chore: remove ConfigMap from ray-job.kueue-toy-sample.yaml #1976
Conversation
|
cc @andrewsykim |
| # The alternative value is "HTTPMode", indicating that KubeRay will submit the Ray job by sending an HTTP request to the RayCluster. | ||
| # submissionMode: "K8sJobMode" | ||
| entrypoint: python /home/ray/samples/sample_code.py | ||
| entrypoint: python -c "import time; [print(i) or time.sleep(1) for i in range(600)]" |
There was a problem hiding this comment.
I like the example used here: https://docs.ray.io/en/master/cluster/kubernetes/getting-started/raycluster-quick-start.html#method-1-execute-a-ray-job-in-the-head-pod
Can this be
entrypoint: python -c "import ray; ray.init(); print(ray.cluster_resources())"
I think an example that uses ray library will be better, even if it's very simple
There was a problem hiding this comment.
The ray.cluster_resources() example is not suitable for showcasing gang scheduling. In this doc, we create two RayJob custom resources and ensure that no Pods belonging to the second RayJob will be created because the ClusterQueue doesn't have enough resources. If the RayJob finishes very quickly, it is possible that users may not be able to observe this behavior.
There was a problem hiding this comment.
Yes, that makes sense. This was more of a general comment that we don't need ConfigMap for most examples. ray-job.sample.yaml for example can probably inline the example right?
There was a problem hiding this comment.
The ray.cluster_resources() example is not suitable for showcasing gang scheduling.
Actually, even for this example, can we run ray.cluster_resources() in between each sleep?
|
LGTM, my other comments are not blocking |
Why are these changes needed?
#1956 (comment)
This PR removes ConfigMap from the YAML. In this doc, we need to run
kubectl create -f ray-job.kueue-toy-sample.yamlmultiple times. If the YAML includes a ConfigMap, users may see a warning message indicating that the ConfigMap already exists. Removing the ConfigMap improves the UX.Related issue number
Checks