A docker image could be built with build-image.sh script.

The image will have following pre-installed components:

• openjdk-jdk;
• mesos native libs;

/opt/kafka-mesos folder will contain following content:

• kafka*.tgz - kafka distro;
• kafka-mesos*.jar - kafka-mesos distro;
• kafka-mesos.sh - shell script to run kafka-mesos;

No default configuration file for kafka-mesos is provided. All configuration params should be specified via cli options.

Building image is done by build-image.sh script. Please refer to help via ./build-image.sh -h.

Example:

# ./build-image.sh

Note: this would not push the image. Push should be done manually after testing.

If you have docker-machine installed, with virtualbox, you can run the following commands:

$ docker-machine create --driver virtualbox dev
$ eval "$(docker-machine env dev)"
$ ./build-image.sh

Running image using docker. Required networking params should be provided. Image has no entry point, so ./kafka-mesos.sh should be specified explicitly.

Example:

# sudo docker run -it -p 7000:7000 --add-host=master:192.168.3.5 `whoami`/kafka-mesos ./kafka-mesos.sh scheduler \
--master=master:5050 --zk=master:2181 --api=http://<accessible-ip>:7000 --storage=zk:/kafka-mesos

Where accessible-ip - is the IP address of running host, accessible from mesos nodes.

Note: if you want to inspect image content you can skip specifying ./kafka-mesos.sh entry point. In that case you will get shell access.

To run image in Marathon it is better to pre-pull image on all required nodes using:

# sudo docker pull <image-name>

After that, scheduler process could be created via Marathon API via POST /v2/apps call. Example:

{
    "container": {
        "type": "DOCKER",
        "docker": {
            "network": "HOST",
            "image": "$imageName"
        }
    },
    "id":"kafka-mesos-scheduler",
    "cpus": 0.5,
    "mem": 256,
    "ports": [7000],
    "cmd": "./kafka-mesos.sh scheduler --master=master:5050 --zk=master:2181 --api=http://master:7000 --storage=zk:/kafka-mesos",
    "instances": 1,
    "constraints": [["hostname", "LIKE", "master"]]
}

Then the cli should be able to connect to url http://master:7000 from your api client machine (typically different than the mesos slaves). Example:

# ./kafka-mesos.sh broker list --api=http://master:7000
no brokers

Now you can configure and start brokers.