一. 概述
kubernetes通过statefulset为zookeeper、etcd等这类有状态的应用程序提供完善支持，statefulset具备以下特性：
- 为pod提供稳定的唯一的网络标识
- 稳定值持久化存储：通过pv/pvc来实现
- 启动和停止pod保证有序：优雅的部署和伸缩性
本文阐述了如何在k8s集群上部署zookeeper和etcd有状态服务，并结合ceph实现数据持久化。
二. 总结
- 使用k8s的statefulset、storageclass、pv、pvc和ceph的rbd，能够很好的支持zookeeper、etcd这样的有状态服务部署到kubernetes集群上。
- k8s不会主动删除已经创建的pv、pvc对象，防止出现误删。
如果用户确定删除pv、pvc对象，同时还需要手动删除ceph段的rbd镜像。
- 遇到的坑
storageclass中引用的ceph客户端用户，必须要有mon rw，rbd rwx权限。如果没有mon write权限，会导致释放rbd锁失败，无法将rbd镜像挂载到其他的k8s worker节点。
- zookeeper使用探针检查zookeeper节点的健康状态，如果节点不健康，k8s将删除pod，并自动重建该pod，达到自动重启zookeeper节点的目的。
因zookeeper 3.4版本的集群配置，是通过静态加载文件zoo.cfg来实现的，所以当zookeeper节点pod ip变动后，需要重启zookeeper集群中的所有节点。
- etcd部署方式有待优化
本次试验中使用静态方式部署etcd集群，如果etcd节点变迁时，需要执行etcdctl member remove/add等命令手动配置etcd集群，严重限制了etcd集群自动故障恢复、扩容缩容的能力。因此，需要考虑对部署方式优化，改为使用DNS或者etcd descovery的动态方式部署etcd，才能让etcd更好的运行在k8s上。
三. zookeeper集群部署
1. 下载镜像
docker pull gcr.mirrors.ustc.edu.cn/google_containers/kubernetes-zookeeper:1.0-3.4.10docker tag gcr.mirrors.ustc.edu.cn/google_containers/kubernetes-zookeeper:1.0-3.4.10 172.16.18.100:5000/gcr.io/google_containers/kubernetes-zookeeper:1.0-3.4.10docker push 172.16.18.100:5000/gcr.io/google_containers/kubernetes-zookeeper:1.0-3.4.10
2. 定义ceph secret
cat << EOF | kubectl create -f -apiVersion: v1data: key: QVFBYy9ndGFRUno4QlJBQXMxTjR3WnlqN29PK3VrMzI1a05aZ3c9PQo=kind: Secretmetadata: creationTimestamp: 2017-11-20T10:29:05Z name: ceph-secret namespace: default resourceVersion: "2954730" selfLink: /api/v1/namespaces/default/secrets/ceph-secret uid: a288ff74-cddd-11e7-81cc-000c29f99475type: kubernetes.io/rbdEOF
3. 定义storageclass rbd存储
cat << EOF | kubectl create -f -apiVersion: storage.k8s.io/v1kind: StorageClassmetadata: name: cephparameters: adminId: admin adminSecretName: ceph-secret adminSecretNamespace: default fsType: ext4 imageFormat: "2" imagefeatures: layering monitors: 172.16.13.223 pool: k8s userId: admin userSecretName: ceph-secretprovisioner: kubernetes.io/rbdreclaimPolicy: DeleteEOF
4. 创建zookeeper集群
使用rbd存储zookeeper节点数据
cat << EOF | kubectl create -f ----apiVersion: v1kind: Servicemetadata: name: zk-hs labels: app: zkspec: ports: - port: 2888 name: server - port: 3888 name: leader-election clusterIP: None selector: app: zk---apiVersion: v1kind: Servicemetadata: name: zk-cs labels: app: zkspec: ports: - port: 2181 name: client selector: app: zk---apiVersion: policy/v1beta1kind: PodDisruptionBudgetmetadata: name: zk-pdbspec: selector: matchLabels: app: zk maxUnavailable: 1---apiVersion: apps/v1beta2 # for versions before 1.8.0 use apps/v1beta1kind: StatefulSetmetadata: name: zkspec: selector: matchLabels: app: zk serviceName: zk-hs replicas: 3 updateStrategy: type: RollingUpdate podManagementPolicy: Parallel template: metadata: labels: app: zk spec: affinity: podAntiAffinity: requiredDuringSchedulingIgnoredDuringExecution: - labelSelector: matchExpressions: - key: "app" operator: In values: - zk topologyKey: "kubernetes.io/hostname" containers: - name: kubernetes-zookeeper imagePullPolicy: Always image: "172.16.18.100:5000/gcr.io/google_containers/kubernetes-zookeeper:1.0-3.4.10" ports: - containerPort: 2181 name: client - containerPort: 2888 name: server - containerPort: 3888 name: leader-election command: - sh - -c - "start-zookeeper \ --servers=3 \ --data_dir=/var/lib/zookeeper/data \ --data_log_dir=/var/lib/zookeeper/data/log \ --conf_dir=/opt/zookeeper/conf \ --client_port=2181 \ --election_port=3888 \ --server_port=2888 \ --tick_time=2000 \ --init_limit=10 \ --sync_limit=5 \ --heap=512M \ --max_client_cnxns=60 \ --snap_retain_count=3 \ --purge_interval=12 \ --max_session_timeout=40000 \ --min_session_timeout=4000 \ --log_level=INFO" readinessProbe: exec: command: - sh - -c - "zookeeper-ready 2181" initialDelaySeconds: 10 timeoutSeconds: 5 livenessProbe: exec: command: - sh - -c - "zookeeper-ready 2181" initialDelaySeconds: 10 timeoutSeconds: 5 volumeMounts: - name: datadir mountPath: /var/lib/zookeeper securityContext: runAsUser: 1000 fsGroup: 1000 volumeClaimTemplates: - metadata: name: datadir annotations: volume.beta.kubernetes.io/storage-class: ceph spec: accessModes: [ "ReadWriteOnce" ] resources: requests: storage: 1GiEOF
查看创建结果
[root@172 zookeeper]# kubectl get noNAME STATUS ROLES AGE VERSION172.16.20.10 Ready50m v1.8.2172.16.20.11 Ready 2h v1.8.2172.16.20.12 Ready 1h v1.8.2[root@172 zookeeper]# kubectl get po -owide NAME READY STATUS RESTARTS AGE IP NODEzk-0 1/1 Running 0 8m 192.168.5.162 172.16.20.10zk-1 1/1 Running 0 1h 192.168.2.146 172.16.20.11[root@172 zookeeper]# kubectl get pv,pvcNAME CAPACITY ACCESS MODES RECLAIM POLICY STATUS CLAIM STORAGECLASS REASON AGEpv/pvc-226cb8f0-d322-11e7-9581-000c29f99475 1Gi RWO Delete Bound default/datadir-zk-0 ceph 1hpv/pvc-22703ece-d322-11e7-9581-000c29f99475 1Gi RWO Delete Bound default/datadir-zk-1 ceph 1hNAME STATUS VOLUME CAPACITY ACCESS MODES STORAGECLASS AGEpvc/datadir-zk-0 Bound pvc-226cb8f0-d322-11e7-9581-000c29f99475 1Gi RWO ceph 1hpvc/datadir-zk-1 Bound pvc-22703ece-d322-11e7-9581-000c29f99475 1Gi RWO ceph 1h
zk-0 pod的rbd的锁信息为
[root@ceph1 ceph]# rbd lock list kubernetes-dynamic-pvc-227b45e5-d322-11e7-90ab-000c29f99475 -p k8s --user adminThere is 1 exclusive lock on this image.Locker ID Address client.24146 kubelet_lock_magic_172.16.20.10 172.16.20.10:0/1606152350
5. 测试pod迁移
尝试将172.16.20.10节点设置为污点，让zk-0 pod自动迁移到172.16.20.12
kubectl cordon 172.16.20.10[root@172 zookeeper]# kubectl get noNAME STATUS ROLES AGE VERSION172.16.20.10 Ready,SchedulingDisabled58m v1.8.2172.16.20.11 Ready 2h v1.8.2172.16.20.12 Ready 1h v1.8.2kubectl delete po zk-0
观察zk-0的迁移过程
