Merge pull request #26913 from bprashanth/petset_e2e

test/e2e/petset.go

@@ -37,6 +37,7 @@ import (
 	"k8s.io/kubernetes/pkg/controller/petset"
 	"k8s.io/kubernetes/pkg/labels"
 	"k8s.io/kubernetes/pkg/runtime"
+	"k8s.io/kubernetes/pkg/util/sets"
 	"k8s.io/kubernetes/pkg/util/wait"
 	utilyaml "k8s.io/kubernetes/pkg/util/yaml"
 	"k8s.io/kubernetes/test/e2e/framework"
@@ -49,9 +50,15 @@ const (
 	zookeeperManifestPath   = "test/e2e/testing-manifests/petset/zookeeper"
 	mysqlGaleraManifestPath = "test/e2e/testing-manifests/petset/mysql-galera"
 	redisManifestPath       = "test/e2e/testing-manifests/petset/redis"
+	// Should the test restart petset clusters?
+	// TODO: enable when we've productionzed bringup of pets in this e2e.
+	restartCluster = false
 )
 
-var _ = framework.KubeDescribe("PetSet", func() {
+// Time: 25m, slow by design.
+// GCE Quota requirements: 3 pds, one per pet manifest declared above.
+// GCE Api requirements: nodes and master need storage r/w permissions.
+var _ = framework.KubeDescribe("PetSet [Slow] [Feature:PetSet]", func() {
 	f := framework.NewDefaultFramework("petset")
 	var ns string
 	var c *client.Client
@@ -81,13 +88,16 @@ var _ = framework.KubeDescribe("PetSet", func() {
 			Expect(err).NotTo(HaveOccurred())
 		})
 
+		AfterEach(func() {
+			if CurrentGinkgoTestDescription().Failed {
+				dumpDebugInfo(c, ns)
+			}
+			framework.Logf("Deleting all petset in ns %v", ns)
+			deleteAllPetSets(c, ns)
+		})
+
 		It("should provide basic identity [Feature:PetSet]", func() {
 			By("creating petset " + psName + " in namespace " + ns)
-			defer func() {
-				err := c.Apps().PetSets(ns).Delete(psName, nil)
-				Expect(err).NotTo(HaveOccurred())
-			}()
-
 			petMounts := []api.VolumeMount{{Name: "datadir", MountPath: "/data/"}}
 			podMounts := []api.VolumeMount{{Name: "home", MountPath: "/home"}}
 			ps := newPetSet(psName, ns, headlessSvcName, 3, petMounts, podMounts, labels)
@@ -99,7 +109,7 @@ var _ = framework.KubeDescribe("PetSet", func() {
 			By("Saturating pet set " + ps.Name)
 			pst.saturate(ps)
 
-			cmd := "echo $(hostname) > /data/hostname"
+			cmd := "echo $(hostname) > /data/hostname; sync;"
 			By("Running " + cmd + " in all pets")
 			pst.execInPets(ps, cmd)
 
@@ -114,10 +124,6 @@ var _ = framework.KubeDescribe("PetSet", func() {
 
 		It("should handle healthy pet restarts during scale [Feature:PetSet]", func() {
 			By("creating petset " + psName + " in namespace " + ns)
-			defer func() {
-				err := c.Apps().PetSets(ns).Delete(psName, nil)
-				Expect(err).NotTo(HaveOccurred())
-			}()
 
 			petMounts := []api.VolumeMount{{Name: "datadir", MountPath: "/data/"}}
 			podMounts := []api.VolumeMount{{Name: "home", MountPath: "/home"}}
@@ -126,6 +132,7 @@ var _ = framework.KubeDescribe("PetSet", func() {
 			Expect(err).NotTo(HaveOccurred())
 
 			pst := petSetTester{c: c}
+
 			pst.waitForRunning(1, ps)
 
 			By("Marking pet at index 0 as healthy.")
@@ -152,17 +159,17 @@ var _ = framework.KubeDescribe("PetSet", func() {
 		})
 	})
 
