Identify proper slurmd address for communications rather than using
  "localhost", which could bind to an address the daemon does not use.

srun doesn't get any notification through TCP that the 
connection was lost, and the slurmctld sees the down node
and ends the job step.

    based upon NodeAddr, ControllerAddr and BackupAddr configuration 
    parameters. Patch from Matthieu Hautreux, CEA.

added support for classification in a cluster.  This doesn't work with sacctmgr yet.  it should be easy to add

    for better fault tolerance (e.g. killed and restarteed SLURM daemons on 
    compute node).

callers to better handle and report errors from this function. Also
update all callers to use new calling style. Update call site in
plugstack.c to report error using plugin_strerror(). Other callers
ignore errors as before.

 * Fix other gratuitous whitespace use that didn't conform to SLURM style

spank_err_t type.

Certain SPANK items that are available in local or allocator context
may not be available until after cmdline options have been processed,
or after a job has been allocated. Return ESPANK_NOT_AVAIL to indicate
to plugins that this is the case.

This patch fixes a potential segfault if SPANK plugins call spank_get_item
for these items from early callbacks such as spank_init.

    nodes are configured in multiple leaf switches.

fix for reservations to create new entries in the database when something about nodes or cpus or flags change

  possibly a configuration error.

    launch tasks).

notify if the user or node_prefix has changed things aren't going to work correctly in the bluegene plugin