Cassandra 2.1.2

In the simplest case, you can start a single node with docker run jeremyot/cassandra. The new node will be configured to listen on the address asigned to the eth0 interface and will accept RPC connectons on all interfaces.

Deploying a cluster is almost as easy. The container includes tooling to automatically bootstrap a cluster using etcd. For example:

ETCD=`docker run -d jeremyot/etcd`
ETCD_ADDR=`docker inspect --format '{{ .NetworkSettings.IPAddress }}' ${ETCD}`
docker run -d jeremyot/cassandra autoscale "${ETCD_ADDR}:4001" service/cassandra
docker run -d jeremyot/cassandra autoscale "${ETCD_ADDR}:4001" service/cassandra --join
docker run -d jeremyot/cassandra autoscale "${ETCD_ADDR}:4001" service/cassandra --join

will start a three node cassandra cluster. You can verify with

CASSANDRA_ADDR=`docker inspect --format "{{ .NetworkSettings.IPAddress }}" \`docker ps -lq\``
docker run -it --rm jeremyot/cassandra nodetool -h $CASSANDRA_ADDR status

You should see a normal nodetool status printout displaying information on all three nodes.

To persist data across restarts, be sure to mount Cassandra's data directory with -v /host/path/to/cassandra:/var/cassandra.

Cassandra may be configured by passing arguments when starting the container. A YAML library is included which will convert arbitrary arguments to YAML and write to cassandra.yaml. YAML paths passed as arguments are flattened and prefixed with --.

For example: --rpc_address=0.0.0.0 becomes rpc_address: 0.0.0.0 and --some.key=value becomes some: {key: value}. To support complex structures, values may also be passed as YAML strings. --some.key="{k1: [1, 2, 3], k2: [2, 3, 4]}" becomes

some:
  key:
    k1: [1, 2, 3]
    k2: [2, 3, 4]

For convenience, a few special arguments are supported:

• --listen_interface="etho": configures cassandra to bind to the specified interface.
• --seeds="10.0.0.2, 10.0.0.3": a shortcut for using SimpleSeedProvider with the specified seeds.
• --etc_seeds="127.0.0.1:4001/v2/keys/service/key": queries etcd for the initial list of keys, registered at the specified path. (used by autoscale). Will start as a standalone node if no results are found.
• --logger=INFO,stdout,R: sets the logger in log4j-server.properties.
• --infer_address=google.com: uses a connection to the specified address to infer which address Cassandra should bind to. useful for containers with multiple ethernet devices. Allows you to specify a remote host (e.g. another node in the Cassandra cluster) and know that Cassandra will be accessible to it.
• --join: Used with --etc_seeds to prevent bootstrapping as a standalone node. Will wait until some seeds are returned from etcd before continuing.

Along with Cassandra, there are a few tools included with this container. They are as follows:

• autoscale: as shown above, uses etcd to bootstrap a cluster. May be followed by the standard cassandra configuration arguments.
• etcdmon: a subcomponent of autoscale, uses etcdmon (https://github.com/JeremyOT/etcdmon) to register the server's status with etcd. Additional arguments may be passed in the form <ETCDMON_PARAMS> -- <CASSANDRA_PARAMS>.
• nodetool: runs nodetool and supports any standard nodetool arguments.
• cqlsh: runs cqlsh and supports any standard cqlsh arguments.

Autoscaling may be used in a few different ways. Called with autoscale 10.0.0.8:4001 service/cassandra, Cassandra will infer its listen_address by making a connection to 10.0.0.8. It will then attempt to retrieve a list of seeds by querying http://10.0.0.8:4001/v2/keys/service/cassandra, using its own address as the only seed if none are found. Finally, etcdmon will update http://10.0.0.8:4001/v2/keys/service/cassandra/<node_address> with value={"host": <node_address>}&ttl=30, pinging etcd every 10 seconds. Alternatively autoscale 10.0.0.8:4001 service/cassandra <interface_name> may be used to bind to the specified interface, and autoscale 10.0.0.8:4001 service/cassandra <remote_address> will infer the address by making a connection to the specified remote address. If the third argument does not begin with --, it will first be treated as an interface, then used to infer the address if no matching interface is found. In all three cases, standard configuration options may be appended.

Note: The address executable, used by the autoscale commands, is built from https://github.com/JeremyOT/address. It is pre-compiled to avoid the need to install Go inside the container.