-	framework.KubeDescribe("Deploy clustered applications", func() {
+	framework.KubeDescribe("Deploy clustered applications [Slow] [Feature:PetSet]", func() {
 		BeforeEach(func() {
 			framework.SkipUnlessProviderIs("gce")
 		})
 
 		AfterEach(func() {
-			// TODO: delete pvs
-			if !CurrentGinkgoTestDescription().Failed {
-				return
+			if CurrentGinkgoTestDescription().Failed {
+				dumpDebugInfo(c, ns)
 			}
-			dumpDebugInfo(c, ns)
+			framework.Logf("Deleting all petset in ns %v", ns)
+			deleteAllPetSets(c, ns)
 		})
 
 		It("should creating a working zookeeper cluster [Feature:PetSet]", func() {
@@ -174,9 +181,11 @@ var _ = framework.KubeDescribe("PetSet", func() {
 			By("Creating foo:bar in member with index 0")
 			pet.write(0, map[string]string{"foo": "bar"})
 
-			By("Restarting pet set " + ps.Name)
-			pst.restart(ps)
-			pst.waitForRunning(ps.Spec.Replicas, ps)
+			if restartCluster {
+				By("Restarting pet set " + ps.Name)
+				pst.restart(ps)
+				pst.waitForRunning(ps.Spec.Replicas, ps)
+			}
 
 			By("Reading value under foo from member with index 2")
 			if v := pet.read(2, "foo"); v != "bar" {
@@ -193,9 +202,11 @@ var _ = framework.KubeDescribe("PetSet", func() {
 			By("Creating foo:bar in member with index 0")
 			pet.write(0, map[string]string{"foo": "bar"})
 
-			By("Restarting pet set " + ps.Name)
-			pst.restart(ps)
-			pst.waitForRunning(ps.Spec.Replicas, ps)
+			if restartCluster {
+				By("Restarting pet set " + ps.Name)
+				pst.restart(ps)
+				pst.waitForRunning(ps.Spec.Replicas, ps)
+			}
 
 			By("Reading value under foo from member with index 2")
 			if v := pet.read(2, "foo"); v != "bar" {
@@ -212,9 +223,11 @@ var _ = framework.KubeDescribe("PetSet", func() {
 			By("Creating foo:bar in member with index 0")
 			pet.write(0, map[string]string{"foo": "bar"})
 
-			By("Restarting pet set " + ps.Name)
-			pst.restart(ps)
-			pst.waitForRunning(ps.Spec.Replicas, ps)
+			if restartCluster {
+				By("Restarting pet set " + ps.Name)
+				pst.restart(ps)
+				pst.waitForRunning(ps.Spec.Replicas, ps)
+			}
 
 			By("Reading value under foo from member with index 2")
 			if v := pet.read(2, "foo"); v != "bar" {
@@ -381,6 +394,7 @@ func petSetFromManifest(fileName, ns string) *apps.PetSet {
 	return &ps
 }
 
+// petSetTester has all methods required to test a single petset.
 type petSetTester struct {
 	c *client.Client
 }
@@ -429,30 +443,36 @@ func (p *petSetTester) deletePetAtIndex(index int, ps *apps.PetSet) {
 	}
 }
 
-func (p *petSetTester) restart(ps *apps.PetSet) {
+func (p *petSetTester) scale(ps *apps.PetSet, count int) error {
 	name := ps.Name
 	ns := ps.Namespace
-	oldReplicas := ps.Spec.Replicas
-	p.update(ns, name, func(ps *apps.PetSet) { ps.Spec.Replicas = 0 })
+	p.update(ns, name, func(ps *apps.PetSet) { ps.Spec.Replicas = count })
 
