Implement OomScoreAdj in kubelet

pkg/kubelet/dockertools/manager.go

@@ -62,6 +62,8 @@ const (
 
 	minimumDockerAPIVersion = "1.20"
 
+	dockerv110APIVersion = "1.21"
+
 	// ndots specifies the minimum number of dots that a domain name must contain for the resolver to consider it as FQDN (fully-qualified)
 	// we want to able to consider SRV lookup names like _dns._udp.kube-dns.default.svc to be considered relative.
 	// hence, setting ndots to be 5.
@@ -503,7 +505,8 @@ func (dm *DockerManager) runContainer(
 	ipcMode string,
 	utsMode string,
 	pidMode string,
-	restartCount int) (kubecontainer.ContainerID, error) {
+	restartCount int,
+	oomScoreAdj int) (kubecontainer.ContainerID, error) {
 
 	dockerName := KubeletContainerName{
 		PodFullName:   kubecontainer.GetPodFullName(pod),
@@ -584,6 +587,11 @@ func (dm *DockerManager) runContainer(
 		SecurityOpt: securityOpts,
 	}
 
+	// If current api version is newer than docker 1.10 requested, set OomScoreAdj to HostConfig
+	if dm.checkDockerAPIVersion(dockerv110APIVersion) >= 0 {
+		hc.OomScoreAdj = oomScoreAdj
+	}
+
 	if dm.cpuCFSQuota {
 		// if cpuLimit.Amount is nil, then the appropriate default value is returned to allow full usage of cpu resource.
 		cpuQuota, cpuPeriod := milliCPUToQuota(cpuLimit.MilliValue())
@@ -1473,7 +1481,20 @@ func (dm *DockerManager) runContainerInPod(pod *api.Pod, container *api.Containe
 	if usesHostNetwork(pod) {
 		utsMode = namespaceModeHost
 	}
-	id, err := dm.runContainer(pod, container, opts, ref, netMode, ipcMode, utsMode, pidMode, restartCount)
+
+	// Set OOM score of the container based on the priority of the container.
+	// Processes in lower-priority pods should be killed first if the system runs out of memory.
+	// The main pod infrastructure container is considered high priority, since if it is killed the
+	// whole pod will die.
+	var oomScoreAdj int
+	if container.Name == PodInfraContainerName {
+		oomScoreAdj = qos.PodInfraOOMAdj
+	} else {
+		oomScoreAdj = qos.GetContainerOOMScoreAdjust(container, int64(dm.machineInfo.MemoryCapacity))
+
+	}
+
+	id, err := dm.runContainer(pod, container, opts, ref, netMode, ipcMode, utsMode, pidMode, restartCount, oomScoreAdj)
 	if err != nil {
 		return kubecontainer.ContainerID{}, fmt.Errorf("runContainer: %v", err)
 	}
@@ -1512,9 +1533,12 @@ func (dm *DockerManager) runContainerInPod(pod *api.Pod, container *api.Containe
 		return kubecontainer.ContainerID{}, fmt.Errorf("can't get init PID for container %q", id)
 	}
 
-	if err := dm.applyOOMScoreAdj(container, containerInfo); err != nil {
+	// Check if current docker version is higher than 1.10. Otherwise, we have to apply OOMScoreAdj instead of using docker API.
-		return kubecontainer.ContainerID{}, fmt.Errorf("failed to apply oom-score-adj to container %q- %v", err, containerInfo.Name)
+	err = dm.applyOOMScoreAdjIfNeeded(container, containerInfo)
+	if err != nil {
+		return kubecontainer.ContainerID{}, err
 	}
+
 	// The addNDotsOption call appends the ndots option to the resolv.conf file generated by docker.
 	// This resolv.conf file is shared by all containers of the same pod, and needs to be modified only once per pod.
 	// we modify it when the pause container is created since it is the first container created in the pod since it holds
@@ -1529,6 +1553,38 @@ func (dm *DockerManager) runContainerInPod(pod *api.Pod, container *api.Containe
 	return id, err
 }
 
+func (dm *DockerManager) applyOOMScoreAdjIfNeeded(container *api.Container, containerInfo *docker.Container) error {
+	// Compare current API version with expected api version
+	result := dm.checkDockerAPIVersion(dockerv110APIVersion)
+	// If current api version is older than OOMScoreAdj requested, use the old way.
+	if result < 0 {
+		if err := dm.applyOOMScoreAdj(container, containerInfo); err != nil {
+			return fmt.Errorf("failed to apply oom-score-adj to container %q- %v", err, containerInfo.Name)
+		}
+	}
+
+	return nil
+}
+
+// Check current docker API version against expected version.
+// Return:
+// 1 : newer than expected version
+// -1: older than expected version
+// 0 : same version
+func (dm *DockerManager) checkDockerAPIVersion(expectedVersion string) int {
+	apiVersion, err := dm.APIVersion()
+	if err != nil {
+		glog.Errorf("failed to get current docker version - %v", err)
+	}
+
+	result, err := apiVersion.Compare(expectedVersion)
+	if err != nil {
+		glog.Errorf("failed to compare current docker version %v with OOMScoreAdj supported Docker version %q - %v",
+			apiVersion, expectedVersion, err)
+	}
+	return result
+}
+
 func addNDotsOption(resolvFilePath string) error {
 	if len(resolvFilePath) == 0 {
 		glog.Errorf("ResolvConfPath is empty.")