Merge remote-tracking branch 'main/master' into local-ssd-provision

deploy/aliyun/templates/local-volume-provisioner.yaml.tpl

@@ -72,10 +72,10 @@ spec:
         resources:
           requests:
             cpu: 100m
-            memory: 100m
+            memory: 100Mi
           limits:
             cpu: 100m
-            memory: 100m
+            memory: 100Mi
         volumeMounts:
         - mountPath: /etc/provisioner/config
           name: provisioner-config

deploy/aws/manifests/local-volume-provisioner.yaml

@@ -63,10 +63,10 @@ spec:
           resources:
             requests:
               cpu: 100m
-              memory: 100m
+              memory: 100Mi
             limits:
               cpu: 100m
-              memory: 100m
+              memory: 100Mi
           volumeMounts:
             - mountPath: /etc/provisioner/config
               name: provisioner-config

manifests/local-dind/local-volume-provisioner.yaml

@@ -56,10 +56,10 @@ spec:
           resources:
             requests:
               cpu: 100m
-              memory: 100m
+              memory: 100Mi
             limits:
               cpu: 100m
-              memory: 100m
+              memory: 100Mi
           volumeMounts:
             - mountPath: /etc/provisioner/config
               name: provisioner-config

tests/actions.go

@@ -104,7 +104,6 @@ const (
 	tidbClusterChartName                      = "tidb-cluster"
 	backupChartName                           = "tidb-backup"
 	statbilityTestTag                         = "stability"
-	metricsPort                               = 8090
 )
 
 type OperatorActions interface {
@@ -157,6 +156,7 @@ type OperatorActions interface {
 	CheckTidbClustersAvailableOrDie(infos []*TidbClusterConfig)
 	CheckOneEtcdDownOrDie(operatorConfig *OperatorConfig, clusters []*TidbClusterConfig, faultNode string)
 	CheckOneApiserverDownOrDie(operatorConfig *OperatorConfig, clusters []*TidbClusterConfig, faultNode string)
+	CheckKubeProxyDownOrDie(clusters []*TidbClusterConfig)
 	RegisterWebHookAndService(context *apimachinery.CertContext, info *OperatorConfig) error
 	RegisterWebHookAndServiceOrDie(context *apimachinery.CertContext, info *OperatorConfig)
 	CleanWebHookAndService(info *OperatorConfig) error
@@ -1540,7 +1540,7 @@ func (oa *operatorActions) checkGrafanaData(clusterInfo *TidbClusterConfig) erro
 	// Grafana ready, init grafana client, no more sync logic because race condition is okay here
 	if clusterInfo.GrafanaClient == nil {
 		grafanaURL := fmt.Sprintf("http://%s.%s:3000", svcName, ns)
-		client, err := metrics.NewClient(grafanaURL, grafanaUsername, grafanaPassword, metricsPort)
+		client, err := metrics.NewClient(grafanaURL, grafanaUsername, grafanaPassword)
 		if err != nil {
 			return err
 		}
@@ -2342,11 +2342,10 @@ func (oa *operatorActions) EventWorker() {
 			ns := clusterEv.ns
 			clusterName := clusterEv.clusterName
 			grafanaURL := fmt.Sprintf("http://%s-grafana.%s:3000", clusterName, ns)
-			client, err := metrics.NewClient(grafanaURL, grafanaUsername, grafanaPassword, metricsPort)
+			client, err := metrics.NewClient(grafanaURL, grafanaUsername, grafanaPassword)
 			if err != nil {
-				retryEvents = append(retryEvents, ev)
-				glog.V(4).Infof("failed to new grafana client: [%s/%s], %v", ns, clusterName, err)
-				continue
+				// If parse grafana URL failed, this error cannot be recovered by retrying, so send error msg and panic
+				slack.NotifyAndPanic(fmt.Errorf("failed to parse grafana URL so can't new grafana client: %s, %v", grafanaURL, err))
 			}
 
 			anno := metrics.Annotation{

tests/cmd/fault-trigger/main.go

@@ -20,8 +20,6 @@ import (
 	_ "net/http/pprof"
 	"time"
 
