pingcap · sre-bot · Apr 7, 2020 · Apr 3, 2020 · Apr 3, 2020 · Apr 7, 2020
diff --git a/hack/e2e.sh b/hack/e2e.sh
@@ -73,6 +73,7 @@ Environments:
     AWS_ACCESS_KEY_ID     (eks only) the aws access key id
     AWS_SECRET_ACCESS_KEY (eks only) the aws secret access key
     AWS_REGION            (eks only) the aws region
+    AWS_ZONE              (eks only) the aws zone
     GINKGO_NODES          ginkgo nodes to run specs, defaults: 1
     GINKGO_PARALLEL       if set to `y`, will run specs in parallel, the number of nodes will be the number of cpus
     GINKGO_NO_COLOR       if set to `y`, suppress color output in default reporter
@@ -197,6 +198,7 @@ GCP_MACHINE_TYPE=${GCP_MACHINE_TYPE:-n1-standard-4}
 AWS_ACCESS_KEY_ID=${AWS_ACCESS_KEY_ID:-}
 AWS_SECRET_ACCESS_KEY=${AWS_SECRET_ACCESS_KEY:-}
 AWS_REGION=${AWS_REGION:-}
+AWS_ZONE=${AWS_ZONE:-}
 KUBE_VERSION=${KUBE_VERSION:-v1.12.10}
 KUBE_WORKERS=${KUBE_WORKERS:-3}
 DOCKER_IO_MIRROR=${DOCKER_IO_MIRROR:-}
@@ -223,6 +225,7 @@ echo "GCP_ZONE: $GCP_ZONE"
 # echo "AWS_ACCESS_KEY_ID: $AWS_ACCESS_KEY_ID"
 # echo "AWS_SECRET_ACCESS_KEY: $AWS_SECRET_ACCESS_KEY"
 echo "AWS_REGION: $AWS_REGION"
+echo "AWS_ZONE: $AWS_ZONE"
 echo "KUBE_VERSION: $KUBE_VERSION"
 echo "KUBE_WORKERS: $KUBE_WORKERS"
 echo "DOCKER_IO_MIRROR: $DOCKER_IO_MIRROR"
@@ -465,10 +468,14 @@ EOF
         )
     fi
 elif [ "$PROVIDER" == "eks" ]; then
+    export KUBE_SSH_USER=ec2-user
     hack::ensure_aws_k8s_tester
     if [ -n "$AWS_REGION" ]; then
         aws configure set default.region "$AWS_REGION"
     fi
+    if [ -z "$AWS_ZONE" ]; then
+        AWS_ZONE=${AWS_REGION}a
+    fi
     if [ -n "$AWS_ACCESS_KEY_ID" ]; then
         aws configure set aws_access_key_id "$AWS_ACCESS_KEY_ID"
     fi
@@ -505,6 +512,10 @@ export GCP_PROJECT
 export GCP_REGION
 export GCP_ZONE
 export GCP_CREDENTIALS
+export AWS_ACCESS_KEY_ID
+export AWS_SECRET_ACCESS_KEY
+export AWS_REGION
+export AWS_ZONE
 export IMAGE_TAG
 export SKIP_GINKGO
 export SKIP_IMAGE_LOAD

diff --git a/hack/run-e2e.sh b/hack/run-e2e.sh
@@ -29,6 +29,7 @@ GCP_REGION=${GCP_REGION:-}
 GCP_ZONE=${GCP_ZONE:-}
 GCP_CREDENTIALS=${GCP_CREDENTIALS:-}
 GCP_SDK=${GCP_SDK:-/google-cloud-sdk}
+KUBE_SSH_USER=${KUBE_SSH_USER:-vagrant}
 IMAGE_TAG=${IMAGE_TAG:-}
 SKIP_IMAGE_LOAD=${SKIP_IMAGE_LOAD:-}
 TIDB_OPERATOR_IMAGE=${TIDB_OPERATOR_IMAGE:-localhost:5000/pingcap/tidb-operator:latest}
@@ -51,6 +52,7 @@ if [ -z "$KUBECONFIG" ]; then
     exit 1
 fi
 
