You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Marty has been adding some --requires=hosts constraints to his jobspecs to force flux to pick compute nodes associated with specific rabbits. However, it would be more ergonomic to simply specify the rabbits desired, since that's a much more compact representation. Something like --requires=hosts:rabbit203 to cause the scheduler to only choose compute nodes on the same rack as rabbit203.
The text was updated successfully, but these errors were encountered:
I'm more just thinking out loud here, but --requires=hosts:rabbit203 would only work (I think) if constraining the resource graph to only vertexes that match a "hostname" of rabbit203 would result in something that could match the rest of the job request. (Given that I have no idea how the datawarp/rabbit matches work in Fluxion, I could be way off here)
An alternative would be to establish a new requires scheme rabbits which, when used (e.g. as --requires=rabbbits:rabbit203), would activate a custom rabbit constraint that would constrain the graph to only the resources with access or that would use the specified rabbits. A drawback to this is that there is no way that I know of to extend the constraints supported by Fluxion at the moment.
I just thought I'd bring these points up in case they were helpful. If the hosts solution works then that might be preferable.
Marty has been adding some
--requires=hosts
constraints to his jobspecs to force flux to pick compute nodes associated with specific rabbits. However, it would be more ergonomic to simply specify the rabbits desired, since that's a much more compact representation. Something like--requires=hosts:rabbit203
to cause the scheduler to only choose compute nodes on the same rack asrabbit203
.The text was updated successfully, but these errors were encountered: