Fail to override the kube-scheduler.

Hi,

I copied the kube-scheduler file of yours to `/etc/kubernetes/manifests/kube-scheduler.yaml. The container initialized successfully, but the container log seems that it cannot request resources as follows:

Is kube-scheduler of v1.11.2 (in your image) not compatible with my k8s cluster (v1.17) or any other problems in my environment?

Thanks!

Hello,

I also seen the same problem with my k8s cluster (v1.16.6-beta.0). I followed the installation instruction, which instruct to modify the kube-scheduler.yaml and use the image registry.cn-hangzhou.aliyuncs.com/acs/kube-scheduler-amd64:v1.11.2

Is it possible to obtain an image that for a latest version of K8 an not v1.11.2? Or can this image be build from the github repo?

Thanks kubectl logs kube-scheduler-sta2node01.sdp.research.bell-labs.com -n kube-system I0727 14:27:57.506380 1 server.go:126] Version: v1.11.2 W0727 14:27:57.509174 1 authorization.go:47] Authorization is disabled W0727 14:27:57.509196 1 authentication.go:55] Authentication is disabled I0727 14:27:57.509207 1 insecure_serving.go:47] Serving healthz insecurely on 127.0.0.1:10251 E0727 14:27:57.586187 1 reflector.go:205] k8s.io/kubernetes/vendor/k8s.io/client-go/informers/factory.go:130: Failed to list *v1beta1.ReplicaSet: the server could not find the requested resource E0727 14:27:57.588845 1 reflector.go:205] k8s.io/kubernetes/vendor/k8s.io/client-go/informers/factory.go:130: Failed to list *v1beta1.StatefulSet: the server could not find the requested resource E0727 14:27:58.588623 1 reflector.go:205] k8s.io/kubernetes/vendor/k8s.io/client-go/informers/factory.go:130: Failed to list *v1beta1.ReplicaSet: the server could not find the requested resource E0727 14:27:58.590786 1 reflector.go:205] k8s.io/kubernetes/vendor/k8s.io/client-go/informers/factory.go:130: Failed to list *v1beta1.StatefulSet: the server could not find the requested resource E0727 14:27:59.591626 1 reflector.go:205] k8s.io/kubernetes/vendor/k8s.io/client-go/informers/factory.go:130: Failed to list *v1beta1.ReplicaSet: the server could not find the requested resource E0727 14:27:59.593165 1 reflector.go:205] k8s.io/kubernetes/vendor/k8s.io/client-go/informers/factory.go:130: Failed to list *v1beta1.StatefulSet: the server could not find the requested resource

I was able to resolve the problem by using my own image in the kube-scheduler.yaml file instead of using the one in the documentation which is for an older version of K8. The image correspond to the version of K8 that I was using v.16

I also had to modify the device-plugin-ds.yaml in order to use api corresponding to K8 version > 1.16

In line 1 replace "apiVersion: extensions/v1beta1" with "apiVersion: apps/v1" and now deploy the deamon set by executing (make sure to execute step 4 below prior to starting the deamon set)

After making these changes, I was able to see how different containers were bin packed into a single GPU based on the gpu-memory resource limit.

use the resource limit gpu-count seems to load the requested number of gpus into a container, however, if a second container also request the gpus, they seem to be shared as well. Not what I expected.