With the release of version 1.9, Ansible now includes a module named gluster_volume for managing Gluster distributed file system volumes. The new module offers significant improvements over using shell commands including less code, better error handling, clearer declarative statements, and idempotency, the ability to apply an operation once and only once.

With my recent patch that adds support for multiple Gluster bricks, or backing directories on the storage nodes, I’ve been able to refactor our Ansible playbooks to use the new gluster_volume module.

Setting up Gluster with Ansible is now dead simple:

The playbook above does the following:

1. Specifes the bricks, volume name, and number of replicas as variables
2. Adds the Gluster repository to yum
3. Installs the Gluster packages
4. Starts the Gluster service
5. Creates a volume
6. Starts the volume

Creating a volume is a bit complicated and warrants some explanation. The bricks and cluster parameters take strings seperated by commas. (The machines specified in the cluster parameter are used for peer probing).The force parameters allows us to specify bricks in the same partition as / (for use in VMs). The present state tells the gluster_volume module to create the volume. The run_once parameter specified that the task is run on only one machine rather than on each machine.

The gluster_volume builds the brick list by iterating over each host and then each brick. For example, the brick list for two hosts and two bricks would be

host1:/brick1 host2:/brick1 host1:/brick2 /host2:/brick2


When R replicas, N nodes, and M bricks are used, Gluster groups every (N * M) / R bricks into a replica. With 2 replicas, the first replica would consist of host1:/brick1 and host2:/brick1 while the second replica would consist of host1:/brick2 and /host2:/brick2. Thus, we must be careful to coordinate the ordering of the bricks and the number of nodes and bricks with the number of replicas.

In the future, I would like to improve the gluster_volume API to make the specification of replica and striping groups more explicit and provide some error checking.