[RayCluster] Adds aliyun.com/gpu-mem to known custom accelerators

[win5923]

Why are these changes needed?

if use aliyun k8s gpu share, gpu key is aliyun.com/gpu-mem

    workerGroupSpecs:
            resources:
              limits:
                aliyun.com/gpu-mem: "1"
                cpu: "1"
                memory: 2Gi
              requests:
                aliyun.com/gpu-mem: "1"
                cpu: "1"
                memory: 2Gi

autoscaler will not work when request gpu resource

(autoscaler +3m13s) Error: No available node types can fulfill resource request {'GPU': 1.0, 'CPU': 1.0}. Add suitable node types to this cluster to resolve this issue.

Adds aliyun's gpu share to the list of known custom accelerators in pod.go

Related issue number

Closes #2484

Checks

• I've made sure the tests are passing.
• Testing Strategy
  • Unit tests
  • Manual tests
  • This PR is not tested :(

[andrewsykim]

Is this the actual resource name recognized by Ray? aliyun.com/gpu-mem -> gpu_share seems like an odd mapping

[win5923]

Apologies, I realized that Ray Resource Name only supports GPU, neuron_cores, TPU, NPU, and HPU. I will mapping aliyun's gpu share to GPU.

Ref: https://docs.ray.io/en/latest/ray-core/scheduling/accelerators.html

[andrewsykim]

yes, I think we should only add custom accelerators here that are in the supported list you referenced

[andrewsykim]

ref: #2631 (comment)

[win5923]

I mapped aliyun.com/gpu-mem to num-gpus. Because gpu share also belongs to the GPU.

[andrewsykim]

Can you first open a PR to update the officially supported accelerators in https://docs.ray.io/en/latest/ray-core/scheduling/accelerators.html to include alyun.com/gpu?

[win5923]

I checked Aliyun's documentation, and I think they also use Nvidia GPU with field called nvidia.com/gpu, while Aliyun GPU Share has another field called aliyun.com/gpu-mem.

Ref: https://www.alibabacloud.com/help/en/eci/user-guide/create-a-pod-by-specifying-the-gpu-specification-1

Can you first open a PR to update the officially supported accelerators in https://docs.ray.io/en/latest/ray-core/scheduling/accelerators.html to include alyun.com/gpu?

So I think we should just focus on the string?