Re: [Condor-users] Problem with multiple machines in a Windows pool

[Ralph Finch <ralphmariafinch@xxxxxxxxx>]

I checked a few things on the my work machine and a couple others which won't delete condor execute dirs..

• SYSTEM privs for the c:\condor directory are Full Control.
• All users in the condor pool have Windows admin privs associated with their accounts and of course the Administrative accounts have Full Control.
  
• c:\condor\bin\condor_rmdir.exe is enabled in every machine's condor_config file, that is, the two lines mentioned below are un-commented.
• All the Win7x64 condor installs were done from the .msi file deliberately to try to avoid install problems.
  
• The problem seems only to exist on Win7x64 machines, no XPs, and only after I implemented run_as_owner and all the CREDD stuff. I'm not saying it's a condor bug, but something I've overlooked in the setup.
  

I'll continue checking some other things as able, I'm in a remote mountain resort area with dodgy WiFi so can remote login to work only a couple times a day.

RF

On Mon, Aug 22, 2011 at 12:56 PM, John (TJ) Knoeller <johnkn@xxxxxxxxxxx> wrote:

It looks like an access rights problem.  the Condor daemons run as SYSTEM so if any of the files or sub-directories of the execute directory
deny delete access to SYSTEM, condor might be unable to cleanup after your job, this message seems to indicate that that is the case.   

ERROR: C:\Condor\execute\dir_3180 still exists after trying to add Full control to ACLs for PRIV_ROOT

There is code in the starter to try and change the ACLs on files in the execute directory after the jobs exits, but that code doesn't work
if the total pathname length of any file exceeds 260 characters.

There is a special program called condor_rmdir that ships with condor (in the bin directory) that can handle  deleting long paths
with SYSTEM deny ACLs, but I suspect that that program is not enabled in your condor_config file.

search for WINDOWS_RMDIR in your config file.  you should see something like this...

# On Windows, the built-in rmdir command will not delete directories
# or files that have Discretionary Access Control (DACLs) that deny
# access to the SYSTEM account. This binary will take ownership of
# files and directories and remove DACLs if needed to delete a
# directory.  Elevated priveleges are needed for it to operate so
# set this only if Condor runs as SYSTEM. (i.e. as a service)
#WINDOWS_RMDIR = $(SBIN)\condor_rmdir.exe
#WINDOWS_RMDIR_OPTIONS = /s /c

try un-commenting the last two lines.

-tj