[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [HTCondor-users] Inconsistent execute dir permissions

Thanks for the quick response.

I think I found the problem, and fortunately it isn't a Condor issue. Looks like the Condor puppet module I'm using was fighting with Condor over the permissions of the execute subdirectories.

Sorry for the noise. I learned a lot more about the StarterLog* entries, at least.



On 3/17/2016 12:26 PM, Todd Tannenbaum wrote:

But on the nodes, I see inconsistent execute directory ownership,
sometimes a mix of slot users and condor. Other times all owned by

Any idea how this would be happening? Log entries to look for? Ever seen
it before? Any config changes to try?

In addition to my questions before, be aware that indeed it could be
normal owner as "condor" temporarily, but it shouldn't stay that way
long (thus all my questions in previous post).

HTCondor does a stat() on the $(EXECUTE) directory to determine how the
job sandbox should be created.   If it is world-writable, we do a
mkdir() as the user. If it is not, we do the mkdir() as condor then
chown() it over to the user or slot-user, since we have no realy choice...

So if you are willing to make the $(EXECUTE) directory world-writable,
HTCondor will avoid the chown from condor to user step.

Another couple quick questions:

- is your $(EXECUTE) directory on NFS with root-squash??

- do you have PRIVSEP_ENABLED=True in your condor_config on the execute
nodes??? (note: PRIVSEP support in HTCondor is deprecated, unsupported,
and soon to be removed from the code....)


John Hover
Group Leader | Grid Group/Experiment Services
RHIC/ATLAS Computing Facility | Brookhaven National Laboratory
jhover@xxxxxxx | 631-344-5828 | http://www.racf.bnl.gov/Members/jhover