GitLab

Found a bad default path for apbasil.  This is now something that should
work on a default Cray install.

This would prevent re-reservation if not fixed.

Rereservations were broken for long (>5 min) startups.  This should
allow the CAPMC scripts to do their thing.

The alps_script_forker has to call BASIL, too.  Make sure they all point
to the same one.

I've gotten a test job to run end to end and relase resources.  ALPS
reservation set confirmed released, node added to pool, run and
released.  Lots of cleanup is needed.  Code is at prototype stage. Not
production ready, but a lot closer to it.

ALPS needs some extra work from the forker.  This is a modified
user-script-forker for ALPS use.  It doesn't inherit from
user-script-forker due to some differences in the informaiton passed.

user_script_forker should not be run on ALPS systems.

A modified version of the user script forker is needed so that we can
confirm the ALPS reservation and set the pg_id from the child.  This
will let apruns from a user script run.

Adds feature requested for #4320.  This will signal the head pid rather
than the process group.  Not recommended for use with csh scripts.

- restructured bgsystem state preservation and restoration
- removed special mode for running diags on Blue Gene systems
- improved error handling in bgsystem's resource assignment code
- refactored bgsystem's update_partition_state.  includes a less aggressive
  partition cleanup algorithm to significantly reduce the probability of
  blocks getting stuck in the deallocating state.
- modified user_script_forker to kill the entire process group rather than
  just the process it forked

Fix for _get_exit_status to not kill booting block jobs (again) and
user_script_forker is now using killpg rather than just a kill.

This got missed in the initial resolution when merging the public and nda repos.

  script, replacing the shell process with the shell running the script.  the
  main benefit is that signals are delivered directly to the shell interpreting
  the script rather than the non-interactive login shell which might ignore
  SIGTERM.
- environment variables in the mpirun config setting are now expanded
- additional state version checking added to the base forker class

  to better tolerate GPFS flakiness
- modified forker status and cleanup interface to better tolerate XML/RPC
  connection drops and ssl timeouts
- updated bgsystem, cqm and the simulator to use the new forker interface (the
  cluster forker and system component still need updating)

  reservation on the partition has expired; as a side effect, a job now
  reserves the partition for the entire walltime just before attempting to fork
  the job
- modified base_forker to execute setpgrp/setsid in the child so that the
  entire process group could be killed
- fixed environment variable settings and propagation for script mode jobs

- added separate forkers for user and system scripts
- modified cqm to use system script forker
- moved mpirun specifics from bgsystem to bg_mpirun_forker
- removed a vast majority of the script job specific code from bgsystem
- a few other fixes here and there

* NOTE: only bgsystem has been updated.  the other system components,
  including the bg simulator, still need work.