DNS-Answers didn't leave the system

[Tom Schmitt]

Thank you for your hints.

I don't think, the syslog is the source of my problems because I don't log
to syslog, instead I log directly into a file. (it's faster than using
syslog, at least during my tests)

Of course there could be a problem with the file-locking if all
named-threads try to write to the logfile, but I don't believe that this
causes my problems because my problems doesn't occure when there were more
logmessages than usual. I couldn't establish any correlation between the
load (or the amount of messages) and the lost DNS-answers.
Anyway, I will have a look and examine the file-locking for any errormessage
from the kernel.


> --- Ursprüngliche Nachricht ---
> Von: Robert Lee <robert at spin.net.uky.edu>
> An: TomSchmitt at gmx.de
> Betreff: Re: DNS-Answers didn't leave the system
> Datum: Fri, 5 May 2006 09:28:43 -0400 (EDT)
> 
> I'm testing  but I recently noticed this behaviour and I'm feel it may
> be related to syslog.  In my case those underscore hostnames were 
> generating lots of log messages when zones updated.   I don't know for
> sure 
> if this is it but you may look to see if syslog could be part of 
> the problem.
> 
> I have also seen this problem when log messages are sent to the console
> and the console is not running a window manager.  I'm now
> testing to see what happens when a burst of messages are sent to syslog
> and running a window manager as in my most recent case.  To avoid the
> delay don't send logs to the console if they don't need to be there. 
> 
> I think that when named is logging messages there is some
> delay added while syslog logs the message to the console..  Most of the 
> time it isn't a problem yet at times the logging of errors/warnings causes
> the server to delay much longer waiting to finish logging when a burst 
> of errors or warnings occurs.
> 
> If you find another reason let me know, since your problems sounds
> like mine.
> 
> Regards,
> 
> Robert

> >Date: Fri, 5 May 2006 09:54:10 +0200 (MEST)
> >From: "Tom Schmitt" <TomSchmitt at gmx.de>
> >To: comp-protocols-dns-bind at isc.org
> >Subject: DNS-Answers didn't leave the system
> >
> >Hi,
> >
> >I found a strange behaviour with Bind 9.3.2 on Solaris 10. Normally all
> >things work well, but sometime there are clients complaining about
> timeouts
> >when they do a nslookup.
> >
> >I looked into it (with the logfiles of named and tcpdump on the
> >networkinterface) and found out: If such a timeout occurs, the incoming
> >request appears in the tcpdump-log and in the query-log of the named too.
> So
> >the named obvious gets the request otherwise it wouldn't be in his
> logfile.
> >But: There is no answer in the tcpdump-log!
> >
> >Additional:
> >- it is definitly no load-problem
> >- it is no clientspecific. It is happening with differrent clients and
> only
> >one time out of a few thousands requests.
> >- there is no error-log-entry in the named-logs
> >- the request whose answer get lost is asking for a existing name for
> which
> >my DNS-server is authorativ.
> >- There was no reload or reconfig at the time of the lost request.
> >
> >I don't have any clue where to look for the soure of the problem anymore.
> Is
> >this a known behaviour? Or have anyone a guess what I should check to
> find
> >the source of the problem?
> >
> >Thanks,
> >Tom.

-- 
"Feel free" - 10 GB Mailbox, 100 FreeSMS/Monat ...
Jetzt GMX TopMail testen: http://www.gmx.net/de/go/topmail