RHOAI 2.22 GPU as a Service Lab overlay #152
Conversation
| @@ -0,0 +1,11 @@ | |||
| apiVersion: kustomize.config.k8s.io/v1alpha1 | |||
There was a problem hiding this comment.
This may be a duplicate file?
| @@ -0,0 +1,40 @@ | |||
| --- | |||
| apiVersion: batch/v1 | |||
| kind: Job | |||
There was a problem hiding this comment.
Is this different from the aws-gpu-machineset component?
It looks like it is deploying the same GPU instance type and my quick check doesn't seem to deploy something different.
If it is different I would prefer to move this portion into its own component.
The main reason for that is that the aws-gpu job is really a work around for us to get GPUs in the demo environment and not something that we would generally use in a customer environment. We want to easily remove the job that creates the GPU machinesets but we may still want to use time-slicing.
| @@ -0,0 +1,25 @@ | |||
| # components-distributed-compute | |||
There was a problem hiding this comment.
Don't forget to update the readme
| components: | ||
| codeflare: | ||
| managementState: Managed | ||
| kueue: |
There was a problem hiding this comment.
I'm wondering if it would make more sense to call this something besides kueue-operator since this is kind of the opposite.
I wouldn't mind updating the current component-distributed-compute to disable the kueue operator and including the Red Hat build of Kueue in the AI Accelerator by default since that will be the preferred method moving forward.
I don't think we have any sort of backwards compatibility concerns since keueu/distributed compute was not really used in any of the existing examples.
Thoughts?
2 participants
What does this PR do?
Add a new overlay for the GPU as a Service Lab, using RHOAI 2.22 and OCP 4.19
Test Plan
Deploy a new demo cluster --> execute bootstrap --> select rhoai-stable-2.22-aws-gpu-time-sliced bootstrap.