[Condor-users] eviction problems with 7.4.2

["Smith, Ian" <I.C.Smith@xxxxxxxxxxxxxxx>]

Dear All,

I've recently been taking a look at checkpointing under the vanilla
universe*. I had everything working fine using Condor 7.0.2 on
the execute hosts (running Win XP SP 3) but when I moved
to 7.4.2 there are problems when jobs get evicted. 

When this happens because of mouse/keyboard activity I see
the machine go through the usual Claimed/Busy -> 
Preempting/Vacating -> Preempting/Killing  -> Owner
states but the job carries on running according to condor_q
(and the log file).

If I look on the execute host, then the 
execute directory has been wiped but condor_q insists that
the job is still running. Eventually when the job starts again
I see a "job disconnected" error in the job's log file. As
well as this, none of the output files get returned to the $(SPOOL)
area.

The execute hosts have this config:

WANT_SUSPEND   = FALSE
WANT_VACATE  = TRUE
START	= ( $(UWCS_START) && $(OfficeHours) \
|| ( $(OfficeHours) == FALSE ) && ( $(ShutdownHours) == FALSE ) )
SUSPEND = FALSE
CONTINUE= $(UWCS_CONTINUE)
PREEMPT= $(UWCS_SUSPEND) && $(OfficeHours) 
KILL= TRUE 

which worked fine with 7.0.2.

Any ideas what may be wrong. Could it be something to do with one
of the daemons not receiving a signal from condor_kbdd ?

regards,

-ian.

* I've written up some detailed instructions on this for the benefit of
our users. If anyone is interested I'll post the link here.

--------------------------------------------
Dr Ian C. Smith,
e-Science Team,
The University of Liverpool,
Computing Services Department