 	var petList *api.PodList
 	pollErr := wait.PollImmediate(petsetPoll, petsetTimeout, func() (bool, error) {
 		petList = p.getPodList(ps)
-		if len(petList.Items) == 0 {
+		if len(petList.Items) == count {
 			return true, nil
 		}
 		return false, nil
 	})
 	if pollErr != nil {
-		ts := []string{}
+		unhealthy := []string{}
 		for _, pet := range petList.Items {
-			if pet.DeletionTimestamp != nil {
-				ts = append(ts, fmt.Sprintf("%v", pet.DeletionTimestamp.Time))
+			delTs, phase, readiness := pet.DeletionTimestamp, pet.Status.Phase, api.IsPodReady(&pet)
+			if delTs != nil || phase != api.PodRunning || !readiness {
+				unhealthy = append(unhealthy, fmt.Sprintf("%v: deletion %v, phase %v, readiness %v", pet.Name, delTs, phase, readiness))
 			}
 		}
-		framework.Failf("Failed to scale petset down to 0, %d remaining pods with deletion timestamps: %v", len(petList.Items), ts)
+		return fmt.Errorf("Failed to scale petset to %d in %v. Remaining pods:\n%v", count, petsetTimeout, unhealthy)
 	}
-	p.update(ns, name, func(ps *apps.PetSet) { ps.Spec.Replicas = oldReplicas })
+	return nil
+}
+
+func (p *petSetTester) restart(ps *apps.PetSet) {
+	oldReplicas := ps.Spec.Replicas
+	ExpectNoError(p.scale(ps, 0))
+	p.update(ps.Namespace, ps.Name, func(ps *apps.PetSet) { ps.Spec.Replicas = oldReplicas })
 }
 
 func (p *petSetTester) update(ns, name string, update func(ps *apps.PetSet)) {
@@ -542,6 +562,74 @@ func (p *petSetTester) setHealthy(ps *apps.PetSet) {
 	}
 }
 
+func deleteAllPetSets(c *client.Client, ns string) {
+	pst := &petSetTester{c: c}
+	psList, err := c.Apps().PetSets(ns).List(api.ListOptions{LabelSelector: labels.Everything()})
+	ExpectNoError(err)
+
+	// Scale down each petset, then delete it completely.
+	// Deleting a pvc without doing this will leak volumes, #25101.
+	errList := []string{}
+	for _, ps := range psList.Items {
+		framework.Logf("Scaling petset %v to 0", ps.Name)
+		if err := pst.scale(&ps, 0); err != nil {
+			errList = append(errList, fmt.Sprintf("%v", err))
+		}
+		framework.Logf("Deleting petset %v", ps.Name)
+		if err := c.Apps().PetSets(ps.Namespace).Delete(ps.Name, nil); err != nil {
+			errList = append(errList, fmt.Sprintf("%v", err))
+		}
+	}
+
+	// pvs are global, so we need to wait for the exact ones bound to the petset pvcs.
+	pvNames := sets.NewString()
+	// TODO: Don't assume all pvcs in the ns belong to a petset
+	pvcPollErr := wait.PollImmediate(petsetPoll, petsetTimeout, func() (bool, error) {
+		pvcList, err := c.PersistentVolumeClaims(ns).List(api.ListOptions{LabelSelector: labels.Everything()})
+		if err != nil {
+			framework.Logf("WARNING: Failed to list pvcs, retrying %v", err)
+			return false, nil
+		}
+		for _, pvc := range pvcList.Items {
+			pvNames.Insert(pvc.Spec.VolumeName)
+			// TODO: Double check that there are no pods referencing the pvc
+			framework.Logf("Deleting pvc: %v with volume %v", pvc.Name, pvc.Spec.VolumeName)
+			if err := c.PersistentVolumeClaims(ns).Delete(pvc.Name); err != nil {
+				return false, nil
+			}
+		}
+		return true, nil
+	})
+	if pvcPollErr != nil {
+		errList = append(errList, fmt.Sprintf("Timeout waiting for pvc deletion."))
+	}
+
+	pollErr := wait.PollImmediate(petsetPoll, petsetTimeout, func() (bool, error) {
+		pvList, err := c.PersistentVolumes().List(api.ListOptions{LabelSelector: labels.Everything()})
+		if err != nil {
+			framework.Logf("WARNING: Failed to list pvs, retrying %v", err)
+			return false, nil
+		}
+		waitingFor := []string{}
+		for _, pv := range pvList.Items {
+			if pvNames.Has(pv.Name) {
+				waitingFor = append(waitingFor, fmt.Sprintf("%v: %+v", pv.Name, pv.Status))
+			}
+		}
+		if len(waitingFor) == 0 {
+			return true, nil
+		}
+		framework.Logf("Still waiting for pvs of petset to disappear:\n%v", strings.Join(waitingFor, "\n"))
+		return false, nil
+	})
+	if pollErr != nil {
+		errList = append(errList, fmt.Sprintf("Timeout waiting for pv provisioner to delete pvs, this might mean the test leaked pvs."))
+	}
+	if len(errList) != 0 {
+		ExpectNoError(fmt.Errorf("%v", strings.Join(errList, "\n")))
+	}
+}
+
 func ExpectNoError(err error) {
 	Expect(err).NotTo(HaveOccurred())
 }

test/e2e/testing-manifests/petset/mysql-galera/petset.yaml

@@ -57,6 +57,7 @@ spec:
             }
         ]'
     spec:
+      terminationGracePeriodSeconds: 0
       containers:
       - name: mysql
         image: gcr.io/google_containers/mysql-galera:e2e
@@ -73,10 +74,12 @@ spec:
         - --defaults-file=/etc/mysql/my-galera.cnf
         - --user=root
         readinessProbe:
-          # TODO: If docker exec is buggy just use nc mysql-id 3306 as a ping.
-          httpGet:
-            path: /healthz
-            port: 8080
+          # TODO: If docker exec is buggy just use gcr.io/google_containers/mysql-healthz:1.0
+          exec:
+            command:
+            - sh
+            - -c
+            - "mysql -u root -e 'show databases;'"
           initialDelaySeconds: 15
           timeoutSeconds: 5
           successThreshold: 2
@@ -85,11 +88,6 @@ spec:
           mountPath: /var/lib/
         - name: config
           mountPath: /etc/mysql
-      - name: healthz
-        image: gcr.io/google_containers/mysql-healthz:1.0
-        args:
-        - --port=8080
-        - --verbose=true
       volumes:
       - name: config
         emptyDir: {}
@@ -104,4 +102,4 @@ spec:
       accessModes: [ "ReadWriteOnce" ]
       resources:
         requests:
-          storage: 10Gi
+          storage: 1Gi

test/e2e/testing-manifests/petset/redis/petset.yaml

@@ -14,9 +14,9 @@ spec:
         pod.alpha.kubernetes.io/init-containers: '[
             {
                 "name": "install",
-                "image": "gcr.io/google_containers/redis-install:0.1",
+                "image": "gcr.io/google_containers/redis-install-3.2.0:e2e",
                 "imagePullPolicy": "Always",
-                "args": ["--version=3.2.0", "--install-into=/opt", "--work-dir=/work-dir"],
+                "args": ["--install-into=/opt", "--work-dir=/work-dir"],
                 "volumeMounts": [
                     {
                         "name": "opt",
@@ -57,6 +57,7 @@ spec:
             }
         ]'
     spec:
+      terminationGracePeriodSeconds: 0
       containers:
       - name: redis
         image: debian:jessie
@@ -94,4 +95,4 @@ spec:
       accessModes: [ "ReadWriteOnce" ]
       resources:
         requests:
-          storage: 10Gi
+          storage: 1Gi

test/e2e/testing-manifests/petset/zookeeper/petset.yaml

@@ -14,9 +14,9 @@ spec:
         pod.alpha.kubernetes.io/init-containers: '[
             {
                 "name": "install",
-                "image": "gcr.io/google_containers/zookeeper-install:0.1",
+                "image": "gcr.io/google_containers/zookeeper-install-3.5.0-alpha:e2e",
                 "imagePullPolicy": "Always",
-                "args": ["--version=3.5.0-alpha", "--install-into=/opt", "--work-dir=/work-dir"],
+                "args": ["--install-into=/opt", "--work-dir=/work-dir"],
                 "volumeMounts": [
                     {
                         "name": "opt",
@@ -47,7 +47,7 @@ spec:
                 "volumeMounts": [
                     {
                         "name": "opt",
-                        "mountPath": "/opt/"
+                        "mountPath": "/opt"
                     },
                     {
                         "name": "workdir",
@@ -61,6 +61,7 @@ spec:
             }
         ]'
     spec:
+      terminationGracePeriodSeconds: 0
       containers:
       - name: zk
         image: java:openjdk-8-jre
@@ -85,7 +86,10 @@ spec:
         - name: datadir
           mountPath: /tmp/zookeeper
         - name: opt
-          mountPath: /opt/
+          mountPath: /opt
+        # Mount the work-dir just for debugging
+        - name: workdir
+          mountPath: /work-dir
       volumes:
       - name: opt
         emptyDir: {}
@@ -100,4 +104,4 @@ spec:
       accessModes: [ "ReadWriteOnce" ]
       resources:
         requests:
-          storage: 10Gi
+          storage: 1Gi

test/images/pets/redis/Dockerfile

+# Copyright 2016 The Kubernetes Authors All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# TODO: get rid of bash dependency and switch to plain busybox.
+# The tar in busybox also doesn't seem to understand compression.
+FROM debian:jessie
+MAINTAINER Prashanth.B <beeps@google.com>
+
+# TODO: just use standard redis when there is one for 3.2.0.
+RUN apt-get update && apt-get install -y wget make gcc
+
+# See README.md
+RUN wget -qO /redis-3.2.0.tar.gz http://download.redis.io/releases/redis-3.2.0.tar.gz && \
+    tar -xzf /redis-3.2.0.tar.gz -C /tmp/ && rm /redis-3.2.0.tar.gz
+
+# Clean out existing deps before installation
+# see https://github.com/antirez/redis/issues/722
+RUN cd /tmp/redis-3.2.0 && make distclean && mkdir -p /redis && \
+    make install INSTALL_BIN=/redis && \
+    mv /tmp/redis-3.2.0/redis.conf /redis/redis.conf && \
+    rm -rf /tmp/redis-3.2.0
+
+ADD on-start.sh /
+# See contrib/pets/peer-finder for details
+RUN wget -qO /peer-finder https://storage.googleapis.com/kubernetes-release/pets/peer-finder
+ADD install.sh /
+RUN chmod -c 755 /install.sh /on-start.sh /peer-finder
+Entrypoint ["/install.sh"]

test/images/pets/redis/Makefile

+# Copyright 2016 The Kubernetes Authors All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+all: push
+
+TAG = e2e
+PREFIX = gcr.io/google_containers/redis-install-3.2.0
+
+container:
+	docker build -t $(PREFIX):$(TAG) .
+
+push: container
+	gcloud docker push $(PREFIX):$(TAG)
+
+clean:
+	docker rmi $(PREFIX):$(TAG)

test/images/pets/redis/README.md

+# Redis petset e2e tester
+
+The image in this directory is the init container for contrib/pets/redis but for one difference, it bakes a specific verfion of redis into the base image so we get deterministic test results without having to depend on a redis download server. Discussing the tradeoffs to either approach (download the version at runtime, or maintain an image per version) are outside the scope of this document.
+
+You can execute the image locally via:
+```
+$ docker run -it gcr.io/google_containers/redis-install-3.2.0:e2e --cmd --install-into=/opt --work-dir=/work-dir
+```
+To share the installation with other containers mount the appropriate volumes as `--install-into` and `--work-dir`, where `install-into` is the directory to install redis into, and `work-dir` is the directory to install the user/admin supplied on-{start,change} hook scripts.
+
+
+[![Analytics](https://kubernetes-site.appspot.com/UA-36037335-10/GitHub/test/images/pets/redis/README.md?pixel)]()

