Drain mode would be useful for a GRAM Service admin wanting to bring down a
machine gracefully for maintenance.  The admin would set the gram service for
"drain mode", meaning no new jobs will be allowed, but all existing jobs will
be processed as normal.  During this time, users submitting jobs should get a
fault type indicating that the gram service is in "drain mode".


Begin forwarded message:

From: Terrence Martin <tmartin@physics.ucsd.edu>
Date: July 1, 2008 10:38:36 PM CDT
To: Alain Roy <roy@cs.wisc.edu>
Cc: Charles Bacon <bacon@mcs.anl.gov>, osg-sites@OPENSCIENCEGRID.ORG
Subject: Re: Stopping new connections to a CE

A drain mode would be handy I have to say. Especially when you are
trying to juggle two CE to maximize access time to the cluster while
being able to perform routine maintenance.  As much as it is in
principal useful to alert to downtime in practice users are not
necessarily going to check the maintenance schedule of all sites
they might submit to. They might not have any choice of site. An
GRAM level "come back a bit later" message would be very useful I
think.