bug: ray command not found following parallel pipeline tutorial

[brunnels]

Describe the bug

I'm following this tutorial: https://github.com/vllm-project/production-stack/blob/main/tutorials/15-basic-pipeline-parallel.md

kubectl -n ai logs vllm-stack-distilgpt2-raycluster-raycluster-head-lrxgm
Running vllm command in the background.
ERROR: init 250 result=11/bin/bash: line 1: ray: command not found

Regular non-parallel modelSpec's are working.

To Reproduce

I have kuberay-operator chart 1.2.0 deployed in the same namespace as vllm-stack chart 0.1.11. I'm using the same config file as the tutorial tutorials/assets/values-15-a-minimal-pipeline-parallel-example-raycluster.yaml.

I also tried the lastest kuberay-operator chart version 1.6.2 and got the same result.

Expected behavior

I should be able to verify the deployment per the tutorial

Additional context

No response

[cupkk]

I can take this. The tutorial uses vllm/vllm-openai, while the generated RayCluster requires the container image to provide the Ray Python package and CLI. I will update the example to a Ray-capable image, document that requirement, and validate the rendered chart.

[brunnels]

@cupkk Your proposed fix didn't resolve the issue for me.

$ k -n ai logs vllm-stack-distilgpt2-raycluster-raycluster-head-bm9mj
Running vllm command in the background.
ERROR: init 250 result=11/bin/bash: line 1: ray: command not found
$ k -n ai get pod vllm-stack-distilgpt2-raycluster-raycluster-head-bm9mj -o yaml | yq .spec.containers[0].image
lmcache/vllm-openai:latest

[brunnels]

@ruizhang0101 @cupkk can you please reopen the issue? This isn't actually resolved because the container image specified in the docs doesn't have ray.