Managing Clusters

Clusters are the core concept of Rift. A cluster is a set of nodes that are managed together. You can run jobs on the cluster, inspect the cluster state, add/remove nodes, add/remove users etc.

The most common use cases for clusters are:

1. Organize nodes into logical groups. For example, you can have a cluster for each team in your organization.
2. Run jobs on nodes with specific hardware. For example, you can have a cluster with GPU nodes and run jobs that require GPU on this cluster.
3. Share nodes with other users. For example, you can pool compute resources together with your friends to run jobs on a larger cluster.

Default Cluster

By default, all nodes are added to the default cluster and commands will work with it by default. You can see information about the default cluster by running the following command:

rift cluster info

Creating a Cluster

Throughout this tutorial we will be using a cluster named my_cluster. To create a new cluster use:

rift cluster create my_cluster

Now the cluster is created and you can inspect the cluster by running

rift cluster -c my_cluster info

To list all clusters you have access to use

rift cluster list

Adding Nodes to the Cluster

After creating a cluster you can add nodes to it. By default, all nodes are added to the default cluster, you can inspect them by invoking:

rift cluster nodes list

Let's add some first node from our default cluster to the my_cluster cluster under the name my_node.

NODE_ID=`rift cluster nodes list | head -n 1 | awk '{print $2}'`
rift cluster -c my_cluster nodes add $NODE_ID my_node

Note that node_id is a unique identifier of the node in the cluster, it is not the same as node_name. Node name is a human-readable name of the node to which you can refer to when running jobs.

Removing Nodes from the Cluster

You can remove nodes from the cluster. For example, let's remove the node that we just added:

rift cluster -c my_cluster nodes remove my_node

Renaming Nodes in the Cluster

You can also rename the node at any point in time. It is equivalent to simply removing and adding the node again. For example, let's add the node that we've just removed back to the cluster under the name my_new_node:

rift cluster -c my_cluster nodes add $NODE_ID my_new_node

However, if node is already in the cluster it is more convenient to use rename command. Let's rename node back to my_node:

rift cluster -c my_cluster nodes rename my_new_node my_node

Adding Users to the Cluster

You can add users to the cluster. Users will be able to run jobs on the cluster. Users are identified by their email address. For example, to add some@user.com to the cluster you can use the command:

rift cluster -c my_cluster users add some@user.com

Now authorized users can run jobs on the cluster. You can list authorized users by running the following command:

rift cluster -c my_cluster users list

Removing the Cluster

Finally, you can remove the cluster by running:

rift cluster remove my_cluster