when I start an interactive job is activated in one of the worker nodes - I see it in the logs on the terminal and running squeue, but then when I run my commands in the terminal they are using the RAM/CPUs of the login node. I checked it using both htop and glances.
salloc -c 2 -t 10:00:00 --mem=4G
What should I check in the slurm configuration file? I expect that should not happen this.
The "login" node is not in the configuration file.
### Nodes
NodeName=node1 CPUs=80 Boards=1 SocketsPerBoard=2 CoresPerSocket=20 ThreadsPerCore=2 RealMemory=480000
NodeName=node2 CPUs=80 Boards=1 SocketsPerBoard=2 CoresPerSocket=20 ThreadsPerCore=2 RealMemory=480000
### Partitions
PartitionName=standard Nodes=node1,node2 Default=Yes MaxCPUsPerNode=76 MaxTime=INFINITE State=UP
When I run with sbatch everything go well.
The salloc command will or will not connect to the (first) node in the allocation depending on the value of the LauchParameters option in slurm.conf's use_interactive_step.
If set, it will
Have salloc use the Interactive Step to launch a shell on an allocated compute node rather than locally to wherever salloc was invoked.
Otherwise you have the behaviour you observe