Skip to content

Update ARC runner skill doc for 5-node cluster #457

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
msaroufim merged 1 commit into from
Mar 5, 2026
Merged

Update ARC runner skill doc for 5-node cluster #457

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
32 changes: 29 additions & 3 deletions .claude/skills/arc-gpu-runners.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -108,7 +108,7 @@ sudo k3s kubectl logs -n arc-systems -l actions.github.com/scale-set-name=arc-ru
- **GPU isolation**: The AMD device plugin exposes `amd.com/gpu` as a k8s resource. Each runner pod requests exactly 1 GPU. Kubernetes guarantees no two pods share a GPU — each gets a unique `/dev/dri/renderD*` device.
- **CPU isolation**: Each pod gets 14 dedicated cores via cgroup limits (`nproc` reports 14 inside the container).
- **RAM isolation**: Each pod gets a 340Gi memory limit enforced by cgroups. Exceeding it triggers OOM kill.
- **Autoscaling**: With `minRunners: 0` and `maxRunners: 8`, runners spin up on demand when GitHub queues jobs and are destroyed after completion (ephemeral runners).
- **Autoscaling**: With `minRunners: 0` and `maxRunners: 40`, runners spin up on demand when GitHub queues jobs and are destroyed after completion (ephemeral runners). The scheduler spreads pods across all 5 nodes.

## Resource Budget (per MI355X node)

Expand All @@ -120,7 +120,7 @@ The MI355X node has 126 allocatable CPUs, ~3TB RAM, and 8 GPUs.
| RAM | 340 Gi |
| GPU | 1x MI355X |

At max capacity (8 runners): 112 cores, 2720 Gi, 8 GPUs. Remaining resources go to system pods.
At max capacity (40 runners across 5 nodes): 8 runners per node, each using 14 cores / 340 Gi / 1 GPU.

## Using in Workflows

Expand Down Expand Up @@ -148,6 +148,32 @@ sudo KUBECONFIG=/etc/rancher/k3s/k3s.yaml helm upgrade arc-runner-set \
--version 0.10.1
```

## Adding New Nodes

ARC is cluster-wide — no per-node setup is needed. When a new node joins the k3s cluster:

1. The AMD GPU device plugin (DaemonSet) auto-deploys to the new node
2. The k8s scheduler can immediately place runner pods on it
3. No changes needed to workflows or the GitHub launcher

The only thing to update is `maxRunners` to reflect the new total GPU count:

```bash
# Example: 3 nodes × 8 GPUs = 24 max runners
sudo KUBECONFIG=/etc/rancher/k3s/k3s.yaml helm upgrade arc-runner-set \
--namespace arc-runners \
--set maxRunners=24 \
--reuse-values \
oci://ghcr.io/actions/actions-runner-controller-charts/gha-runner-scale-set \
--version 0.10.1
```

Verify the new node has GPUs registered:

```bash
sudo k3s kubectl describe node <new-node-name> | grep amd.com/gpu
```

## Troubleshooting

- **403 on token**: PAT needs `repo` scope or fine-grained "Administration" read/write permission
Expand All @@ -157,7 +183,7 @@ sudo KUBECONFIG=/etc/rancher/k3s/k3s.yaml helm upgrade arc-runner-set \

## Current Cluster Info

- **Node**: mia1-p02-g29 (+ 4 more nodes in the k3s cluster)
- **Nodes**: mia1-p02-g29, mia1-p02-g52, mia1-p02-g53, mia1-p02-g55, mia1-p02-g56 (5-node k3s cluster)
- **GPUs**: 8x AMD Instinct MI355X per node
- **CPU**: AMD EPYC 9575F 64-Core (128 threads, 2 sockets)
- **RAM**: ~3 TB per node
Expand Down
Loading