flux-mini: need specific option for setting system attributes #4481
Comments
|
I like these modifications to the simplify the |
grondo
added a commit
to grondo/flux-core
that referenced
this issue
Aug 10, 2022
Problem: The flux-mini --setattr option is almost always used to set attributes in the system dictionary, and therefore its use can be very repetitive and inefficient. Prepend "system." to the KEY specified in --setattar=KEY=VAL when it doesn't already begin with "system." or "user.". To force a different top-level target value, also allow keys to be prefixed with "." (e.g. --setattr=.foo will set attributes.foo=1). Fixes flux-framework#4481
grondo
added a commit
to grondo/flux-core
that referenced
this issue
Aug 10, 2022
Problem: The flux-mini --setattr option is almost always used to set attributes in the system dictionary, and therefore its use can be very repetitive and inefficient. Prepend "system." to the KEY specified in --setattar=KEY=VAL when it doesn't already begin with "system." or "user.". To force a different top-level target value, also allow keys to be prefixed with "." (e.g. --setattr=.foo will set attributes.foo=1). Fix one use of --setattr=foo=val in the testsuite with --setattr=.foo=val. Fixes flux-framework#4481
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
The use of "system" attributes under
attributes.systemin jobspec may become more prevalent as we move forward. E.g. one use case is to set system jobspec attributes that optionally affect the behavior of the job prolog and epilog.However, the current method to set these attributes requires the use of a long option
--setattrand explicit designation of the attribute under thesystemdictionary, e.g.--setattr=system.prolog.clear-ssd. This could get tedious if a user is setting many of these attributes.We should add a new option that sets system attributes specifically, perhaps with a short option equivalent (similar to what these commands provide for
-o, --setopt=OPTION.Another idea might be to have the
system.prefix implied if the attribute doesn't already start withsystem.oruser.. I'm just not sure if that approach would get us into trouble later.The text was updated successfully, but these errors were encountered: