Merge pull request #45602 from dashpole/enable_memcg_for_all_tests

Automatic merge from submit-queue (batch tested with PRs 45569, 45602, 45604, 45478, 45550)

Enable kernel memcg notification for node and cluster GCI/COS testing.

Sets --experimental-kernel-memcg-notification=true when running on the GCI/COS image.  It sets this for master and nodes for cluster e2e tests, and for the node in node e2e tests.

Issue #42676 

cc @dchen1107 @Random-Liu

cluster/gce/config-test.sh

@@ -149,6 +149,12 @@ TEST_CLUSTER_RESYNC_PERIOD="${TEST_CLUSTER_RESYNC_PERIOD:---min-resync-period=3m
 TEST_CLUSTER_API_CONTENT_TYPE="${TEST_CLUSTER_API_CONTENT_TYPE:-}"
 
 KUBELET_TEST_ARGS="${KUBELET_TEST_ARGS:-} --max-pods=110 --serialize-image-pulls=false --outofdisk-transition-frequency=0 ${TEST_CLUSTER_API_CONTENT_TYPE}"
+if [[ "${NODE_OS_DISTRIBUTION}" == "gci" ]]; then
+  NODE_KUBELET_TEST_ARGS=" --experimental-kernel-memcg-notification=true"
+fi
+if [[ "${MASTER_OS_DISTRIBUTION}" == "gci" ]]; then
+  MASTER_KUBELET_TEST_ARGS=" --experimental-kernel-memcg-notification=true"
+fi
 APISERVER_TEST_ARGS="${APISERVER_TEST_ARGS:-} --runtime-config=extensions/v1beta1 ${TEST_CLUSTER_DELETE_COLLECTION_WORKERS} ${TEST_CLUSTER_MAX_REQUESTS_INFLIGHT}"
 CONTROLLER_MANAGER_TEST_ARGS="${CONTROLLER_MANAGER_TEST_ARGS:-} ${TEST_CLUSTER_RESYNC_PERIOD} ${TEST_CLUSTER_API_CONTENT_TYPE}"
 SCHEDULER_TEST_ARGS="${SCHEDULER_TEST_ARGS:-} ${TEST_CLUSTER_API_CONTENT_TYPE}"

cluster/gce/gci/configure-helper.sh

@@ -681,6 +681,7 @@ function start-kubelet {
     flags+=" --port=${KUBELET_PORT}"
   fi
   if [[ "${KUBERNETES_MASTER:-}" == "true" ]]; then
+    flags+="${MASTER_KUBELET_TEST_ARGS:-}"
     flags+=" --enable-debugging-handlers=false"
     flags+=" --hairpin-mode=none"
     if [[ "${REGISTER_MASTER_KUBELET:-false}" == "true" ]]; then
@@ -691,6 +692,7 @@ function start-kubelet {
       flags+=" --pod-cidr=${MASTER_IP_RANGE}"
     fi
   else # For nodes
+    flags+="${NODE_KUBELET_TEST_ARGS:-}"
     flags+=" --enable-debugging-handlers=true"
     flags+=" --api-servers=https://${KUBERNETES_MASTER_NAME}"
     if [[ "${HAIRPIN_MODE:-}" == "promiscuous-bridge" ]] || \

test/e2e_node/jenkins/jenkins-memcg-serial.properties

@@ -13,7 +13,7 @@ CLEANUP=true
 # gci-docker-version specifies docker version in GCI image.
 GCE_INSTANCE_METADATA="user-data<${GCI_CLOUD_INIT},gci-docker-version=${DOCKER_VERSION},gci-update-strategy=update_disabled"
 GINKGO_FLAGS='--focus="MemoryEviction" --skip=""'
-TEST_ARGS='--kubelet-flags=--experimental-kernel-memcg-notification=true --feature-gates=DynamicKubeletConfig=true'
+TEST_ARGS='--kubelet-flags=--feature-gates=DynamicKubeletConfig=true'
 KUBELET_ARGS='--cgroups-per-qos=true --cgroup-root=/'
 PARALLELISM=1
 TIMEOUT=1h

test/e2e_node/remote/node_e2e.go

@@ -108,9 +108,9 @@ func tarAddCOSMounter(tar string) error {
 	return nil
 }
 
-// updateCOSMounterPath updates kubelet flags to set gci mounter path. This will only take effect for
+// updateCOSKubeletFlags updates kubelet flags to set gci mounter path, and enables memcg notifications. This will only take effect for
 // GCI/COS image.
-func updateCOSMounterPath(args, host, workspace string) (string, error) {
+func updateCOSKubeletFlags(args, host, workspace string) (string, error) {
 	// Determine if tests will run on a GCI/COS node.
 	output, err := SSH(host, "cat", "/etc/os-release")
 	if err != nil {
@@ -145,7 +145,7 @@ func updateCOSMounterPath(args, host, workspace string) (string, error) {
 		return args, fmt.Errorf("unabled to chmod 544 GCI/COS mounter script. Err: %v, Output:\n%s", err, output)
 	}
 	// Insert args at beginning of test args, so any values from command line take precedence
-	args = fmt.Sprintf("--kubelet-flags=--experimental-mounter-path=%s ", mounterPath) + args
+	args = fmt.Sprintf("--kubelet-flags='--experimental-kernel-memcg-notification=true --experimental-mounter-path=%s'", mounterPath) + args
 	return args, nil
 }
 
@@ -164,7 +164,7 @@ func (n *NodeE2ERemote) RunTest(host, workspace, results, junitFilePrefix, testA
 	// Kill any running node processes
 	cleanupNodeProcesses(host)
 
-	testArgs, err := updateCOSMounterPath(testArgs, host, workspace)
+	testArgs, err := updateCOSKubeletFlags(testArgs, host, workspace)
 	if err != nil {
 		return "", err
 	}