slurm_cancel_job RPC now takes signal argument.

   prolog is run as root and failure is reported as a non-zero return code
   from the launch tasks or launch batch job messages
   epilog is run as result of revoke credential message

New RPC added for slurmd to send batch_sid to slurmctld.

New RPC defined to map local pid into slurm job id.

   response message

Also add exclude=node_list to allocate RPCs.

job time limit. Defined pack/unpack/free functions for RPC.

 o srun -d option updated to set slurmd-debuggging level
 o slurmd sends task launch resp message with list of pids
   (for totalview support)
 o Three Ctrl-C's in rapid succession to srun forcefully terminates the job.
 o Changed --kill-off to --no-kill

than single PID). Used for TotalView support.

call REQUEST_IMMEDIATE_RESOURCE_ALLOCATION. This immediate flag also
applies to the slurm_allocate_and_run_job_step() function.

Change pack32_array use 32 bit counter (rather than 16 bits).

initiation.
Establish limit of MAX_TASKS_PER_NODE and add matching error code.

support (RPC xfer and printing) of some missing control info.

Major clean-up of slurm_protocol_pack (not finished, but about half done).

   along with other general bugfixes and cleanup.

 o cbuf.[ch] list.[ch] : begin using Chris' list and cbuf routines
   from the LSD-Tools project.

 o srun/srun.c : *Always* wait for I/O to complete. This still needs work.

 o srun/launch.c : code cleanup

SlurmdLogFile, and SlurmdSpoolDir.

Add local_pid to task_launch_response RPC.

bytes in a loop (for speed).

bad message. Functions now return SLURM_ERROR on bad unpack instead
of assert failure.

(so shutdown command can shutdown both servers).

 o make dist now makes a proper distribution

 o rewrite of slurmd.c -- extensive changes therein
 o launch msg contains port numbers instead of slurm_addrs

   changed to err/in/out, fixes for inclusion of pthreads and SSL libraries
   etc.