Vultr Open Cluster Manager

Your cluster manager is ready!

• Your server's IP address is: use.your.ip.
• The root password is: use.your.root.password.

Usage:

• SSH to your new cluster manager.

• Inspect and edit /root/config.yml to your specifications. Some things to note below:

  • instance_plan : This is the Vultr SKU that you will be deploying for your cluster nodes.
  • instance_gpu : If you will be using Slurm, you must specify the GPU model from the plan chosen to be used in the Slurm configuration.
  • instance_gpu_count : If you will be using Slurm, you must specify the GPU count from the plan chosen to be used in the Slurm configuration.
  • instance_slurm_memory : Slurm needs to know how much of the cluster node's RAM can be used. Generally set this to 15% less than the total available.
  • os_id : ID of the Operating system to be installed on cluster nodes. Query to https://api.vultr.com/v2/os can get available OS provided by Vultr. Default is Ubuntu 22.04 LTS
  • instance_region : Autofilled with the region of the cluster manager instance. If you change this, the automatically created and attached VPC will be invalid.
  • hostprefix : Prefix of each cluster node's hostname. Defaults to #region#-cluster-node
  • hostsuffix: Suffix of each cluster node's hostname. Defaults to gpu.local

• You may wish to have the rest completed automatically in which case you can run /root/build-cluster.sh

• Change into the Terraform directory. cd /root/terraform
• Initialize Terraform. terraform init
• Check the Terraform plan. terraform plan
• Apply the Terraform plan. terraform apply
• Wait for cluster nodes to be built and come online before proceeding.
• Change into the Ansible directory. cd /root/ansible
• Run the Ansible playbook. ansible-playbook -i hosts cluster.yml This will perform the following actions:
  • Update package repo on all cluster nodes and the manager.
  • Install Grafana Alloy on all cluster nodes and the manager.
  • Configure Grafana Alloy to send logs to the provided Loki instance (if provided).
  • Install and configure the Slurm Daemon slurmd on all cluster nodes.
  • Install and configure the Slurm Controller slurmctld on the manager.
  • Brings online a Grafana and Prometheus docker container on the manager. See /root/docker-compose.yml.
  • Install Prometheus Node Exporter on all cluster nodes.
  • Adds Node Exporter dashboard to local Grafana instance.
• Connect to your Grafana interface at http://use.your.ip:3000/.
  • Your Grafana username is: admin.
  • Your Grafana password is: Grafana Password.