Forward step failure reason back to slurmd before in some cases it would just be SLURM_FAILURE returned.

and change in sbatch log message

  reduce iteration counts when running in memory leak test mode

    to set default job parameters or perform other site-configurable actions at
    job submit time.
svn merge -r20006:20019 https://eris.llnl.gov/svn/slurm/branches/slurm-2.2.job_template

logic to handle resize of jobs for steps and suspended jobs.  suspended job logic hasn't been tested

    addition to the currently accepted "minutes", "hours", etc.

default partition was not idle. new logic tests every node in the 
default partition for an idle one.

  and use this to rebuilt the node bitmap on slurmctld econfigure or
  restart. This is needed to insure consistency with the other bitmaps
  (e.g. cpu count per allocated node).

ok, it appears things are handled correctly for jobs when they resize, it is still needed to add steps resizing