non-kernel PIDs to the system container.
 	attemptsRemaining := 10

pkg/kubelet/kubelet.go

@@ -898,42 +898,42 @@ func (kl *Kubelet) StartGarbageCollection() {
 // initializeModules will initialize internal modules that do not require the container runtime to be up.
 // Note that the modules here must not depend on modules that are not initialized here.
 func (kl *Kubelet) initializeModules() error {
-	// Promethues metrics.
+	// Step 1: Move Kubelet to a container, if required.
+	if kl.resourceContainer != "" {
+		// Fixme: I need to reside inside ContainerManager interface.
+		err := util.RunInResourceContainer(kl.resourceContainer)
+		if err != nil {
+			glog.Warningf("Failed to move Kubelet to container %q: %v", kl.resourceContainer, err)
+		}
+		glog.Infof("Running in container %q", kl.resourceContainer)
+	}
+
+	// Step 2: Promethues metrics.
 	metrics.Register(kl.runtimeCache)
 
-	// Step 1: Setup filesystem directories.
+	// Step 3: Setup filesystem directories.
 	if err := kl.setupDataDirs(); err != nil {
 		return err
 	}
 
-	// Step 2: If the container logs directory does not exist, create it.
+	// Step 4: If the container logs directory does not exist, create it.
 	if _, err := os.Stat(containerLogsDir); err != nil {
 		if err := kl.os.Mkdir(containerLogsDir, 0755); err != nil {
 			glog.Errorf("Failed to create directory %q: %v", containerLogsDir, err)
 		}
 	}
 
-	// Step 3: Move Kubelet to a container, if required.
-	if kl.resourceContainer != "" {
-		// Fixme: I need to reside inside ContainerManager interface.
-		err := util.RunInResourceContainer(kl.resourceContainer)
-		if err != nil {
-			glog.Warningf("Failed to move Kubelet to container %q: %v", kl.resourceContainer, err)
-		}
-		glog.Infof("Running in container %q", kl.resourceContainer)
-	}
-
-	// Step 4: Start the image manager.
+	// Step 5: Start the image manager.
 	if err := kl.imageManager.Start(); err != nil {
 		return fmt.Errorf("Failed to start ImageManager, images may not be garbage collected: %v", err)
 	}
 
-	// Step 5: Start container manager.
+	// Step 6: Start container manager.
 	if err := kl.containerManager.Start(kl.nodeConfig); err != nil {
 		return fmt.Errorf("Failed to start ContainerManager %v", err)
 	}
 
-	// Step 6: Start out of memory watcher.
+	// Step 7: Start out of memory watcher.
 	if err := kl.oomWatcher.Start(kl.nodeRef); err != nil {
 		return fmt.Errorf("Failed to start OOM watcher %v", err)
 	}