zk.md

Zookeeper:

# A headless service to create DNS records
apiVersion: v1
kind: Service
metadata:
  name: zk
  labels:
    app: zookeeper
spec:
  # We really don't care about ports
  ports:
  - port: 2888
    name: peer
  - port: 3888
    name: leader-election
  # *.zk.default.svc.cluster.local
  clusterIP: None
  selector:
    app: zookeeper
---
# First quorum member
apiVersion: v1
kind: ReplicationController
metadata:
  name: zk-server1
  labels:
    app: zookeeper
spec:
  replicas: 1
  selector:
    id: zk-server1
  template:
    metadata:
      annotations:
        # s1.zk.default.svc.cluster.local
        pod.beta.kubernetes.io/hostname: s1
        pod.beta.kubernetes.io/subdomain: zk
      labels:
        app: zookeeper
        id: zk-server1
    spec:
      # Node label only needed because we're using local storage.
      # Run `kubectl label node foo role=server1` to set.
      nodeSelector:
        role: server1
      containers:
      - image: jplock/zookeeper
        imagePullPolicy: Always
        name: zookeeper
        # Replace with more robust code to grow/shrink members.
        env:
        - name: ZK_SERVER_ID
          value: "1"
        command:
          - sh
          - -c
          - "echo -e \"server.1=s1.zk.default.svc.cluster.local:2888:3888\nserver.2=s2.zk.default.svc.cluster.local:2888:3888\n\" >> /opt/zookeeper/conf/zoo.cfg; \
             mkdir -p /tmp/zookeeper; echo ${ZK_SERVER_ID} > /tmp/zookeeper/myid; \
             /opt/zookeeper/bin/zkServer.sh start-foreground"
        # For this to be useful the pod needs to land on the same node everytime,
        # this will happen because of the node label.
        volumeMounts:
        - name: datadir
          mountPath: /tmp/zookeeper
        ports:
        - containerPort: 2888
          name: peer
        - containerPort: 3888
          name: leader-election
      volumes:
      - name: datadir
        hostPath:
          path: /var/lib/zookeeper
---
# Second quorum member
apiVersion: v1
kind: ReplicationController
metadata:
  name: zk-server2
  labels:
    app: zookeeper
spec:
  replicas: 1
  selector:
    id: zk-server2
  template:
    metadata:
      annotations:
        # s2.zk.default.svc.cluster.local
        pod.beta.kubernetes.io/hostname: s2
        pod.beta.kubernetes.io/subdomain: zk
      labels:
        app: zookeeper
        id: zk-server2
    spec:
      # Node label only needed because we're using local storage.
      # Run `kubectl label node foo role=server2` to set.
      nodeSelector:
        role: server2
      containers:
      - image: jplock/zookeeper
        imagePullPolicy: Always
        name: zookeeper
        # Replace with more robust code to grow/shrink members.
        env:
        - name: ZK_SERVER_ID
          value: "2"
        command:
          - sh
          - -c
          - "echo -e \"server.1=s1.zk.default.svc.cluster.local:2888:3888\nserver.2=s2.zk.default.svc.cluster.local:2888:3888\n\" >> /opt/zookeeper/conf/zoo.cfg; \
             mkdir -p /tmp/zookeeper; echo ${ZK_SERVER_ID} > /tmp/zookeeper/myid; \
             /opt/zookeeper/bin/zkServer.sh start-foreground"
        # For this to be useful the pod needs to land on the same node everytime,
        # this will happene because of the node label.
        volumeMounts:
        - name: datadir
          mountPath: /tmp/zookeeper
        ports:
        - containerPort: 2888
          name: peer
        - containerPort: 3888
          name: leader-election
      volumes:
      - name: datadir
        hostPath:
          path: /var/lib/zookeeper

Eg:

$ kubectl label node nodename1 role=server1
$ kubectl label node nodename2 role=server2
$ kubectl create -f zookeeper.yaml

This will create a zookeeper ensemble statically configured. Killing a pod will bring it back on the same node. It will use the same hostPath as its datadir. You can also use pvcs for storage instead of hostPath.