[root@172 zookeeper]# kubectl get po -owide -wNAME READY STATUS RESTARTS AGE IP NODEzk-0 1/1 Running 0 14m 192.168.5.162 172.16.20.10zk-1 1/1 Running 0 1h 192.168.2.146 172.16.20.11zk-0 1/1 Terminating 0 16m 192.168.5.162 172.16.20.10zk-0 0/1 Terminating 0 16m172.16.20.10zk-0 0/1 Terminating 0 16m 172.16.20.10zk-0 0/1 Terminating 0 16m 172.16.20.10zk-0 0/1 Terminating 0 16m 172.16.20.10zk-0 0/1 Terminating 0 16m 172.16.20.10zk-0 0/1 Pending 0 0s zk-0 0/1 Pending 0 0s 172.16.20.12zk-0 0/1 ContainerCreating 0 0s 172.16.20.12zk-0 0/1 Running 0 3s 192.168.3.4 172.16.20.12
此时zk-0正常迁移到172.16.20.12
[root@ceph1 ceph]# rbd lock list kubernetes-dynamic-pvc-227b45e5-d322-11e7-90ab-000c29f99475 -p k8s --user adminThere is 1 exclusive lock on this image.Locker ID Address client.24146 kubelet_lock_magic_172.16.20.10 172.16.20.10:0/1606152350 [root@ceph1 ceph]# rbd lock list kubernetes-dynamic-pvc-227b45e5-d322-11e7-90ab-000c29f99475 -p k8s --user adminThere is 1 exclusive lock on this image.Locker ID Address client.24154 kubelet_lock_magic_172.16.20.12 172.16.20.12:0/3715989358
之前在另外一个ceph集群测试这个zk pod迁移的时候，总是报错无法释放lock，经分析应该是使用的ceph账号没有相应的权限，所以导致释放lock失败。记录的报错信息如下：
Nov 27 10:45:55 172 kubelet: W1127 10:45:55.551768 11556 rbd_util.go:471] rbd: no watchers on kubernetes-dynamic-pvc-f35a411e-d317-11e7-90ab-000c29f99475Nov 27 10:45:55 172 kubelet: I1127 10:45:55.694126 11556 rbd_util.go:181] remove orphaned locker kubelet_lock_magic_172.16.20.12 from client client.171490: err exit status 13, output: 2017-11-27 10:45:55.570483 7fbdbe922d40 -1 did not load config file, using default settings.Nov 27 10:45:55 172 kubelet: 2017-11-27 10:45:55.600816 7fbdbe922d40 -1 Errors while parsing config file!Nov 27 10:45:55 172 kubelet: 2017-11-27 10:45:55.600824 7fbdbe922d40 -1 parse_file: cannot open /etc/ceph/ceph.conf: (2) No such file or directoryNov 27 10:45:55 172 kubelet: 2017-11-27 10:45:55.600825 7fbdbe922d40 -1 parse_file: cannot open ~/.ceph/ceph.conf: (2) No such file or directoryNov 27 10:45:55 172 kubelet: 2017-11-27 10:45:55.600825 7fbdbe922d40 -1 parse_file: cannot open ceph.conf: (2) No such file or directoryNov 27 10:45:55 172 kubelet: 2017-11-27 10:45:55.602492 7fbdbe922d40 -1 Errors while parsing config file!Nov 27 10:45:55 172 kubelet: 2017-11-27 10:45:55.602494 7fbdbe922d40 -1 parse_file: cannot open /etc/ceph/ceph.conf: (2) No such file or directoryNov 27 10:45:55 172 kubelet: 2017-11-27 10:45:55.602495 7fbdbe922d40 -1 parse_file: cannot open ~/.ceph/ceph.conf: (2) No such file or directoryNov 27 10:45:55 172 kubelet: 2017-11-27 10:45:55.602496 7fbdbe922d40 -1 parse_file: cannot open ceph.conf: (2) No such file or directoryNov 27 10:45:55 172 kubelet: 2017-11-27 10:45:55.651594 7fbdbe922d40 -1 auth: unable to find a keyring on /etc/ceph/ceph.client.k8s.keyring,/etc/ceph/ceph.keyring,/etc/ceph/keyring,/etc/ceph/keyring.bin,: (2) No such file or directoryNov 27 10:45:55 172 kubelet: rbd: releasing lock failed: (13) Permission deniedNov 27 10:45:55 172 kubelet: 2017-11-27 10:45:55.682470 7fbdbe922d40 -1 librbd: unable to blacklist client: (13) Permission denied
k8s rbd volume的实现代码：
