How to Manage Your GPU Cluster

Eli Mernit

December 6, 2024 1 min read

This is the final feature in our first launch week, where we ship a brand new feature every day, five days in a row.

If you're self-hosting Beam, you'll need a CLI to manage the machines in your cluster. Today, we're shipping an upgraded suite of management commands for self-hosted users to easily manage machines and workers.

What Can You Do With It?

For context, Beta9 is the open-source project that powers Beam. You can self-host Beam and run your workloads on arbitrary hardware.

Connect Machines to Beam

You can connect arbitrary hardware to Beam using the machine create command. This command will print a cURL command, which you'll run on the node in order to connect the machine to the control plane running Beam.

$ beta9 machine create --pool lambda-a100-40

=> Created machine with ID: '9541cbd2'. Use the following command to set up the node:

#!/bin/bash
sudo curl -L -o agent https://release.beam.cloud/agent/agent && \
sudo chmod +x agent && \
sudo ./agent --token "AUTH_TOKEN" \
  --machine-id "9541cbd2" \
  --tailscale-url "" \
  --tailscale-auth "AUTH_TOKEN" \
  --pool-name "lambda-a100-40" \
  --provider-name "lambda"

Cordon and Drain Workers

If you've used Kubernetes before, you're already familiar with the concept of cordon and drain. If you aren't, we'll give a quick refresher:

Cordon: This marks a node as unschedulable so that no new workloads are assigned to it. Typically, you'll cordon a node when it's experiencing issues or requires maintenance.

Drain: This gracefully evicts all existing workloads from the node, usually to prepare it for maintenance or upgrades.

We've added cordon and drain commands in the beta9 CLI, giving you more control over your workers:

$ beta9 worker cordon 3e1fa6f3
$ beta9 worker drain 3e1fa6f3
$ beta9 worker uncordon 3e1fa6f3

List Workers

Here's the command to list all the workers connected to your Beam or Beta9 cluster:

ID         Pool          Status      Priority   CPU       Memory       GPUs
 ────────────────────────────────────────────────────────────────────────
  a1f4b9c7   build         available   0          7,000m    27.00 GiB    0
  e3c8d572   build         available   0          7,000m    27.00 GiB    0
  92b1f0e8   default       available   1          16,000m   32.00 GiB    0
  f8d37c5b   default       available   1          2,000m    16.00 GiB    0
  8c72e6da   default       available   1          2,000m    16.00 GiB    0
  b2d8a1fc   default       available   1          2,000m    16.00 GiB    0
  d4e3c9ba   default       available   1          2,000m    16.00 GiB    0

What's Next

As a next step, check out the Beta9 repo. You can self-host Beta9 and connect it to your own hardware.

If you have any feedback on the workflow or feature requests, we’d love to hear from you in our Slack Community.

This is Launch 5/5 for this week! You can follow along with our upcoming launches on Twitter.

Eli Mernit

Published December 6, 2024