Closed
Description
In the CPU container we create a user id and group id for user 1000 here
https://github.com/kubeflow/kubeflow/blob/master/components/k8s-model-server/images/Dockerfile.cpu#L53
We set that userid and groupid as part of the deployment here
https://github.com/kubeflow/kubeflow/blob/master/kubeflow/tf-serving/tf-serving.libsonnet#L143
We should probably do something similar for the GPU container. Otherwise I think users might have to override the securityContext in the deployment so as to avoid setting user id and group id.