if lock { // check if lock is already held for this host by matching lock_id and rbd lock id if strings.Contains(output, lock_id) { // this host already holds the lock, exit glog.V(1).Infof("rbd: lock already held for %s", lock_id) return nil } // clean up orphaned lock if no watcher on the image used, statusErr := util.rbdStatus(&b) if statusErr == nil && !used { re := regexp.MustCompile("client.* " + kubeLockMagic + ".*") locks := re.FindAllStringSubmatch(output, -1) for _, v := range locks { if len(v) > 0 { lockInfo := strings.Split(v[0], " ") if len(lockInfo) > 2 { args := []string{ "lock", "remove", b.Image, lockInfo[1], lockInfo[0], "--pool", b.Pool, "--id", b.Id, "-m", mon} args = append(args, secret_opt...) cmd, err = b.exec.Run("rbd", args...) # 执行rbd lock remove命令时返回了错误信息 glog.Infof("remove orphaned locker %s from client %s: err %v, output: %s", lockInfo[1], lockInfo[0], err, string(cmd)) } } } } // hold a lock: rbd lock add args := []string{ "lock", "add", b.Image, lock_id, "--pool", b.Pool, "--id", b.Id, "-m", mon} args = append(args, secret_opt...) cmd, err = b.exec.Run("rbd", args...) } 可以看到，rbd lock remove操作被拒绝了，原因是没有权限rbd: releasing lock failed: (13) Permission denied。
6. 测试扩容
zookeeper集群节点数从2个扩为3个。
集群节点数为2时，zoo.cfg的配置中定义了两个实例zookeeper@zk-0:/opt/zookeeper/conf$ cat zoo.cfg #This file was autogenerated DO NOT EDITclientPort=2181dataDir=/var/lib/zookeeper/datadataLogDir=/var/lib/zookeeper/data/logtickTime=2000initLimit=10syncLimit=5maxClientCnxns=60minSessionTimeout=4000maxSessionTimeout=40000autopurge.snapRetainCount=3autopurge.purgeInteval=12server.1=zk-0.zk-hs.default.svc.cluster.local:2888:3888server.2=zk-1.zk-hs.default.svc.cluster.local:2888:3888
使用kubectl edit statefulset zk命令修改replicas=3，start-zookeeper --servers=3,
[root@172 zookeeper]# kubectl get po -owide -wNAME READY STATUS RESTARTS AGE IP NODEzk-0 1/1 Running 0 1h 192.168.5.170 172.16.20.10zk-1 1/1 Running 0 1h 192.168.3.12 172.16.20.12zk-2 0/1 Pending 0 0szk-2 0/1 Pending 0 0s 172.16.20.11zk-2 0/1 ContainerCreating 0 0s 172.16.20.11zk-2 0/1 Running 0 1s 192.168.2.154 172.16.20.11zk-2 1/1 Running 0 11s 192.168.2.154 172.16.20.11zk-1 1/1 Terminating 0 1h 192.168.3.12 172.16.20.12zk-1 0/1 Terminating 0 1h 172.16.20.12zk-1 0/1 Terminating 0 1h 172.16.20.12zk-1 0/1 Terminating 0 1h 172.16.20.12zk-1 0/1 Terminating 0 1h 172.16.20.12zk-1 0/1 Pending 0 0s zk-1 0/1 Pending 0 0s 172.16.20.12zk-1 0/1 ContainerCreating 0 0s 172.16.20.12zk-1 0/1 Running 0 2s 192.168.3.13 172.16.20.12zk-1 1/1 Running 0 20s 192.168.3.13 172.16.20.12zk-0 1/1 Terminating 0 1h 192.168.5.170 172.16.20.10zk-0 0/1 Terminating 0 1h 172.16.20.10zk-0 0/1 Terminating 0 1h 172.16.20.10zk-0 0/1 Terminating 0 1h 172.16.20.10zk-0 0/1 Terminating 0 1h 172.16.20.10zk-0 0/1 Pending 0 0s zk-0 0/1 Pending 0 0s 172.16.20.10zk-0 0/1 ContainerCreating 0 0s 172.16.20.10zk-0 0/1 Running 0 2s 192.168.5.171 172.16.20.10zk-0 1/1 Running 0 12s 192.168.5.171 172.16.20.10
可以看到zk-0/zk-1都重启了，这样可以加载新的zoo.cfg配置文件，保证集群正确配置。
