gpu-manger are always scheduled to master node

[ryanqin01]

What happened:
I use two GPU machines to deploy a k8s cluster using TKE. One as master node and one as worker node. When I try to install gpu manger plugin, it will always fail. Since the gpu-quota-admission pod is always scheduled to master node. I have to manually change gpu-manger.yaml and force the gpu-quota-admission pod to be deployed in worker node.

What you expected to happen:

How to reproduce it (as minimally and precisely as possible):

Anything else we need to know?:

Environment:

• TKE version: 1.2.4
• Global or business cluster:
• Kubernetes version (use kubectl version):
• Install addons:
• Others:

[QianChenglong]

Seems like gpu-quota-admision need copy config to /etc/kubernetes.

@mYmNeo Maybe explain why？

[mYmNeo]

The affinity spec tell the scheduler to do so. https://github.com/tkestack/tke/blob/v1.2.4/pkg/platform/controller/addon/gpumanager/template/template.go#L165