Currently the node definitions are constructed using ansible facts. At least in some situations this doesn't appear entirely satisfactory to slurm, e.g. slurmd -C shows ... Boards=1 ... and nodes are getting set DOWN.
This PR runs slurmd -C on all compute nodes, then uses values from the first-in-play in each partition (iaw existing logic) to provide node definitions.
This is sort of Trust On First Use that the node configuration is in fact correct.
An alternative is only to specify NodeName and not the expected CPU parameters at all:
Only the NodeName must be supplied in the configuration file. All other node configuration information is optional.
This would have 2x disadvantages:
Slurm cannot detect node misconfiguration
Scheduling is slower:
Establishing baseline configurations will also speed Slurm's scheduling process by permitting it to compare job requirements against these (relatively few) configuration parameters and possibly avoid having to check job requirements against every individual node's configuration. The resources checked at node registration time are: CPUs, RealMemory and TmpDisk.
Currently the node definitions are constructed using ansible facts. At least in some situations this doesn't appear entirely satisfactory to slurm, e.g.
slurmd -C
shows... Boards=1 ...
and nodes are getting set DOWN.This PR runs
slurmd -C
on all compute nodes, then uses values from the first-in-play in each partition (iaw existing logic) to provide node definitions.This is sort of Trust On First Use that the node configuration is in fact correct.
An alternative is only to specify NodeName and not the expected CPU parameters at all:
This would have 2x disadvantages:
Quotes from https://slurm.schedmd.com/slurm.conf.html.