Kai Zhang

Use arena to create a single node model development box

2

besides submit training jobs, user also wants to create a development box which contains jupyter, math lib, frameworks, in order to dev and debug algorithms before starting to train in...

It should be helpful for data scientists to use command like "arena create data imagenet-full" to create, index and manage different training datasets for different training jobs. Then when use...

add cpu, mem info in "arena top node" and "arena top job"

1

cpu and mem resource info is useful for both non-gpu and gpu jobs. these should be included in "arena top "

arena should submit jobs with support for customized label, toleration, securityContext in podSpec

1

more and more user cases are request specify customized label, toleration, securityContext, priorityClass, etc to job underlaying Pods. arena should give a unified mechanism to meet those customization requirement.

add event info of tfjob level in "arena get jobname -e "

1

"arena get jobname -e" only shows events of chief worker pod's event. while some meaningful events info of job level should also be shown. e.g. when ResourceQuotas enabled, if job...

add support for looking up command details of previous jobs

1

to support look up previous jobs command details is helpful to diagnose client problems or reference for re-run

'arena version' should also contain info of charts, operators installed in environment

1

arena version command should contain more info of which versions of charts, apiVersion of tfjob, mpijob etc, and corresponding operators release version deployed in cluster. it's helpful to quickly identify...