+echo "KUBE_SSH_USER: $KUBE_SSH_USER"
 echo "TIDB_OPERATOR_IMAGE: $TIDB_OPERATOR_IMAGE"
 echo "TIDB_BACKUP_MANAGER_IMAGE: $TIDB_BACKUP_MANAGER_IMAGE"
 echo "E2E_IMAGE: $E2E_IMAGE"
@@ -123,67 +125,15 @@ for ((i = 1; i <= 32; i++)) {
 EOF
         done
     elif [ "$PROVIDER" == "gke" ]; then
-        # disks are created under /mnt/stateful_partition directory
-        # https://cloud.google.com/container-optimized-os/docs/concepts/disks-and-filesystem
-        for n in $($KUBECTL_BIN --context "$KUBECONTEXT" get nodes -ojsonpath='{range .items[*]}{.metadata.name}{"\n"}{end}'); do
-            gcloud compute ssh e2e@$n --command 'sudo bash -c '"'"'
-test -d /mnt/stateful_partition/disks || mkdir -p /mnt/stateful_partition/disks
-df -h /mnt/stateful_partition/disks
-test -d /mnt/disks || mkdir -p /mnt/disks
-cd /mnt/disks
-for ((i = 1; i <= 32; i++)) {
-    if [ ! -d vol$i ]; then
-        mkdir vol$i
-    fi
-    if ! mountpoint vol$i &>/dev/null; then
-        if [ ! -d /mnt/stateful_partition/disks/vol$i ]; then
-            mkdir /mnt/stateful_partition/disks/vol$i
-        fi
-        mount --bind /mnt/stateful_partition/disks/vol$i vol$i
-    fi
-}
-'"'"
-        done
+        echo "info: provider is $PROVIDER, skipped"
     elif [ "$PROVIDER" == "eks" ]; then
-        while IFS=$'\n' read -r line; do
-            read -r id dns <<< $line
-            echo "info: prepare disks on $dns"
-            ssh -T -o "StrictHostKeyChecking no" -i ~/.ssh/kube_aws_rsa ec2-user@$dns <<'EOF'
-sudo bash -c '
-test -d /mnt/disks || mkdir -p /mnt/disks
-df -h /mnt/disks
-if mountpoint /mnt/disks &>/dev/null; then
-    echo "info: /mnt/disks is a mountpoint"
-else
-    echo "info: /mnt/disks is not a mountpoint, creating local volumes on the rootfs"
-fi
-cd /mnt/disks
-for ((i = 1; i <= 32; i++)) {
-    if [ ! -d vol$i ]; then
-        mkdir vol$i
-    fi
-    if ! mountpoint vol$i &>/dev/null; then
-        mount --bind vol$i vol$i
-    fi
-}
-echo "info: increase max open files for containers"
-if ! grep -qF "OPTIONS" /etc/sysconfig/docker; then
-    echo 'OPTIONS="--default-ulimit nofile=1024000:1024000"' >> /etc/sysconfig/docker
-fi
-systemctl restart docker
-'
-EOF
-        done <<< "$(e2e::__eks_instances)"
+        echo "info: provider is $PROVIDER, skipped"
     fi
     echo "info: installing local-volume-provisioner"
     $KUBECTL_BIN --context $KUBECONTEXT apply -f ${ROOT}/manifests/local-dind/local-volume-provisioner.yaml
     e2e::__wait_for_ds kube-system local-volume-provisioner
 }
 
-function e2e::__eks_instances() {
-    aws ec2 describe-instances --filter Name=tag:eks:cluster-name,Values=$CLUSTER --query 'Reservations[*].Instances[*].{InstanceId:InstanceId,PublicDnsName:PublicDnsName}' --output text
-}
-
 function e2e::__ecr_url() {
     local account_id=$(aws sts get-caller-identity --output text | awk '{print $1}')
     local region=$(aws configure get region)
@@ -211,6 +161,13 @@ function e2e::setup_helm_server() {
     $HELM_BIN version
 }
 