test/images/pets/redis/install.sh

+#! /bin/bash
+
+# Copyright 2016 The Kubernetes Authors All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# This volume is assumed to exist and is shared with parent of the init
+# container. It contains the redis installation.
+INSTALL_VOLUME="/opt"
+
+# This volume is assumed to exist and is shared with the peer-finder
+# init container. It contains on-start/change configuration scripts.
+WORK_DIR="/work-dir"
+
+VERSION="3.2.0"
+
+for i in "$@"
+do
+case $i in
+    -i=*|--install-into=*)
+    INSTALL_VOLUME="${i#*=}"
+    shift
+    ;;
+    -w=*|--work-dir=*)
+    WORK_DIR="${i#*=}"
+    shift
+    ;;
+    *)
+    # unknown option
+    ;;
+esac
+done
+
+echo installing config scripts into "${WORK_DIR}"
+mkdir -p "${WORK_DIR}"
+cp /on-start.sh "${WORK_DIR}"/
+cp /peer-finder "${WORK_DIR}"/
+
+echo installing redis-"${VERSION}" into "${INSTALL_VOLUME}"
+mkdir -p "${INSTALL_VOLUME}"
+mv /redis "${INSTALL_VOLUME}"/redis

test/images/pets/redis/on-start.sh

+#!/bin/bash
+
+# Copyright 2016 The Kubernetes Authors All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+set -e
+
+CFG=/opt/redis/redis.conf
+HOSTNAME=$(hostname)
+DATADIR="/data"
+# Port on which redis listens for connections.
+PORT=6379
+
+# Ping everyone but ourself to see if there's a master. Only one pet starts at
+# a time, so if we don't see a master we can assume the position is ours.
+while read -ra LINE; do
+    if [[ "${LINE}" == *"${HOSTNAME}"* ]]; then
+        sed -i -e "s|^bind.*$|bind ${LINE}|" ${CFG}
+    elif [ "$(/opt/redis/redis-cli -h $LINE info | grep role | sed 's,\r$,,')" = "role:master" ]; then
+        # TODO: More restrictive regex?
+        sed -i -e "s|^.*slaveof.*$|slaveof ${LINE} ${PORT}|" ${CFG}
+    fi
+done
+
+# Set the data directory for append only log and snapshot files. This should
+# be a persistent volume for consistency.
+sed -i -e "s|^.*dir .*$|dir ${DATADIR}|" ${CFG}
+
+# The append only log is written for every SET operation. Without this setting,
+# redis just snapshots periodically which is only safe for a cache. This will
+# produce an appendonly.aof file in the configured data dir.
+sed -i -e "s|^appendonly .*$|appendonly yes|" ${CFG}
+
+# Every write triggers an fsync. Recommended default is "everysec", which
+# is only safe for AP applications.
+sed -i -e "s|^appendfsync .*$|appendfsync always|" ${CFG}
+
+

test/images/pets/zookeeper/Dockerfile

+# Copyright 2016 The Kubernetes Authors All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# TODO: get rid of bash dependency and switch to plain busybox.
+# The tar in busybox also doesn't seem to understand compression.
+FROM debian:jessie
+MAINTAINER Prashanth.B <beeps@google.com>
+
+RUN apt-get update && apt-get install -y wget netcat
+
+ADD on-start.sh /
+# See contrib/pets/peer-finder for details
+RUN wget -qO /peer-finder https://storage.googleapis.com/kubernetes-release/pets/peer-finder
+
+# See README.md
+RUN wget -q -O /zookeeper-3.5.0-alpha.tar.gz http://apache.mirrors.pair.com/zookeeper/zookeeper-3.5.0-alpha/zookeeper-3.5.0-alpha.tar.gz && \
+    tar -xzf /zookeeper-3.5.0-alpha.tar.gz -C /tmp/ && mv /tmp/zookeeper-3.5.0-alpha /zookeeper && rm /zookeeper-3.5.0-alpha.tar.gz
+
+ADD install.sh /
+RUN chmod -c 755 /install.sh /on-start.sh /peer-finder
+Entrypoint ["/install.sh"]

test/images/pets/zookeeper/Makefile

+# Copyright 2016 The Kubernetes Authors All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+all: push
+
+TAG = e2e
+PREFIX = gcr.io/google_containers/zookeeper-install-3.5.0-alpha
+
+container:
+	docker build -t $(PREFIX):$(TAG) .
+
+push: container
+	gcloud docker push $(PREFIX):$(TAG)
+
+clean:
+	docker rmi $(PREFIX):$(TAG)

