dev: real-GPU kind cluster with DRA passthrough - #179
Closed
ShubyM wants to merge 1 commit into
Closed
Conversation
One command turns a Linux box with NVIDIA GPUs into a single-node cluster whose GPUs are real, allocatable DRA devices -- the same gpu.nvidia.com DeviceClass and ResourceSlices a production DRA cluster publishes. hack/kind/kind-up.sh is idempotent with preflight checks and offered fixes (nvidia runtime as docker default, volume-mount device injection); bootstrap.sh provisions a fresh Ubuntu VM from nothing. Self-contained: no manifests, no workloads, no opinions about what runs on it. Anything that consumes DRA can be developed against it.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
One command turns a Linux box with NVIDIA GPUs into a single-node Kubernetes cluster whose GPUs are real, allocatable DRA devices — the same
gpu.nvidia.comDeviceClass and ResourceSlices a production DRA cluster publishes, with no cloud cluster involved.hack/kind/kind-up.sh— idempotent: creates theopenrl-gpukind cluster (k8s 1.34, DRA GA) with host GPUs passed through via nvidia-container-toolkit volume-mount device injection, installs the NVIDIA DRA driver, waits for the DeviceClass. Preflight checks with offered fixes.hack/kind/bootstrap.sh— provisions a fresh Ubuntu VM from nothing (docker, kind, kubectl, helm, nvidia-container-toolkit, uv).docs/setup/kind-gpu-cluster.md— how the passthrough works and the fidelity caveats.Deliberately self-contained: no manifests, no workloads, no opinions about what runs on it. Anything that consumes DRA (#178's scheduler, the FFT stack, plain ResourceClaim experiments) can be developed against it.
Verified on a g2-standard-24 (2× L4): cluster up, both GPUs visible in the node, DeviceClass registered, claims allocated by the real driver.
🤖 Generated with Claude Code