Since this changes the data structure and RPC, I made the change in
the code base for version 14.11 rather than 14.03, but did not see any
significant issues with your patch.
I did move some things around in the header file to keep fields in
alphabetic order, which makes it easier to find values using a debugger.
There isn't a need to copy the node names before printing them unless
you want to report allocated I/O nodes, which will not be available in
this case (see src/squeue/print.c)
I also cleared this field once the job is actually allocated resources
and starts (src/slurmctld/node_scheduler.c)
But these are really minor issues.
The final commit is here:
https://github.com/SchedMD/slurm/commit/0c3aea36278fe510d00be3114b35abbcaf3c9a76
Thanks!
Quoting Martins Innus <[email protected]>:
Hello,
I am working on trying to get better information on how jobs are
being scheduled by the backfill algorithm. And have started with the
attached patch to expose information to squeue that is already
calculated by the scheduler. Since it breaks the ABI for messages
from the controller, I don't expect that this would be merged in the
nearterm, but is there any interest at some point to have this
available? Also, I would appreciate some feedback on whether I'm
doing the memory management correctly. It wasn't clear to me
exactly where memory should be freed.
Thanks
Martins