The First Cry of Atom Today is the first day of the rest of my life.

Multiple Hadoop Cluster on Docker

As a hadoop developer, there are several times when I want to create multiple node hadoop cluster more easily. First I came up with using VirtualBox and Vagrant. But it was very slow to launch one cluster. Besides the more nodes we added, the slower launching time be. I cannot wait to check each change and debug it. Every developers may think so.

Docker is an open platform for building, shipping and running distributed applications. This software sufficiently fits this use case. Docker is a container service not a virtual machine. So it is light and faster launching. I searched hadoop cluster image for docker. But I cannot find good one which seems to be maintained active. Some images cannot be used now at latest Docker or MacOSX.

This is the reason why I crated docker-hadoop-cluster. This is the modified version of sequenceiq/hadoop-docker. With docker-hadoop-cluster, you can build multiple node hadoop cluster on your laptop about one minute. This is not for production only for development and tutorial. So docker-hadoop-cluster can deploy arbitrary hadoop package build by yourself. Here is the instruction how to use this image.

Build images

hadoop-base is a base image of hadoop service. This image includes JDK, hadoop package configurations etc. This image also can include your self-build hadoop package. In order to bind your hadoop package, hadoop-X.Y.Z.tar.gz package assumed be put under hadoop-base directory. Then it’s the time to build image.

$ cp hadoop-3.0.0-SNAPSHOT.tar.gz hadoop-base
$ cd hadoop-base
$ docker build -t lewuathe/hadoop-base .

Next one is the image for master server. This image includes master service such as namenode and resource manager.

$ cd hadoop-master
$ docker build -t lewuathe/hadoop-master .

The last one is for slave services. This image includes slave service such as datanode and node manager.

$ cd hadoop-slave
$ docker build -t lewuathe/hadoop-slave .

Run cluster

docker-hadoop-cluster includes launching script.

## Create hadoop cluster with 5 slave nodes.
$ bin/build_cluster.sh --slaves 5 launch

## Destroy your previsout hadoop cluster.
$ bin/build_cluster.sh --slaves 5 destroy

Welcome patches

The biggest thing I want to do was to create maintained image for hadoop cluster. Although it might be difficult to do this alone, anyone can contribute and help this images. This is why I published this image. If you have any questions or are interested in this projects, please let me know. Thank you!