This is work towards full select (although we use "nodes") support as present in PBSPro.

Torque server originally did not support counted resources with the exception of processes/processors (ppn).

This feature adds support for counting any known resource (see src/server/resc_def_all.c for complete list) per node.

Resources can be manually set on nodes using qmgr:

set node node1 resources_total.vmem = 40G

Set resources can be the requested on job submit:

qsub -l nodes=1:vmem=20G

Server counts each resource and won't allow job runs that go over the total limit.

Because nodes already report certain resources in their status, server also supports reading resources from this reports.

Two server attributes are present resources_to_store and resources_mappings.

For example this node reports several resources:

status = ...,idletime=23801,totmem=7638716kb,availmem=6704332kb,physmem=3445760kb,ncpus=2,loadave=0.00,gres=,netload=4542446503,state=free,...

Let's say we don't want to manually set memory on each node, but we want to have total node memory stored in the vmem resource (not totmem as reported by the node).

Using qmgr:

set server resources_to_store=totmem
set server resources_mappings=totmem=vmem

Each node will now have a resource vmem that will be read from the nodes status report.

We can still override this value by manually setting it.