How should I apply this plugin without docker layer but just use the containerd ?

[metaimagine]

Seem the Kubernetes will remove the docker after the version 1.20. How to migrate the gpushare in the new contain runtime?

[chaiyd]

me too

[wsxiaozhang]

the gpushare device plugin and scheduler extender themselves don't rely on docker or containerd. what you should do is just update your containerd config to use nvidia-container-runtime, pls refer to https://developer.aliyun.com/article/782035 for more details. as long as you can assign nvidia.com/gpu to your pod with your containerd enabled k8s，you can use gpushare there as well. BTW, the containerd compatibility has been resolved by default in Alibaba cloud K8s service (aka. ACK).