NVIDIA MPS not available under gpu_sharing_strategy field of google_container_cluster && google_container_node_pool resources

[whyevenquestion1t]

Community Note

• Please vote on this issue by adding a 👍 reaction to the original issue to help the community and maintainers prioritize this request.
• Please do not leave +1 or me too comments, they generate extra noise for issue followers and do not help prioritize the request.
• If you are interested in working on this issue or have submitted a pull request, please leave a comment.
• If an issue is assigned to a user, that user is claiming responsibility for the issue.
• Customers working with a Google Technical Account Manager or Customer Engineer can ask them to reach out internally to expedite investigation and resolution of this issue.

Description

https://cloud.google.com/kubernetes-engine/docs/how-to/nvidia-mps-gpus#deploy GCP has a Multi Processing Service feature that has a configuration of gpu-sharing-strategy=mps through the gcloud CLI. Unfortunately, that is the only way to deploy it. I am looking forward to being able to deploy a GKE cluster and configure a node pool with MPS through terraform.

New or Affected Resource(s)

• google_container_node_pool && google_container_cluster

Potential Terraform Configuration

gpu_sharing_config {
        gpu_sharing_strategy       = "MPS"
        max_shared_clients_per_gpu = var.max_shared_clients_per_gpu
      }

References

https://cloud.google.com/kubernetes-engine/docs/how-to/nvidia-mps-gpus#enable-node-pool

b/335463855

[melinath]

This is already supported as gpu_sharing_config, which takes arbitrary string values - the docs just need to be updated.

[whyevenquestion1t]

This is already supported as gpu_sharing_config, which takes arbitrary string values - the docs just need to be updated.

Thank you! Setting gpu_sharing_strategy = "mps" worked for me