[Condor-users] Problem with START configuration for allocating whole machine - VMs stuck in owner state after job is removed

[David Brodbeck <brodbd@xxxxxxxxxxxxxxxx>]

I'm trying to do something along the lines of what's described here,  
to provide a single-slot option for large jobs:
http://nmi.cs.wisc.edu/node/1482
I'm implementing it on condor 6.8.8, so I've changed the syntax to use  
VirtualMachineID instead of SlotID and vm1 instead of Slot1.

However, I'm running into trouble with machines not recovering when  
the large jobs finish.

It works in that jobs marked with "+RequiresWholeMachine = True" will  
only run in vm1, and the other VMs on each machine are marked as being  
in "Owner" status once the job starts.  However, when I remove the  
RequiresWholeMachine job from the queue, the other VMs get stuck in  
the "Owner" state and never return to being unclaimed.  Here's my  
START condition:

START	= ( ( $(CPUIdle) || \
                      (State != "Unclaimed" && State != "Owner")) \
	&& (VirtualMachineID == 1 || vm1_RequiresWholeMachine =!= True) && \
        (TARGET.RequiresWholeMachine =!= True || VirtualMachineID ==  
1) )

Can anyone spot what I'm doing wrong that's preventing VMs from  
returning to the "Unclaimed" state once the RequiresWholeMachine job  
is removed?  They seem to stay that way until I run 'condor_reconfig'  
to force a reload.

--

David Brodbeck
System Administrator, Linguistics
University of Washington