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

Re: [Condor-users] Condor_LOGS

Hi Eric and alian,
here's what i am trying to do.
                         I was trying to run a code developed inernally. I am using CIFS/NFS mount to
mount the file system to compute nodes depending on the os. In most of the cases the jobs complete. How ever on windows systems the jobs get killed for a variety of  reasons. I was trying to get the schedd to determine which nodes had a history of jobs completed and schedule the jobs to these nodes. I have gobal evirment counter which is local to each system and central manager which takes care of this.
                       I was wondering if there is a built in way to do this. I  get thelog files and parse for completed jobs and increment the counter gobally.
which is hitting the performance of the Cenrtral manager. I thought a decission tree would help in choosing the node. But the decission tree generated is not generic and has a error rate of 10%.  If i have the logs written to the cntral manager i can regenerate my decission tree once every hr or so.
                      condor_fetch log works. but again i have to run  3 cron jobs every hr or so.  there is some one else woring on problem detection but i am not sure if it is the same domain.
                      is there a qos feature implemented in current release?. How can one get access to condor source and test integrating our own developments.
On 7/22/05, Nick Partner <nick@xxxxxxxxxxxxxxxxxxxx> wrote:

That's very true. Although it wouldn't exactly be hard to include syslog.h either and I think would be a very useful addition to logging in condor.  For windows, logging to the eventvwr is just as easy. It really would help a lot of us if condor provided a more consist logging mechanism. Like you say, there are many tools which can parse these standard formats


-----Original Message-----
From: Erik Paulson <epaulson@xxxxxxxxxxx>
Date: Fri, 22 Jul 2005 15:24:14
To:Condor-Users Mail List < condor-users@xxxxxxxxxxx>
Subject: Re: [Condor-users] Condor_LOGS

On Fri, Jul 22, 2005 at 03:19:26PM -0500, Alain Roy wrote:
> At 09:54 AM 7/22/2005 -0500, you wrote:
> >Hello All,
> >             Is there a option one can set to instruct the nodes to write
> >the log files to central manager.
> No, there isn't, sorry.
> I would argue that there shouldn't be: the logs are there to do debugging,
> and the disk is the most reliable way to ensure that we have logs. What if
> the central manager is down, or a network connection can't be made, or the
> central manager gets too busy writing log files?

And you can always use a shared filesystem, or condor_fetchlog to retreieve
a remote log. It would also be trivial to write a script that tailed() the
condor logfile and threw every line into syslog, and then use any of the
billion syslog concentrators.

Condor-users mailing list
Condor-users mailing list