[kubectl-plugin] add --worker-gpu flag for cluster creation #2675
Conversation
Signed-off-by: Andrew Sy Kim <[email protected]>
|
@chiayi PTAL |
| @@ -85,6 +89,7 @@ func TestConvertRayClusterApplyConfigToYaml(t *testing.T) { | |||
| WorkerReplicas: 3, | |||
| WorkerCPU: "2", | |||
| WorkerMemory: "10Gi", | |||
| WorkerGPU: "0", | |||
There was a problem hiding this comment.
Do you think it is worth it to check the yaml output when the GPU is non-zero? Or is it enough to see that the value is properly set in the applyconfig?
There was a problem hiding this comment.
Good point, will add a test for when GPU is set to 1
There was a problem hiding this comment.
ah nvm, I already updated the test TestGenerateRayCluterApplyConfig to set GPU to 1
There was a problem hiding this comment.
Oh oops my bad, missed that.
| @@ -98,6 +99,18 @@ func (rayClusterSpecObject *RayClusterSpecObject) generateRayClusterSpec() *rayv | |||
| corev1.ResourceMemory: resource.MustParse(rayClusterSpecObject.WorkerMemory), | |||
| })))))) | |||
|
|
|||
| gpuResource := resource.MustParse(rayClusterSpecObject.WorkerGPU) | |||
| if !gpuResource.IsZero() { | |||
There was a problem hiding this comment.
Since we are adding the GPU option, should we add the option to specify the gke-accelerator?
I do think this could also be done on a later date though.
There was a problem hiding this comment.
We probably want to generalize it further and support an option to specify node selectors
Why are these changes needed?
Add the option to specify worker GPU count. For now we default to nvidia.com/gpu but we can support other GPU types in the future.
Related issue number
#2608
Checks