Following this example, you will create a functional Apache Storm cluster using Kubernetes and Docker.

You will setup an Apache ZooKeeper service, a Storm master service (a.k.a. Nimbus server), and a set of Storm workers (a.k.a. supervisors).

For the impatient expert, jump straight to the tl;dr section.

This example assumes you have a Kubernetes cluster installed and running, and that you have installed the kubectl command line tool somewhere in your path. Please see the getting started for installation instructions for your platform.

ZooKeeper is a distributed coordination service that Storm uses as a bootstrap and for state storage.

Use the examples/storm/zookeeper.json file to create a pod running the ZooKeeper service.

$ kubectl create -f examples/storm/zookeeper.json

Then, use the examples/storm/zookeeper-service.json file to create a logical service endpoint that Storm can use to access the ZooKeeper pod.

$ kubectl create -f examples/storm/zookeeper-service.json

You should make sure the ZooKeeper pod is Running and accessible before proceeding.

$ kubectl get pods
POD                 IP                  CONTAINER(S)        IMAGE(S)             HOST                          LABELS                      STATUS
zookeeper           192.168.86.4        zookeeper           mattf/zookeeper      172.18.145.8/172.18.145.8     name=zookeeper              Running

$ kubectl get services
NAME                LABELS                                    SELECTOR            IP                  PORT
kubernetes          component=apiserver,provider=kubernetes   <none>              10.254.0.2          443
kubernetes-ro       component=apiserver,provider=kubernetes   <none>              10.254.0.1          80
zookeeper           name=zookeeper                            name=zookeeper      10.254.139.141      2181

$ echo ruok | nc 10.254.139.141 2181; echo
imok

The Nimbus service is the master (or head) service for a Storm cluster. It depends on a functional ZooKeeper service.

Use the examples/storm/storm-nimbus.json file to create a pod running the Nimbus service.

$ kubectl create -f examples/storm/storm-nimbus.json

Then, use the examples/storm/storm-nimbus-service.json file to create a logical service endpoint that Storm workers can use to access the Nimbus pod.

$ kubectl create -f examples/storm/storm-nimbus-service.json

Ensure that the Nimbus service is running and functional.

$ kubectl get services
NAME                LABELS                                    SELECTOR            IP                  PORT
kubernetes          component=apiserver,provider=kubernetes   <none>              10.254.0.2          443
kubernetes-ro       component=apiserver,provider=kubernetes   <none>              10.254.0.1          80
zookeeper           name=zookeeper                            name=zookeeper      10.254.139.141      2181
nimbus              name=nimbus                               name=nimbus         10.254.115.208      6627

$ sudo docker run -it -w /opt/apache-storm mattf/storm-base sh -c '/configure.sh 10.254.139.141 10.254.115.208; ./bin/storm list'
...
No topologies running.

The Storm workers (or supervisors) do the heavy lifting in a Storm cluster. They run your stream processing topologies and are managed by the Nimbus service.

The Storm workers need both the ZooKeeper and Nimbus services to be running.

Use the examples/storm/storm-worker-controller.json file to create a ReplicationController that manages the worker pods.

$ kubectl create -f examples/storm/storm-worker-controller.json

One way to check on the workers is to get information from the ZooKeeper service about how many clients it has.

$  echo stat | nc 10.254.139.141 2181; echo
Zookeeper version: 3.4.6--1, built on 10/23/2014 14:18 GMT
Clients:
 /192.168.48.0:44187[0](queued=0,recved=1,sent=0)
 /192.168.45.0:39568[1](queued=0,recved=14072,sent=14072)
 /192.168.86.1:57591[1](queued=0,recved=34,sent=34)
 /192.168.8.0:50375[1](queued=0,recved=34,sent=34)
 /192.168.45.0:39576[1](queued=0,recved=34,sent=34)

Latency min/avg/max: 0/2/2570
Received: 23199
Sent: 23198
Connections: 5
Outstanding: 0
Zxid: 0xa39
Mode: standalone
Node count: 13

There should be one client from the Nimbus service and one per worker. Ideally, you should get stat output from ZooKeeper before and after creating the ReplicationController.

(Pull requests welcome for alternative ways to validate the workers)

kubectl create -f zookeeper.json

kubectl create -f zookeeper-service.json

Make sure the ZooKeeper Pod is running (use: kubectl get pods).

kubectl create -f storm-nimbus.json

kubectl create -f storm-nimbus-service.json

Make sure the Nimbus Pod is running.

kubectl create -f storm-worker-controller.json