Merge pull request #7235 from nojnhuh/azure-gpu-label

azure: use new label to identity GPU Nodes
kubernetes · Sep 4, 2024 · 8e11684 · 8e11684
2 parents 2c08685 + 9ae428a
commit 8e11684
Show file tree

Hide file tree

Showing 2 changed files with 2 additions and 2 deletions.
diff --git a/cluster-autoscaler/cloudprovider/azure/README.md b/cluster-autoscaler/cloudprovider/azure/README.md
@@ -54,7 +54,7 @@ k8s.io_cluster-autoscaler_node-template_resources_cpu: 3800m
 k8s.io_cluster-autoscaler_node-template_resources_memory: 11Gi
 ```
 
-> **_NOTE_**: GPU autoscaling on VMSS is informed by the presence of BOTH the `accelerator` AND `kubernetes.azure.com/accelerator` Node labels. A VMSS with GPUs whose Nodes do not have BOTH labels may not be scaled correctly. In a future release of cluster-autoscaler, the `accelerator` label will no longer be used and only the `kubernetes.azure.com/accelerator` label will be required.
+> **_NOTE_**: GPU autoscaling on VMSS is informed by the presence of the `kubernetes.azure.com/accelerator` Node label. A VMSS with GPUs whose Nodes do not have the label may not be scaled correctly. The `accelerator` label was used for this purpose in versions 1.31 and older.
 
 #### Autoscaling options
 

diff --git a/cluster-autoscaler/cloudprovider/azure/azure_cloud_provider.go b/cluster-autoscaler/cloudprovider/azure/azure_cloud_provider.go
@@ -75,7 +75,7 @@ func (azure *AzureCloudProvider) Name() string {
 
 // GPULabel returns the label added to nodes with GPU resource.
 func (azure *AzureCloudProvider) GPULabel() string {
-	return legacyGPULabel // Use legacy to avoid breaking, for now
+	return GPULabel
 }
 
 // GetAvailableGPUTypes return all available GPU types cloud provider supports