test/images/pets/zookeeper/README.md

+# Zookeeper petset e2e tester
+
+The image in this directory is the init container for contrib/pets/zookeeper but for one difference, it bakes a specific verfion of zookeeper into the base image so we get deterministic test results without having to depend on a zookeeper download server. Discussing the tradeoffs to either approach (download the version at runtime, or maintain an image per version) are outside the scope of this document.
+
+You can execute the image locally via:
+```
+$ docker run -it gcr.io/google_containers/zookeeper-install-3.5.0-alpha:e2e --cmd --install-into=/opt --work-dir=/work-dir
+```
+To share the installation with other containers mount the appropriate volumes as `--install-into` and `--work-dir`, where `install-into` is the directory to install zookeeper into, and `work-dir` is the directory to install the user/admin supplied on-{start,change} hook scripts.
+
+
+[![Analytics](https://kubernetes-site.appspot.com/UA-36037335-10/GitHub/test/images/pets/zookeeper/README.md?pixel)]()

test/images/pets/zookeeper/install.sh

+#! /bin/bash
+
+# Copyright 2016 The Kubernetes Authors All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# This volume is assumed to exist and is shared with parent of the init
+# container. It contains the zookeeper installation.
+INSTALL_VOLUME="/opt"
+
+# This volume is assumed to exist and is shared with the peer-finder
+# init container. It contains on-start/change configuration scripts.
+WORKDIR_VOLUME="/work-dir"
+
+# As of April-2016 is 3.4.8 is the latest stable, but versions 3.5.0 onward
+# allow dynamic reconfiguration.
+VERSION="3.5.0-alpha"
+
+for i in "$@"
+do
+case $i in
+    -i=*|--install-into=*)
+    INSTALL_VOLUME="${i#*=}"
+    shift
+    ;;
+    -w=*|--work-dir=*)
+    WORKDIR_VOLUME="${i#*=}"
+    shift
+    ;;
+    *)
+    # unknown option
+    ;;
+esac
+done
+
+echo installing config scripts into "${WORKDIR_VOLUME}"
+mkdir -p "${WORKDIR_VOLUME}"
+cp /on-start.sh "${WORKDIR_VOLUME}"/
+cp /peer-finder "${WORKDIR_VOLUME}"/
+
+echo installing zookeeper-"${VERSION}" into "${INSTALL_VOLUME}"
+mkdir -p "${INSTALL_VOLUME}"
+mv /zookeeper "${INSTALL_VOLUME}"/zookeeper
+cp "${INSTALL_VOLUME}"/zookeeper/conf/zoo_sample.cfg "${INSTALL_VOLUME}"/zookeeper/conf/zoo.cfg
+
+# TODO: Should dynamic config be tied to the version?
+IFS="." read -ra RELEASE <<< "${VERSION}"
+if [ $(expr "${RELEASE[1]}") -gt 4 ]; then
+    echo zookeeper-"${VERSION}" supports dynamic reconfiguration, enabling it
+    echo "standaloneEnabled=false" >> "${INSTALL_VOLUME}"/zookeeper/conf/zoo.cfg
+    echo "dynamicConfigFile="${INSTALL_VOLUME}"/zookeeper/conf/zoo.cfg.dynamic" >> "${INSTALL_VOLUME}"/zookeeper/conf/zoo.cfg
+fi
+
+# TODO: This is a hack, netcat is convenient to have in the zookeeper container
+# I want to avoid using a custom zookeeper image just for this. So copy it.
+NC=$(which nc)
+if [ "${NC}" != "" ]; then
+    echo copying nc into "${INSTALL_VOLUME}"
+    cp "${NC}" "${INSTALL_VOLUME}"
+fi

