job is in a pending state, then send the request directly to the slurmctld
    daemon and do not attempt to send the request to slurmd daemons, which are
    not running the job anyway.