-
Notifications
You must be signed in to change notification settings - Fork 117
Closed
Description
The idea is that we need to extend the feature that we already have for the flexible node allocation inside a reservation, to whole virtual partitions of the system. Moreover, the following combination need to be supported:
- If
--nodelistis passed, the test should be launched on the selected nodes. - If
--exclude-nodesis passed, the test should be launched on all the nodes of the virtual partition except the selected ones. - Respect maximum job size (e.g., 2400 nodes)
- By default, submit to the available nodes. If
--nodelistis passed, then these nodes must be used. - What if both
--nodelistand--exclude-nodesare passed and the sets intersect?
From the design's point of view, this functionality (for reservations) is part of the Slurm backend. I think it would be good if we extend the Job API so as the backends automatically figure out the number of tasks for the test. Again, implementations for this feature, we will only provide for the Slurm backend.