新的zoo.cfg配置文件记录了3个实例：[root@172 ~]# kubectl exec zk-0 -- cat /opt/zookeeper/conf/zoo.cfg#This file was autogenerated DO NOT EDITclientPort=2181dataDir=/var/lib/zookeeper/datadataLogDir=/var/lib/zookeeper/data/logtickTime=2000initLimit=10syncLimit=5maxClientCnxns=60minSessionTimeout=4000maxSessionTimeout=40000autopurge.snapRetainCount=3autopurge.purgeInteval=12server.1=zk-0.zk-hs.default.svc.cluster.local:2888:3888server.2=zk-1.zk-hs.default.svc.cluster.local:2888:3888server.3=zk-2.zk-hs.default.svc.cluster.local:2888:3888
7. 测试缩容
缩容的时候，zk集群也自动重启了所有的zk节点，缩容过程如下：
[root@172 ~]# kubectl get po -owide -wNAME READY STATUS RESTARTS AGE IP NODEzk-0 1/1 Running 0 5m 192.168.5.171 172.16.20.10zk-1 1/1 Running 0 6m 192.168.3.13 172.16.20.12zk-2 1/1 Running 0 7m 192.168.2.154 172.16.20.11zk-2 1/1 Terminating 0 7m 192.168.2.154 172.16.20.11zk-1 1/1 Terminating 0 7m 192.168.3.13 172.16.20.12zk-2 0/1 Terminating 0 8m172.16.20.11zk-1 0/1 Terminating 0 7m 172.16.20.12zk-2 0/1 Terminating 0 8m 172.16.20.11zk-1 0/1 Terminating 0 7m 172.16.20.12zk-1 0/1 Terminating 0 7m 172.16.20.12zk-1 0/1 Terminating 0 7m 172.16.20.12zk-1 0/1 Pending 0 0s zk-1 0/1 Pending 0 0s 172.16.20.12zk-1 0/1 ContainerCreating 0 0s 172.16.20.12zk-1 0/1 Running 0 2s 192.168.3.14 172.16.20.12zk-2 0/1 Terminating 0 8m 172.16.20.11zk-2 0/1 Terminating 0 8m 172.16.20.11zk-1 1/1 Running 0 19s 192.168.3.14 172.16.20.12zk-0 1/1 Terminating 0 7m 192.168.5.171 172.16.20.10zk-0 0/1 Terminating 0 7m 172.16.20.10zk-0 0/1 Terminating 0 7m 172.16.20.10zk-0 0/1 Terminating 0 7m 172.16.20.10zk-0 0/1 Pending 0 0s zk-0 0/1 Pending 0 0s 172.16.20.10zk-0 0/1 ContainerCreating 0 0s 172.16.20.10zk-0 0/1 Running 0 3s 192.168.5.172 172.16.20.10zk-0 1/1 Running 0 13s 192.168.5.172 172.16.20.10
四. etcd集群部署
1. 创建etcd集群
cat << EOF | kubectl create -f -apiVersion: v1kind: Servicemetadata: name: "etcd" annotations: # Create endpoints also if the related pod isn't ready service.alpha.kubernetes.io/tolerate-unready-endpoints: "true"spec: ports: - port: 2379 name: client - port: 2380 name: peer clusterIP: None selector: component: "etcd"---apiVersion: apps/v1beta1kind: StatefulSetmetadata: name: "etcd" labels: component: "etcd"spec: serviceName: "etcd" # changing replicas value will require a manual etcdctl member remove/add # command (remove before decreasing and add after increasing) replicas: 3 template: metadata: name: "etcd" labels: component: "etcd" spec: containers: - name: "etcd" image: "172.16.18.100:5000/quay.io/coreos/etcd:v3.2.3" ports: - containerPort: 2379 name: client - containerPort: 2380 name: peer env: - name: CLUSTER_SIZE value: "3" - name: SET_NAME value: "etcd" volumeMounts: - name: data mountPath: /var/run/etcd command: - "/bin/sh" - "-ecx" - | IP=$(hostname -i) for i in $(seq 0 $((${CLUSTER_SIZE} - 1))); do while true; do echo "Waiting for ${SET_NAME}-${i}.${SET_NAME} to come up" ping -W 1 -c 1 ${SET_NAME}-${i}.