Re: [HTCondor-users] DAGMan write jobs log into *.nodes.log even if log file specified for all jobs

["R. Kent Wenger" <wenger@xxxxxxxxxxx>]

On Tue, 12 Nov 2013, Tom Downes wrote:

> I expressed some concern a while back about these logs. I can't remember the
> details, but the underlying issue was that was large DAGs can generate tens
> of thousands of separate open/close operations on these logs over the course
> of a day. The way things ran here, the default location of these logs ended
> up on NFS.

Okay, I guess one thing is that we'd recommend not having the logs on NFS, 
especially the default/workflow log.  (We already have this ticket:
https://htcondor-wiki.cs.wisc.edu/index.cgi/tktview?tn=3930,4, I guess
that should be a pretty high priority so it's easy for you guys to *not* 
have the default/workflow log on NFS.)

There are two places that would be doing opens/closes on the log files:
1) The schedd
2) DAGMan

The default/workflow log should really cut down on the opens/closes that 
DAGMan does -- in most cases it will probably just open/close the log file 
once or a handful of times.

Obviously the schedd will be doing more opens/closes if it's writing to 
two log files instead of one for each job (although not all events are 
written to the workflow log, so it's not twice as many).

> The way I read the manual at the time, I wasn't expecting the log files to
> show up at all given the Log-ging settings in the submit file. I am busy
> with other matters right now, but you might take a 2nd look at Condor 8.x
> era e-mails from me (perhaps on the LIGO list).

Hmm -- I guess one option would be to have the schedd *not* write to the 
log file specified in the job's submit file when a default/workflow log 
file is specified by DAGMan.  (I'm not sure I really like that idea, 
though.)

Would this be significanly less of a problem if it were easier to at least 
move the default/workflow log off of NFS?

Kent