---
2.3.1:
  - added 'stage' option to submit mode, which allows submission in a 'holding'
    state.  thanks to Juan.Fernandez@eu.Takata.com for this fix!
2.3.0:
  - added ability to set state=holding
  - added holding to statuslister
  - ENV['RQ_JOB'] now has yaml dump of entire job when run
  - some minor mods to updater/deleter
2.2.0:
  - added start, stop, shutdown, and feeder modes - thanks to mikael borg for
    suggestion
  - improved query usage
  - added credits section to help
2.1.0:
  - enhanced status reporting
  - fixed bug where rotating failed due to prohibition of deletion of running
    jobs added in 2.0.0.  the fixed basically adds a 'force' flag which is used
    by rotator.rb
  - made rotate a bit smarter: if a directory is given the rotation is
    automatically named and placed there.  if a pathname is given the rotation
    is placed there, and if no pathname is given one is automatically generated.
  - made pidfile generation even more robust : a pidfile is only considered o.k.
    when the lock has been obtained AND the the pid (if any) found in the
    pidfile is not running.  if a non-existent pid is found in the pidfile AND
    the lock could not be obtained it is assumed the nfslocking setup on that
    host is fubar and an auto-recovery takes place
  - fixed small bug which allowed the runner field of non-pending jobs to be
    updated
2.0.0:
  - deletion of running jobs is now prohibited (silently ignored actually).  rq
    has no central process and no communication whatsoever is done between
    nodes.  because of this it is not possible for delettion of a running job
    from the queue to somehow be signaled to the possibly remote node.  however,
    it is NEVER an error to kill a job on some host - rq will simply report the
    failed exit_status - thus one can easily set up a simple command, perhaps
    using ssh keys, to kill a remote job and THEN delete it from the queue if
    this behaviour is truely needed.  in simple terms a job may be removed from
    the queue BEFORE it starts running or AFTER is finishes - no matter what the
    exit_status.
  - when submitting a runner(s) can be specified.  feeders respect this
    'request' and only feeders whose hostname matches the list will run the job.
    eg.  jobs may be submitted to a specific host.
  - submitted jobs can now marked restartable (--restartable).  feeders respect
    this and restart any jobs found dead when starting - this means a node
    running a job can be rebooted mid-job and the job will restart when the
    feeder starts again.
  - fixed bug with pidfile generation on machines sharing the same home
    directory via NFS - big thanks to David Desmarest.
  - status reporting has been augmented and now includes some rough metrics for
    the cluster
1.1.0:
  - some refactoring on all the MainHelpers
  - added rotate mode
1.0.0:
  - bug fixed where 42 unix socket files in /tmp were created for every
    jobrunnerdaemon but only one was removed at_exit.  this probably isn't a big
    deal since they are all zero sized and thus will not fill up /tmp, but it's
    ugly nonetheless.  the fix doesn't affect rq's functionality at all.
  - added (gasp) a small bit of api doccumention - need much more work but it's
    a start
0.1.7:
  - massive code re-org into libs so that each main program delegator and many
    supporting classes are in their own own class/file
  - fixed bug where, under certain conditions, lock refreshers might not be
    killed leading to many refresher process being started on a machine
  - added 'tag' field to db and command line tool - useful to mark jobs for easy
    query later
0.1.6:
  - added JobRunnerDaemon drb object to avoid forking in sqlite transactions 
  - added update mode 
  - added backup mode 
  - added lock mode 
  - added query mode 
  - wrote meta-installer script for rq_all.tgz 
0.1.5:
  - fixed nfs sillyname bugette 
  - improvements made to lockd recovery mode
  - fixed submit mode priority bug when using yaml input
  - improved delete mode for both unix/yaml input
  - fixed possibilty of silent death in when in feed mode
0.1.3:
  - many optimizations to the scheduling algorithm
  - many opimizations in the locking algorithm
  - improved logging
  - automatic log rolling
  - added automatic lockd recovery code for sick lockd daemons
  - added ability to restart on SIGHUP
  - added ability to finish all pending jobs on SIGTERM (kill) 
  - added code to fix sqlite nfs sillyname bug
  - surpassed 1,000,000 jobs run
0.0.0:
  - initial version