${SET_NAME}.default.svc.cluster.local > /dev/null && break sleep 1s done done PEERS="" for i in $(seq 0 $((${CLUSTER_SIZE} - 1))); do PEERS="${PEERS}${PEERS:+,}${SET_NAME}-${i}=http://${SET_NAME}-${i}.${SET_NAME}.default.svc.cluster.local:2380" done # start etcd. If cluster is already initialized the `--initial-*` options will be ignored. exec etcd --name ${HOSTNAME} \ --listen-peer-urls http://${IP}:2380 \ --listen-client-urls http://${IP}:2379,http://127.0.0.1:2379 \ --advertise-client-urls http://${HOSTNAME}.${SET_NAME}:2379 \ --initial-advertise-peer-urls http://${HOSTNAME}.${SET_NAME}:2380 \ --initial-cluster-token etcd-cluster-1 \ --initial-cluster ${PEERS} \ --initial-cluster-state new \ --data-dir /var/run/etcd/default.etcd## We are using dynamic pv provisioning using the "standard" storage class so## this resource can be directly deployed without changes to minikube (since## minikube defines this class for its minikube hostpath provisioner). In## production define your own way to use pv claims. volumeClaimTemplates: - metadata: name: data annotations: volume.beta.kubernetes.io/storage-class: ceph spec: accessModes: - "ReadWriteOnce" resources: requests: storage: 1GiEOF 创建完成之后的po,pv,pvc清单如下：
[root@172 etcd]# kubectl get po -owide NAME READY STATUS RESTARTS AGE IP NODEetcd-0 1/1 Running 0 15m 192.168.5.174 172.16.20.10etcd-1 1/1 Running 0 15m 192.168.3.16 172.16.20.12etcd-2 1/1 Running 0 5s 192.168.5.176 172.16.20.10
2. 测试缩容
kubectl scale statefulset etcd --replicas=2[root@172 ~]# kubectl get po -owide -wNAME READY STATUS RESTARTS AGE IP NODEetcd-0 1/1 Running 0 17m 192.168.5.174 172.16.20.10etcd-1 1/1 Running 0 17m 192.168.3.16 172.16.20.12etcd-2 1/1 Running 0 1m 192.168.5.176 172.16.20.10etcd-2 1/1 Terminating 0 1m 192.168.5.176 172.16.20.10etcd-2 0/1 Terminating 0 1m172.16.20.10
检查集群健康
kubectl exec etcd-0 -- etcdctl cluster-healthfailed to check the health of member 42c8b94265b9b79a on http://etcd-2.etcd:2379: Get http://etcd-2.etcd:2379/health: dial tcp: lookup etcd-2.etcd on 10.96.0.10:53: no such hostmember 42c8b94265b9b79a is unreachable: [http://etcd-2.etcd:2379] are all unreachablemember 9869f0647883a00d is healthy: got healthy result from http://etcd-1.etcd:2379member c799a6ef06bc8c14 is healthy: got healthy result from http://etcd-0.etcd:2379cluster is healthy
发现缩容后，etcd-2并没有从etcd集群中自动删除，可见这个etcd镜像对自动扩容缩容的支持并不够好。
我们手工删除掉etcd-2[root@172 etcd]# kubectl exec etcd-0 -- etcdctl member remove 42c8b94265b9b79aRemoved member 42c8b94265b9b79a from cluster[root@172 etcd]# kubectl exec etcd-0 -- etcdctl cluster-health member 9869f0647883a00d is healthy: got healthy result from http://etcd-1.etcd:2379member c799a6ef06bc8c14 is healthy: got healthy result from http://etcd-0.etcd:2379cluster is healthy
3. 测试扩容
从etcd.yaml的启动脚本中可以看出，扩容时新启动一个etcd pod时参数--initial-cluster-state new，该etcd镜像并不支持动态扩容，可以考虑使用基于dns动态部署etcd集群的方式来修改启动脚本，这样才能支持etcd cluster动态扩容。