+# Used by non-kind providers to tag image with its id. This can force our e2e
+# process to pull correct image even if IfNotPresent is used in an existing
+# environment, e.g. testing in the same cluster.
+function e2e::image_id_tag() {
+    docker image inspect -f '{{.Id}}' "$1" | cut -d ':' -f 2 | head -c 10
+}
+
 function e2e::image_load() {
     local images=(
         $TIDB_OPERATOR_IMAGE
@@ -226,9 +183,9 @@ function e2e::image_load() {
     elif [ "$PROVIDER" == "gke" ]; then
         unset DOCKER_CONFIG # We don't need this and it may be read-only and fail the command to fail
         gcloud auth configure-docker
-        GCP_TIDB_OPERATOR_IMAGE=gcr.io/$GCP_PROJECT/tidb-operator:$CLUSTER-$IMAGE_TAG
-        GCP_TIDB_BACKUP_MANAGER_IMAGE=gcr.io/$GCP_PROJECT/tidb-backup-image:$CLUSTER-$IMAGE_TAG
-        GCP_E2E_IMAGE=gcr.io/$GCP_PROJECT/tidb-operator-e2e:$CLUSTER-$IMAGE_TAG
+        GCP_TIDB_OPERATOR_IMAGE=gcr.io/$GCP_PROJECT/tidb-operator:$CLUSTER-$(e2e::image_id_tag $TIDB_OPERATOR_IMAGE)
+        GCP_TIDB_BACKUP_MANAGER_IMAGE=gcr.io/$GCP_PROJECT/tidb-backup-image:$CLUSTER-$(e2e::image_id_tag $TIDB_BACKUP_MANAGER_IMAGE)
+        GCP_E2E_IMAGE=gcr.io/$GCP_PROJECT/tidb-operator-e2e:$CLUSTER-$(e2e::image_id_tag $E2E_IMAGE)
         docker tag $TIDB_OPERATOR_IMAGE $GCP_TIDB_OPERATOR_IMAGE
         docker tag $E2E_IMAGE $GCP_E2E_IMAGE
         docker tag $TIDB_BACKUP_MANAGER_IMAGE $GCP_TIDB_BACKUP_MANAGER_IMAGE
@@ -253,9 +210,9 @@ function e2e::image_load() {
         local ecrURL=$(e2e::__ecr_url)
         echo "info: logging in $ecrURL"
         aws ecr get-login-password | docker login --username AWS --password-stdin $ecrURL
-        AWS_TIDB_OPERATOR_IMAGE=$ecrURL/e2e/tidb-operator:$CLUSTER-$IMAGE_TAG
-        AWS_TIDB_BACKUP_MANAGER_IMAGE=$ecrURL/e2e/tidb-backup-manager:$CLUSTER-$IMAGE_TAG
-        AWS_E2E_IMAGE=$ecrURL/e2e/tidb-operator-e2e:$CLUSTER-$IMAGE_TAG
+        AWS_TIDB_OPERATOR_IMAGE=$ecrURL/e2e/tidb-operator:$CLUSTER-$(e2e::image_id_tag $TIDB_OPERATOR_IMAGE)
+        AWS_TIDB_BACKUP_MANAGER_IMAGE=$ecrURL/e2e/tidb-backup-manager:$CLUSTER-$(e2e::image_id_tag $TIDB_BACKUP_MANAGER_IMAGE)
+        AWS_E2E_IMAGE=$ecrURL/e2e/tidb-operator-e2e:$CLUSTER-$(e2e::image_id_tag $E2E_IMAGE)
         docker tag $TIDB_OPERATOR_IMAGE $AWS_TIDB_OPERATOR_IMAGE
         docker tag $TIDB_BACKUP_MANAGER_IMAGE $AWS_TIDB_BACKUP_MANAGER_IMAGE
         docker tag $E2E_IMAGE $AWS_E2E_IMAGE
@@ -363,16 +320,19 @@ docker_args=(
     -v $KUBECONFIG:/etc/kubernetes/admin.conf:ro
     --env KUBECONFIG=/etc/kubernetes/admin.conf
     --env KUBECONTEXT=$KUBECONTEXT
+    --env KUBE_SSH_USER=$KUBE_SSH_USER
 )
 
 if [ "$PROVIDER" == "eks" ]; then
     e2e_args+=(
         --provider=aws
-        --gce-zone="${AWS_REGION}"
+        --gce-zone="${AWS_ZONE}" # reuse gce-zone to configure aws zone
     )
-    # aws credential is required to get token for EKS
     docker_args+=(
+        # aws credential is required to get token for EKS
         -v $HOME/.aws:/root/.aws
+        # ~/.ssh/kube_aws_rsa must be mounted into e2e container to run ssh
+        -v $HOME/.ssh/kube_aws_rsa:/root/.ssh/kube_aws_rsa
     )
 elif [ "$PROVIDER" == "gke" ]; then
     e2e_args+=(
@@ -393,6 +353,8 @@ elif [ "$PROVIDER" == "gke" ]; then
     fi
     docker_args+=(
         -v ${GCP_SDK}:/google-cloud-sdk
+        # ~/.ssh/google_compute_engine must be mounted into e2e container to run ssh
+        -v $HOME/.ssh/google_compute_engine:/root/.ssh/google_compute_engine
     )
 else
     e2e_args+=(

diff --git a/manifests/gke/local-ssd-provision/local-ssd-provision.yaml b/manifests/gke/local-ssd-provision/local-ssd-provision.yaml
@@ -12,6 +12,7 @@ metadata:
   name: local-provisioner-config
   namespace: kube-system
 data:
+  setPVOwnerRef: "true"
   nodeLabelsForPV: |
     - kubernetes.io/hostname
   storageClassMap: |

diff --git a/manifests/local-dind/local-volume-provisioner.yaml b/manifests/local-dind/local-volume-provisioner.yaml
@@ -12,6 +12,7 @@ metadata:
   name: local-provisioner-config
   namespace: kube-system
 data:
+  setPVOwnerRef: "true"
   nodeLabelsForPV: |
     - kubernetes.io/hostname
   storageClassMap: |

diff --git a/pkg/pdapi/pdapi.go b/pkg/pdapi/pdapi.go
@@ -155,6 +155,8 @@ type PDClient interface {
 	SetStoreLabels(storeID uint64, labels map[string]string) (bool, error)
 	// DeleteStore deletes a TiKV store from cluster
 	DeleteStore(storeID uint64) error
+	// SetStoreState sets store to specified state.
+	SetStoreState(storeID uint64, state string) error
 	// DeleteMember deletes a PD member from cluster
 	DeleteMember(name string) error
 	// DeleteMemberByID deletes a PD member from cluster
@@ -403,6 +405,30 @@ func (pc *pdClient) DeleteStore(storeID uint64) error {
 	return fmt.Errorf("failed to delete store %d: %v", storeID, string(body))
 }
 
+// SetStoreState sets store to specified state.
+func (pc *pdClient) SetStoreState(storeID uint64, state string) error {
+	apiURL := fmt.Sprintf("%s/%s/%d/state?state=%s", pc.url, storePrefix, storeID, state)
+	req, err := http.NewRequest("POST", apiURL, nil)
+	if err != nil {
+		return err
+	}
+	res, err := pc.httpClient.Do(req)
+	if err != nil {
+		return err
+	}
+	defer httputil.DeferClose(res.Body)
+
+	if res.StatusCode == http.StatusOK || res.StatusCode == http.StatusNotFound {
+		return nil
+	}
+	body, err := ioutil.ReadAll(res.Body)
+	if err != nil {
+		return err
+	}
+
+	return fmt.Errorf("failed to delete store %d: %v", storeID, string(body))
+}
+
 func (pc *pdClient) DeleteMemberByID(memberID uint64) error {
 	var exist bool
 	members, err := pc.GetMembers()
@@ -666,6 +692,7 @@ const (
 	GetTombStoneStoresActionType       ActionType = "GetTombStoneStores"
 	GetStoreActionType                 ActionType = "GetStore"
 	DeleteStoreActionType              ActionType = "DeleteStore"
+	SetStoreStateActionType            ActionType = "SetStoreState"
 	DeleteMemberByIDActionType         ActionType = "DeleteMemberByID"
 	DeleteMemberActionType             ActionType = "DeleteMember "
 	SetStoreLabelsActionType           ActionType = "SetStoreLabels"
@@ -790,6 +817,15 @@ func (pc *FakePDClient) DeleteStore(id uint64) error {
 	return nil
 }
 