-	"github.com/pingcap/tidb-operator/tests/pkg/util"
-
 	"github.com/golang/glog"
 	"github.com/pingcap/tidb-operator/tests/pkg/fault-trigger/api"
 	"github.com/pingcap/tidb-operator/tests/pkg/fault-trigger/manager"
@@ -30,17 +28,13 @@ import (
 )
 
 var (
-	port             int
-	pprofPort        int
-	kubeProxyImage   string
-	hostnameOverride string
+	port      int
+	pprofPort int
 )
 
 func init() {
 	flag.IntVar(&port, "port", 23332, "The port that the fault trigger's http service runs on (default 23332)")
 	flag.IntVar(&pprofPort, "pprof-port", 6060, "The port that the pprof's http service runs on (default 6060)")
-	flag.StringVar(&kubeProxyImage, "kube-proxy-image", "k8s.gcr.io/kube-proxy:v1.12.2", "The kube proxy image (default k8s.gcr.io/kube-proxy:v1.12.2)")
-	flag.StringVar(&hostnameOverride, "hostname-override", "", "If non-empty, will use this string as identification instead of the actual hostname")
 
 	flag.Parse()
 }
@@ -49,16 +43,7 @@ func main() {
 	logs.InitLogs()
 	defer logs.FlushLogs()
 
-	mgr := manager.NewManager(kubeProxyImage)
-	hostname, err := util.GetHostname(hostnameOverride)
-	if err != nil {
-		glog.Fatalf("get hostname failed, err: %v", err)
-	}
-	err = mgr.UpdateKubeProxyDaemonset(hostname)
-	if err != nil {
-		glog.Fatalf("update kube-proxy daemonset failed, err: %v", err)
-	}
-
+	mgr := manager.NewManager()
 	server := api.NewServer(mgr, port)
 
 	go wait.Forever(func() {

tests/cmd/stability/main.go

@@ -23,7 +23,7 @@ import (
 
 	"github.com/pingcap/tidb-operator/tests/pkg/apimachinery"
 
-	"k8s.io/api/core/v1"
+	v1 "k8s.io/api/core/v1"
 
 	"github.com/golang/glog"
 	"github.com/jinzhu/copier"
@@ -332,12 +332,16 @@ func run() {
 	// stop one etcd node and k8s/operator/tidbcluster is available
 	faultEtcd := tests.SelectNode(cfg.ETCDs)
 	fta.StopETCDOrDie(faultEtcd)
-	defer fta.StartETCDOrDie(faultEtcd)
 	// TODO make the pause interval as a argument
 	time.Sleep(3 * time.Minute)
 	oa.CheckOneEtcdDownOrDie(operatorCfg, allClusters, faultEtcd)
 	fta.StartETCDOrDie(faultEtcd)
 
+	// stop all kube-proxy and k8s/operator/tidbcluster is available
+	fta.StopKubeProxyOrDie()
+	oa.CheckKubeProxyDownOrDie(allClusters)
+	fta.StartKubeProxyOrDie()
+
 	//clean temp dirs when stability success
 	err := cfg.CleanTempDirs()
 	if err != nil {

tests/failover.go

@@ -452,42 +452,72 @@ func (oa *operatorActions) CheckOneEtcdDownOrDie(operatorConfig *OperatorConfig,
 		if err != nil {
 			return err
 		}
-		glog.V(4).Infof("all clusters is available")
+		glog.V(4).Infof("all clusters are available")
+		return nil
+	})
+}
+
+func (oa *operatorActions) CheckKubeProxyDownOrDie(clusters []*TidbClusterConfig) {
+	glog.Infof("checking k8s/tidbCluster status when kube-proxy down")
+
+	KeepOrDie(3*time.Second, 10*time.Minute, func() error {
+		err := oa.CheckK8sAvailable(nil, nil)
+		if err != nil {
+			return err
+
+		}
+		glog.V(4).Infof("k8s cluster is available.")
+		err = oa.CheckTidbClustersAvailable(clusters)
+		if err != nil {
+			return err
+		}
+		glog.V(4).Infof("all clusters are available.")
 		return nil
 	})
 }
 
