slurm.conf man page

BLUEGENE - fixed race condition with preemption where if the wind blows the right way the slurmctld could lock up when preempting jobs to run others.

This is needed by select/bluegene to synchronize block state
for completing jobs.

with respect to BlueGene systems. Formerly logged bogus inconsistencies.

Fixed issue where QOS priority wasn't re-normalized until a slurmctld restart when a QOS priority was changed.

Made it so QOS with UsageFactor set to 0 would make it so jobs running under that QOS wouldn't add time to fairshare or association/qos limits.

    job. Formerly would display information about one job, but update next
    selected job.

    until the job is released. Patch from Rod Schultz, Bull.

    attempts from 60 seconds to 29 seconds. This should eliminate a possible
    synchronization problem with gang scheduling that could result in job
    step creation requests only occuring when a job is suspended.

BLUEGENE - better checking small blocks in dynamic mode whether a full midplane job could run or not.

    running to be more clear and only print when --verbose option is used.

it's partition is configured "Shared=EXCLUSIVE" (which is redundant).

    after a job completed.

    size get queued).

    specific reservation.

salloc: notify terminal foreground process

This fixes another bug observed in salloc child process cleanup. I found
that some shells, e.g. zsh, do not forward all signals to their children.

The patch fixes the problem that
 * command_pid is still active but does not equal tpgid,
 * tpgid is not the same as salloc's process group,
 * tpgid is very unlikely to come from another process, since we block
   the suspend/TSTP signal,
 * signalling command_pid does not automatically imply that the active
   terminal foreground process is also signalled,
 * hence send a HUP to signify "death of controlling process".

This setup fixed the problem on zsh. I then went and tested a more complex setup:

Before:
-------
palu2:0 ~>ps  f -o pid,pgid,tpgid,ppid,stat,tty,cmd
  PID  PGID TPGID  PPID STAT TT       CMD
21117 21117 21597 21116 Ss   pts/9    -bash
21260 21260 21597 21117 Sl   pts/9     \_ ./slurm_build/git/src/salloc/salloc -v --time=00:01:00 -N17 zsh
21266 21266 21597 21260 S    pts/9         \_ zsh
21323 21323 21597 21266 S    pts/9             \_ /bin/bash
21397 21397 21597 21323 S    pts/9                 \_ -bin/tcsh
21526 21526 21597 21397 S    pts/9                     \_ /bin/sh
21597 21597 21597 21526 S+   pts/9                         \_ aprun -N1 -n17 sleep 12345
21601 21597 21597 21597 S+   pts/9                             \_ aprun -N1 -n17 sleep 12345

After the timeout:
------------------
palu2:0 ~>ps  f -o pid,pgid,tpgid,ppid,stat,tty,cmd
  PID  PGID TPGID  PPID STAT TT       CMD
21323 21323 21117     1 S    pts/9    /bin/bash
21397 21397 21117 21323 S    pts/9     \_ -bin/tcsh
21526 21526 21117 21397 S    pts/9         \_ /bin/sh

==> The 'dangerous' aprun terminal foreground process group 21597 has been removed, while the child
    subprocess groups 21323, 21397, and 21526 now exist as orph01_salloc-Bug-Fix-nested-terminal-foreground-process.diff
aned groups, to be cleaned up by init.

I noticed that the S_{JOB,STEP}_ALLOC_* comments in spank.h don't
document the return-by-value type as is done for the other spank_item
enumerations.