test/images/pets/zookeeper/on-start.sh

+#! /bin/bash
+
+# Copyright 2016 The Kubernetes Authors All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+set -eo pipefail
+
+# This script configures zookeeper cluster member ship for version of zookeeper
+# >= 3.5.0. It should not be used with the on-change.sh script in this example.
+# As of April-2016 is 3.4.8 is the latest stable.
+
+# Both /opt and /tmp/zookeeper are assumed to be volumes shared with the parent.
+# The format of each line in the dynamic config file is:
+# server.<1 based index>=<server-dns-name>:<peer port>:<election port>[:role];[<client port address>:]<client port>
+# <1 based index> is the server index that matches the id in datadir/myid
+# <peer port> is the port on which peers communicate to agree on updates
+# <election port> is the port used for leader election
+# [:role] can be set to observer, participant by default
+# <client port address> is optional and defaults to 0.0.0.0
+# <client port> is the port on which the server accepts client connections
+
+CFG=/opt/zookeeper/conf/zoo.cfg.dynamic
+CFG_BAK=/opt/zookeeper/conf/zoo.cfg.bak
+MY_ID_FILE=/tmp/zookeeper/myid
+HOSTNAME=$(hostname)
+
+while read -ra LINE; do
+    PEERS=("${PEERS[@]}" $LINE)
+done
+
+# Don't add the first member as an observer
+if [ ${#PEERS[@]} -eq 1 ]; then
+    # We need to write our index in this list of servers into MY_ID_FILE.
+    # Note that this may not always coincide with the hostname id.
+    echo 1 > "${MY_ID_FILE}"
+    echo "server.1=${PEERS[0]}:2888:3888;2181" > "${CFG}"
+    # TODO: zkServer-initialize is the safe way to handle changes to datadir
+    # because simply starting will create a new datadir, BUT if the user changed
+    # pod template they might end up with 2 datadirs and brief split brain.
+    exit
+fi
+
+# Every subsequent member is added as an observer and promoted to a participant
+echo "" > "${CFG_BAK}"
+i=0
+LEADER=$HOSTNAME
+for peer in "${PEERS[@]}"; do
+    let i=i+1
+    if [[ "${peer}" == *"${HOSTNAME}"* ]]; then
+      MY_ID=$i
+      MY_NAME=${peer}
+      echo $i > "${MY_ID_FILE}"
+      echo "server.${i}=${peer}:2888:3888:observer;2181" >> "${CFG_BAK}"
+    else
+      if [[ $(echo srvr | /opt/nc "${peer}" 2181 | grep Mode) = "Mode: leader" ]]; then
+        LEADER="${peer}"
+      fi
+      echo "server.${i}=${peer}:2888:3888:participant;2181" >> "${CFG_BAK}"
+    fi
+done
+
+# zookeeper won't start without myid anyway.
+# This means our hostname wasn't in the peer list.
+if [ ! -f "${MY_ID_FILE}" ]; then
+  exit 1
+fi
+
+# Once the dynamic config file is written it shouldn't be modified, so the final
+# reconfigure needs to happen through the "reconfig" command.
+cp ${CFG_BAK} ${CFG}
+
+# TODO: zkServer-initialize is the safe way to handle changes to datadir
+# because simply starting will create a new datadir, BUT if the user changed
+# pod template they might end up with 2 datadirs and brief split brain.
+/opt/zookeeper/bin/zkServer.sh start
+
+# TODO: We shouldn't need to specify the address of the master as long as
+# there's quorum. According to the docs the new server is just not allowed to
+# vote, it's still allowed to propose config changes, and it knows the
+# existing members of the ensemble from *its* config.
+ADD_SERVER="server.$MY_ID=$MY_NAME:2888:3888:participant;0.0.0.0:2181"
+/opt/zookeeper/bin/zkCli.sh reconfig -s "${LEADER}":2181 -add "${ADD_SERVER}"
+
+# Prove that we've actually joined the running cluster
+ITERATION=0
+until $(echo config | /opt/nc localhost 2181 | grep "${ADD_SERVER}" > /dev/null); do
+  echo $ITERATION] waiting for updated config to sync back to localhost
+  sleep 1
+  let ITERATION=ITERATION+1
+  if [ $ITERATION -eq 20 ]; then
+    exit 1
+  fi
+done
+
+/opt/zookeeper/bin/zkServer.sh stop