 func (oa *operatorActions) CheckOneApiserverDownOrDie(operatorConfig *OperatorConfig, clusters []*TidbClusterConfig, faultNode string) {
 	glog.Infof("check k8s/operator/tidbCluster status when one apiserver down")
 	affectedPods := map[string]*corev1.Pod{}
-	apiserverPod, err := GetApiserverPod(oa.kubeCli, faultNode)
+	apiserverPod, err := GetKubeApiserverPod(oa.kubeCli, faultNode)
 	if err != nil {
-		slack.NotifyAndPanic(fmt.Errorf("can't find apiserver in node:%s", faultNode))
+		slack.NotifyAndPanic(fmt.Errorf("can't find apiserver in k8s cluster"))
 	}
 	if apiserverPod != nil {
 		affectedPods[apiserverPod.GetName()] = apiserverPod
 	}
-	controllerPod, err := GetControllerManagerPod(oa.kubeCli, faultNode)
+
+	controllerPod, err := GetKubeControllerManagerPod(oa.kubeCli, faultNode)
 	if err != nil {
-		glog.Infof("can't find controllerManager in node:%s", faultNode)
+		slack.NotifyAndPanic(fmt.Errorf("can't find kube-controller-manager in k8s cluster"))
 	}
 	if controllerPod != nil {
 		affectedPods[controllerPod.GetName()] = controllerPod
 	}
-	schedulerPod, err := GetSchedulerPod(oa.kubeCli, faultNode)
+
+	schedulerPod, err := GetKubeSchedulerPod(oa.kubeCli, faultNode)
 	if err != nil {
-		glog.Infof("can't find scheduler in node:%s", faultNode)
+		slack.NotifyAndPanic(fmt.Errorf("can't find kube-scheduler in k8s cluster"))
 	}
 	if schedulerPod != nil {
 		affectedPods[schedulerPod.GetName()] = schedulerPod
 	}
-	dnsPod, err := GetDNSPod(oa.kubeCli, faultNode)
+
+	dnsPod, err := GetKubeDNSPod(oa.kubeCli, faultNode)
 	if err != nil {
-		slack.NotifyAndPanic(fmt.Errorf("can't find controller-manager in node:%s", faultNode))
+		slack.NotifyAndPanic(fmt.Errorf("can't find kube-dns in k8s cluster"))
 	}
 	if dnsPod != nil {
 		affectedPods[dnsPod.GetName()] = dnsPod
 	}
+
+	proxyPod, err := GetKubeProxyPod(oa.kubeCli, faultNode)
+	if err != nil {
+		slack.NotifyAndPanic(fmt.Errorf("can't find kube-proxy in k8s cluster"))
+	}
+	if proxyPod != nil {
+		affectedPods[dnsPod.GetName()] = proxyPod
+	}
 	KeepOrDie(3*time.Second, 10*time.Minute, func() error {
 		err := oa.CheckK8sAvailable(map[string]string{faultNode: faultNode}, affectedPods)
 		if err != nil {
@@ -555,7 +585,7 @@ func (oa *operatorActions) CheckK8sAvailable(excludeNodes map[string]string, exc
 			}
 			podState := GetPodStatus(&pod)
 			if podState != string(corev1.PodRunning) {
-				return false, fmt.Errorf("pod:[%s/%s] is unavailable,state is %s", pod.GetName(), pod.GetNamespace(), podState)
+				return false, fmt.Errorf("pod:[%s/%s] is unavailable,state is %s", pod.GetNamespace(), pod.GetName(), podState)
 			}
 		}
 		return true, nil