The commands below effectively creates jobs with the same name:

run -g 0 -n aries-00 jet-dev jetd/basic-cuda12.1 -- bash -c 'sleep infinity'
run -g 0 -n aries-01 jet-dev jetd/basic-cuda12.1 -- bash -c 'sleep infinity'
run -g 0 -n aries-02 jet-dev jetd/basic-cuda12.1 -- bash -c 'sleep infinity'

When not specifying a node list, this will return [error] 14 container of same name already exists! instead (not sure if this behavior is consistent or depending on scheduling).

Another behavior is when you kill jet-dev, all jobs with the name would be killed without warning.

A skim into the client code suggests these behaviors are due to the ws server-side design. Can you clarify if these behaviors are expected?

You can now run a portfwd to a container which does not exist. This could cause some confusing error.

Should be easy adding a check at

Proposals:

1. Allow aries portfwd to run in the background (with file-logging, quiet, or verbose options).
2. (Even better) Make ariesd a system service so that portfwd (registered on a list) can check status and re-register themselves (even if you restart the client). This could allow port-forwarding requests to be passed with the run command and stay persistent.