+func (pc *FakePDClient) SetStoreState(id uint64, state string) error {
+	if reaction, ok := pc.reactions[SetStoreStateActionType]; ok {
+		action := &Action{ID: id}
+		_, err := reaction(action)
+		return err
+	}
+	return nil
+}
+
 func (pc *FakePDClient) DeleteMemberByID(id uint64) error {
 	if reaction, ok := pc.reactions[DeleteMemberByIDActionType]; ok {
 		action := &Action{ID: id}

diff --git a/tests/actions.go b/tests/actions.go
@@ -294,6 +294,7 @@ type OperatorConfig struct {
 	ValidatingEnabled         bool
 	Cabundle                  string
 	BackupImage               string
+	AutoFailover              *bool
 }
 
 type TidbClusterConfig struct {
@@ -408,7 +409,6 @@ func (oi *OperatorConfig) OperatorHelmSetString(m map[string]string) string {
 	set := map[string]string{
 		"operatorImage":                                oi.Image,
 		"tidbBackupManagerImage":                       oi.BackupImage,
-		"controllerManager.autoFailover":               "true",
 		"scheduler.logLevel":                           "4",
 		"testMode":                                     strconv.FormatBool(oi.TestMode),
 		"admissionWebhook.cabundle":                    oi.Cabundle,
@@ -442,6 +442,9 @@ func (oi *OperatorConfig) OperatorHelmSetString(m map[string]string) string {
 	if oi.Enabled(features.AdvancedStatefulSet) {
 		set["advancedStatefulset.create"] = "true"
 	}
+	if oi.AutoFailover != nil {
+		set["controllerManager.autoFailover"] = strconv.FormatBool(*oi.AutoFailover)
+	}
 
 	arr := make([]string, 0, len(set))
 	for k, v := range set {

diff --git a/tests/e2e/e2e.go b/tests/e2e/e2e.go
@@ -34,6 +34,7 @@ import (
 	"github.com/pingcap/tidb-operator/tests"
 	e2econfig "github.com/pingcap/tidb-operator/tests/e2e/config"
 	utilimage "github.com/pingcap/tidb-operator/tests/e2e/util/image"
+	utilnode "github.com/pingcap/tidb-operator/tests/e2e/util/node"
 	v1 "k8s.io/api/core/v1"
 	storagev1 "k8s.io/api/storage/v1"
 	apiextensionsclientset "k8s.io/apiextensions-apiserver/pkg/client/clientset/clientset"
@@ -122,6 +123,14 @@ func setupSuite() {
 		e2elog.Logf("WARNING: Waiting for all daemonsets to be ready failed: %v", err)
 	}
 
+	ginkgo.By("Initializing all nodes")
+	nodeList, err := c.CoreV1().Nodes().List(metav1.ListOptions{})
+	framework.ExpectNoError(err)
+	for _, node := range nodeList.Items {
+		framework.Logf("Initializing node %q", node.Name)
+		framework.ExpectNoError(utilnode.InitNode(&node))
+	}
+
 	// By using default storage class in GKE/EKS (aws), network attached storage
 	// which be used and we must clean them later.
 	// We set local-storage class as default for simplicity.