Release Slurm-GCP 6.5.6

[tpdownes]

6.5.6

• Fix bug where dynamic nodes with hostname set to FQDN could not be found by slurmsync script; in current architecture, this impacts login nodes. The resulting Exception was uncaught; Solution truncates all nodenames to short hostname before performing comparisons requiring short hostname

Extends work performed in e33c07e.

Observed results

✅ Test 1: confirm nodeset startup-script is working by activating a service at boot

srun -N64 systemctl is-active nvidia-dcgm.service
active
active
...

✅ Test 2: confirm exception no longer appears in slurmsync logs and the login node appears as idle to sudo sinfo (x-login partition is "root only")

$ sudo cat /var/log/slurm/slurmsync.log
2024-05-29 15:32:45,153 INFO: Restarting slurmctld to make changes take effect.
$ sudo sinfo
PARTITION AVAIL  TIMELIMIT  NODES  STATE NODELIST
x-login      up   infinite      1   idle a3mega-login-001.c.hpc-toolkit-gsc.internal
a3mega*      up   infinite     64   idle a3mega-a3meganodeset-[0-63]
debug        up   infinite      4  idle~ a3mega-debugnodeset-[0-3]

[tpdownes]

Manually deploying examples/hpc-slurm.yaml from the HPC Toolkit shows no errors/exceptions in the sync log:

# cat slurmsync.log 
2024-05-29 16:49:03,412 INFO: Restarting slurmctld to make changes take effect.