Slurm
Admin notes:
Nodes that go down will stay down until they come back to life with a valid configuration.
To get slurm using the node again, assume root and use:
scontrol update NodeName=huff6 State=RESUME
To deactivate a node, set its state to drain.
The GUI tool, sview, is a nice way to keep an eye on the cluster, but it's a little tricky to get running remotely.
The process goes like this:
ssh -X username@huff.hrpc.vcu.edu
once there, assume root:
sudo -i
then enable X forwarding for root:
xauth merge /lustre/home/username/.Xauthority
It may complain that no such file exists if you've never run a Xwindow program before.
You can run one now, or simply touch .Xauthority in your home dir (NOT as root) then relog in.
finally, run the program:
sview &
Most functions can be found in the left-click menu.
Adding new users to huff
If password hash was provided:
useradd -c 'Name, User' -p '$6$GciBq5oV$8s76Qfytu3lvgRdJaxQVGDEUqVmm83g9zTJ3KgKoDLEs41AIR/' -d /lustre/home/username username
If no password hash available:
useradd -c 'name' -d /lustre/home/user userpasswd !$chage -d 0 !$
NB: They will not be able to run slurm jobs until they're in the database.
To add them to the database:
sacctmgr add account !$sacctmgr add user !$ account=!$
# If GPU membersacctmgr modify user !$ set qos+=gpu
# to check:sacctmgr show assoc format=cluster,account,user,qos
Afterward, execute:
./propagate_users.sh
The new version will prompt you to press a key before it commits the changes.