Starting an ArangoDB cluster the easy way

THIS REPO HAS BEEN MOVED TO

https://github.com/arangodb-helper/ArangoDBStarter

Please use that one only from now on.

Building

Just do

make local

and the executable is in ./bin named after the current OS & architecture (e.g. arangodb-linux-amd64). You can copy the binary anywhere in your PATH. A link to the binary for the local OS & architecture is made to ./arangodb. This program will run on Linux, OSX or Windows.

Starting a cluster

Install ArangoDB in the usual way as binary package. Then:

On host A:

arangodb

This will use port 4000 to wait for colleagues (3 are needed for a resilient agency). On host B: (can be the same as A):

arangodb --join A

This will contact A on port 4000 and register. On host C: (can be same as A or B):

arangodb --join A

This will contact A on port 4000 and register.

From the moment on when 3 have joined, each will fire up an agent, a coordinator and a dbserver and the cluster is up. Ports are shown on the console.

Additional servers can be added in the same way.

If two or more of the arangodb instances run on the same machine, one has to use the --dataDir option to let each use a different directory.

The arangodb program will find the ArangoDB executable and the other installation files automatically. If this fails, use the --arangod and --jsdir options described below.

Running in Docker

The executable can be run inside Docker. In that case it will also run all servers in a Docker container.

First make sure the docker images are build using:

make docker 

When running in Docker it is important to care about the volume mappings on the container. Typically you will start the executable in docker with the following commands.

export IP=<IP of docker host>
docker volume create arangodb1
docker run -it --name=adb1 --rm -p 4000:4000 \
    -v arangodb1:/data \
    -v /var/run/docker.sock:/var/run/docker.sock \
    arangodb/arangodb-starter \
    --dockerContainer=adb1 --ownAddress=$IP

The executable will show the commands needed to run the other instances.

Note that the commands above create a docker volume. If you're running on Linux it is also possible to use a host mapped volume. Make sure to map it on /data.

Common options

• --dataDir path

path is the directory in which all data is stored. (default "./")

In the directory, there will be a single file setup.json used for restarts and a directory for each instances that runs on this machine. Different instances of arangodb must use different data directories.

• --join addr

join a cluster with master at address addr (default "")

• --agencySize int

number of agents in agency (default 3).

This number has to be positive and odd, and anything beyond 5 probably does not make sense. The default 3 allows for the failure of one agent.

• --ownAddress addr

addr is the address under which this server is reachable from the outside.

Usually, this option does not have to be specified. Only in the case that --agencySize is set to 1 (see below), the master has to know under which address it can be reached from the outside. If you specify localhost here, then all instances must run on the local machine.

• --docker image

image is the name of a Docker image to run instead of the normal executable. For each started instance a Docker container is launched. Usually one would use the Docker image arangodb/arangodb.

• --dockerContainer containerName

containerName is the name of a Docker container that is used to run the executable. This argument is required when running the executable in docker.

Esoteric options

• --masterPort int

port for arangodb master (default 4000).

This is the port used for communication of the arangodb instances amongst each other.

• --arangod path

path to the arangod executable (default varies from platform to platform, an executable is searched in various places).

This option only has to be specified if the standard search fails.

• --jsDir path

path to JS library directory (default varies from platform to platform, this is coupled to the search for the executable).

This option only has to be specified if the standard search fails.

• --startCoordinator bool

This indicates whether or not a coordinator instance should be started (default true).

• --startDBserver bool

This indicates whether or not a DBserver instance should be started (default true).

• --rr path

path to rr executable to use if non-empty (default ""). Expert and debugging only.

• --verbose bool

show more information (default false).

• --dockerUser user

user is an expression to be used for docker run with the --user option. One can give a user id or a user id and a group id, separated by a colon. The purpose of this option is to limit the access rights of the process in the Docker container.

• --dockerEndpoint endpoint

endpoint is the URL used to reach the docker host. This is needed to run the executable in docker. The default value is "unix:///var/run/docker.sock".

• --dockerNetHost bool

If dockerNetHost is set, all docker container will be started with the --net=host option.

• --dockerPrivileged bool

If dockerPrivileged is set, all docker container will be started with the --privileged option turned on.

Future plans

• deploy this program as a Docker image
• bundle this program with the usual distribution
• make port usage configurable
• support SSL
• support authentication

Technical explanation as to what happens

The procedure is essentially that the first instance of arangodb (aka the "master") offers an HTTP service on port 4000 for peers to register. Every instance that registers becomes a slave. As soon as there are agencySize peers, every instance of arangodb starts up an agent (if it is one of the first 3), a DBserver, and a coordinator. The necessary command line options to link the arangod instances up are generated automatically. The cluster bootstraps and can be used.

Whenever an arangodb instance shuts down, it shuts down the arangod instances under its control as well. When the arangodb is started again, it recalls the old configuration from the setup.json file in its data directory, starts up its arangod instances again (with their data) and they join the cluster.

All network addresses are discovered from the HTTP communication between the arangodb instances. The ports used 4001(/4006/4011) for the agent, 4002(/4007/4012) for the coordinator, 4003(/4008/4013) for the DBserver) need to be free. If more than one instance of an arangodb are started on the same machine, the second will increase all these port numbers by 5 and so on.

In case the executable is running in Docker, it will use the Docker API to retrieve the port number of the Docker host to which the 4000 port number is mapped. The containers started by the executable will all map the port they use to the exact same host port.

Feedback

Feedback is very welcome in the form of github issues, pull requests or simply emails to me:

Max Neunhöffer <max@arangodb.com>