overwhelm the one slurmd for the entire system).

 which shows some negative values.

will fail, but job steps are not properly supported on BGL anyway.

it orphaned).

Map all nodes in cluster to a single front-end node.
Don't repeat ping/register/kill/etc. RPCs to all pseudo nodes, just the
  front-end.
Treat single message for some RPCs as representing all nodes in the
  cluster: register, ping responce, epilog complete, etc.
Srun maps all tasks for a job step onto a single node request.

Map all nodes in cluster to a single front-end node.
Don't repeat ping/register/kill/etc. RPCs to all pseudo nodes, just the
  front-end.
Treat single message for some RPCs as representing all nodes in the
  cluster: register, ping responce, epilog complete, etc.

numbers.

(when the record of the job is completely gone, not ESLURM_ALREADY_DONE)

setting the reply filel descriptor to -1 (seen in Blue Gene/L testing).

Don't strip binaries to avoid breaking TotalView
Fix Munge error retry logic bug
Change srun interrupt handling to work with TotalView

Don't strip binaries to avoid breaking TotalView
Fix Munge error retry logic bug

Don't strip binaries to avoid breaking TotalView

results on AIX.

   when init_setproctitle() is called.

  - handle negative return code w/ errno == EINTR from sigwait()
  - don't create launch threads detached when running under parallel
    debugger

/usr/lib/slurm/jobcomp_script.so