Re: [Condor-users] Using Condor (Windows) with Linux File Servers

[Ari Silver <ari_ag@xxxxxxxxx>]

--- Matt Hope <matthew.hope@xxxxxxxxx> wrote:

> On 12/15/05, Ari Silver <ari_ag@xxxxxxxxx> wrote:
> > --- Matt Hope <matthew.hope@xxxxxxxxx> wrote:
> >
> > > On 12/15/05, Ari Silver <ari_ag@xxxxxxxxx>
> wrote:
> > > > Hello,
> > > >
> > > > I'm trying to get a Windows 2003 Server pool
> > > running
> > > > Condor to use linux file servers. The primary
> > > reason
> > > > for this is that the executables I wish to run
> > > require
> > > > a number of dlls and I don't want to have to
> > > specify
> > > > which dlls should be transferred each time a
> job
> > > is
> > > > run.
> > > <snip>
> > > > 007 (008.000.000) 12/15 10:02:58 Shadow
> exception!
> > > >        Error from starter on
> > > machine504.mydomain.com: Failed
> > > > to open standard output file
> > > >
> '//bonnie/home/silvera/CondorJobs/demo/demo.out':
> > > > Invalid argument (errno 22)
> > > >        0  -  Run Bytes Sent By Job
> > > >        0  -  Run Bytes Received By Job
> > > >
> > > > I made sure the directories have the necessary
> > > > permissions (777). Can anyone shed some light
> on
> > > this?
> > >
> > > On windows jobs execute as a user with no
> > > priviledges whatsoever. this
> > > includes accessing network level resources such
> as
> > > unc shares.
> 
> apologies - I misread you as saying you wanted to
> read/write to these
> directories from the executing machines. Ignore the
> advice in previous
> mail.
> 
> > I ran condor_store_cred add on the execute
> machines
> > and I am still experiencing the same problem. How
> is
> > Condor able to write to the .log file in the
> directory
> > in question, but cannot write to the .out file in
> the
> > same directory?
> 
> The out file (and err file) is written to at the end
> by copying back
> the stderr and stdout redirected files. The log file
> is written to
> repeatedly as the job is on going.
> This can sometimes trigger some nasty behaviour with
> timeouts on sessions.
> 
> The job's userlog is written to in a very different
> maner than the
> file transfer mechanism - is there anything special
> (Active directory
> permissions, security dongles or somesuch)
> controlling access to the
> network?
> 
> I *never* recommend to any of the users of my pool
> that they use a
> location remote of their submission (and thence
> shadow machine) for
> placing files being copied back from the excute
> machine.
> If for some reason beyond condors control windows
> has an issue
> accessing the network at the copy back point then
> the entire job will
> restart.
> 
> Painful in the extreme and nigh on impossible to
> really debug - I
> wouldn't go there if I were you.
> 
> Matt
> 


Ok. Given what was said in the previous post, how
would you suggest I solve the problem of having
executables with dependencies on a number of dlls
without placing the binaries (dlls) on each execute
machine in my pool?

I'm not sure how to get a networked file system
working and simply performing a net use ... will not
really help because although I can keep the binaries
in one location, I still have to specify each dll to
copy over to the execute machine.

Any ideas?

Thanks.

-Ari




__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com