Get number of cpus on allocated job via environment variable

In slurm we can pass an environment variable to a job to get the number of cpus allocated to the job.


Count of CPUs available to the job on the nodes in the allocation, using the format CPU_count [(x number_of_nodes )][, CPU_count [(x number_of_nodes )] …]. For example: SLURM_JOB_CPUS_PER_NODE=‘72(x2),36’ indicates that on the first and second nodes (as listed by SLURM_JOB_NODELIST) the allocation has 72 CPUs, while the third node has 36 CPUs. NOTE : The select/linear plugin allocates entire nodes to jobs, so the value indicates the total count of CPUs on allocated nodes. The select/cons_res and select/cons_tres plugins allocate individual CPUs to jobs, so this number indicates the number of CPUs allocated to the job.

Is there a similar env in pbspro ? Thanks

Do you need only an environment variable? because there are alternative ways.
At python I use for example:
import psutil
number_of_cpus = len(psutil.Process().cpu_affinity())

Note you can also watch which PCU ID is assigned.
I think there is no such PBS environment variable.

You can use OMP_NUM_THREADS, which is pretty standard for multithreading applications.

thanks for your replies. I also found out ${N_CPUS}, thus setting to export OMP_NUM_THREADS=${NCPUS}. I’m curious about how is the standard to way to set memory limitation as well ? thanks


you can use a hook to modify the job’s environment at submission time.
This is what I used to set some values as memory and cpus:

import pbs
E = pbs.event()
J = E.job
R = J.Resource_List
V = J.Variable_List
if R["select"] != None:
    sel = repr(R["select"])
    ncpus = 0
    mem = pbs.size(0)
    chunks = sel.split("+")
    for chunk in chunks:
        mult = 1
        for c in chunk.split(":"):
            kv = c.split("=")
            if len(kv) == 1:
                mult = int(kv[0])
            elif len(kv) == 2:
                if kv[0] == "ncpus":
                    ncpus += mult * int(kv[1])
                elif kv[0] == "mem":
                    size = pbs.size_to_kbytes(pbs.size(kv[1]))*1024
                    mem += pbs.size(mult*int(size))

if ncpus == 0:
    if R["ncpus"] != None:
        ncpus = int(R["ncpus"])
if pbs.size_to_kbytes(mem)*1024 == 0:
    if R["mem"] != None:
        m = pbs.size(R["mem"])
        mkb = pbs.size_to_kbytes(m)*1024
        mem = pbs.size(mkb)

V["PBS_MEM"] = mem
V["PBS_NPROCS"] = ncpus

V is the list of variables exported to the job’s env.

thus setting to export OMP_NUM_THREADS=${NCPUS} .

Hmm, OMP_NUM_THREADS should already be set by PBS. At least in my case, it is.