Update 2014-11-07

I've revamped cassandra-docker with a new entrypoint. It is available on Github as tobert/cassandra:2.1.1 and tobert/cassandra:2.0.11. The instructions are fairly similar but the syntax and paths have changed. The README.md has further details.

TL;DR: use volumes for /var/lib/cassandra

As a fan of Linux containers and evangelist for Apache Cassandra, I get a lot of questions about running Cassandra in Docker containers. I've run production Cassandra clusters with cgroups in the past and had good luck with it, but normally don't have much use for namespaces. One place where full containers really helps is running clusters on a single machine.

A new image

I looked around at some of the community Docker images there are some out there, but I wanted something with less moving pieces. Ideally, users (you, hopefully!) can pull an image and be up and running with Cassandra without having to see a git repo. I also wanted to go through the process of building a new Docker image since I haven't done it in a really long time. The new code can be found at https://github.com/tobert/cassandra-docker

Single-node Quick Start

If all you want to do is run a single-node Cassandra instance in docker, this should get you going. This command will pull down the tobert/dsc208 image I've published in the public registry and then run it with data stored in the host's /srv/cassandra directory.

docker pull tobert/dsc20
mkdir /srv/cassandra
docker run -v /srv/cassandra:/var/lib/cassandra dsc208

Cluster Quick Start

The process starts out the same way for a cluster, but there's a little more work to do to pass in the seeds. You will need to create a data directory for each node in the cluster, then tell Docker to use that as a volume inside each container. To get gossip working one node needs to be started first so it can be used as a seed for the others.

docker pull tobert/dsc208

mkdir /srv/{cass0,cass1,cass2}

# pass in heap settings with envvars
HEAP="-e MAX_HEAP_SIZE=1G -e HEAP_NEWSIZE=256M"

docker run -d --name cass0 $HEAP -v /srv/cass0:/var/lib/cassandra dsc208
IP=$(< /srv/cass0/etc/listen_address.txt)

# you only need to set SEEDS the first time, but it doesn't hurt
docker run -d --name cass1 $HEAP -e SEEDS=$IP -v /srv/cass1:/var/lib/cassandra dsc208
docker run -d --name cass2 $HEAP -e SEEDS=$IP -v /srv/cass2:/var/lib/cassandra dsc208

# wait a minute or two for the gossip to converge
nodetool -h $IP status

Building the Image

This image uses a simple Dockerfile. It should be straightforward.

git clone https://github.com/tobert/cassandra-docker
cd cassandra-docker
docker pull ubuntu:raring
docker build -t dsc208 .

Conclusion

Running Cassandra in Docker isn't ideal, but it can be quite handy for development and QA work. For now, I still don't recommend dockerizing production Cassandra clusters. This may change when the tooling gets better and it's clear that a stable environment can be provided. After all, your database is responsible for all of the important state in applications and IMO should be treated with a little more care.

Edit: running Cassandra in Docker is pretty safe these days. As mentioned before, using volumes for the data storage is a must for durability and performance. In addition, I also recommend avoiding the bridge/NAT networking and run Cassandra containers with --net=host. This provides the simplest way to connect to the outside world and guarantees a stable IP address to the guest. Host networking also has the lowest overhead performance-wise so your cluster should perform nearly as well as it does on bare metal.

If you find any issues, a Github issue against https://github.com/tobert/cassandra-docker would be great or hit me on Twitter: @AlTobey.