Basasuya
Basasuya
@retrry can you run ```kubectl -n ray-system describe clusterrole kuberay-operator```? the correct condition you will see ``` ... rayjobs.ray.io [] [] [create delete get list patch update watch] ``` the...
OK,i think the problem is deployment is not updated in helmchart
@Jeffwan cc
@asm582 I think start time is when these job custom resource is created. Because the job crd supports cluster is already running or create cluster through `clusterSelector` (reference https://github.com/ray-project/kuberay/pull/470) so...
@simon-mo same question with [atinsood](https://github.com/atinsood), what's the difference with ray job stop and stop_job api in Job Submission SDK ？
@atinsood 1 ray job == 1 RayJob CR, if clusterSelector is used, lifecycle of RayJob CR is independent of lifecycle of the cluster. if not, the cluster would be deleted...
My preliminary conclusion is that using klog to write to file and stdout. I will research on the writing file feature of klog
After testing, I found Dockerfile is using gcr.io/distroless/static:nonroot as based image, which can not exec into the container. Writing to file is meaningless because I can not see them. So...
manual testing through replace the based image in dockerfile. the previous image can not exec into it.
@kevin85421 sorry for the slow reply, I have rebase the master again.