Deploy Apache Ray

Things on this page are fragmentary and immature notes/thoughts of the author. Please read with your own judgement!

Manual Cluster Setup¶

It is very easy to set up Apache Ray on a small cluster manually. The instructions at Manual Cluster Setup is very clear. Just one comment. You do not need to configure SSH keys for nodes in the cluster for manual configuration. When you start the master node manually (using the command ray start --head), it will print a command to add new nodes into the cluster. The command (e.g., ray start --address='10.148.186.18:12030' --redis-password='5241590000000000') includes the Redis password for authentication to join the cluster.

By default, the Ray dashboard is visitable only on the head node. To make it visitable by public IP, you can include the option --webui-host=0.0.0.0 when starting Ray on the head node.

Private Cluster¶

python/ray/autoscaler/local/example-full.yaml

The Ray autoscaler installs Ray on the head and worker nodes by default. That is if you don’t set values for the fileds head_setup_commands and setup_commands, Ray is automatically installed/updated on the head and worker nodes. You can disable this behavior by set a dummy command (e.g., echo "Starting Ray ...") for the 2 fileds.

https://ray.readthedocs.io/en/latest/using-ray-on-a-cluster.html

https://ray.readthedocs.io/en/latest/autoscaling.html

https://ray.readthedocs.io/en/latest/autoscaling.html#quick-start-private-cluster

https://ray.readthedocs.io/en/latest/using-ray-on-a-cluster.html