Hi Yuriy,

That segmentation fault deep in the libc remembers me we did see
something on those lines while troubleshooting a different issue
on your collector box:

http://www.mail-archive.com/[email protected]/msg01475.html

At that stage, the only thing that did help against these was to
disable (or relax) compiler optimizations while building the code
- ie. --enable-relax or --enable-debug. 

Did you recompile the package and/or upgrade? If this is the case
can you please give these a try? Otherwise ...

If you get code from the CVS, i've just committed a patch that
classifies those "orphan fragment" log messages as debug. Indeed
the tip Sergio gave stands valid - enable internal buffering on
your probes.

Cheers,
Paolo


On Wed, May 05, 2010 at 06:32:28PM +0300, Yavetskiy Yuriy wrote:
> Hello.
>
> I have disabled buffer in previous configuration.
> I enabled buffer, setting next sizes:
> plugin_buffer_size: 2024000
> plugin_pipe_size: 80960000
> But on this config nfacctd dies in 20-30 seconds after start with next  
> message:
> kernel:[3607959.925286] nfacctd[7172]: segfault at 7f346c5fd000 ip  
> 00007f346f85c903 sp 00007fffdb934da8 error 4 in  
> libc-2.10.2.so[7f346f7e0000+14a000]
> And also during this 20-30 seconds, I see 3-10 old warnings, insted 50  
> with buffer disabled.
>
> Sergio Charpinel Jr. wrote:
>> Hi,
>>
>> I have a different setup , as I dont use balanced links.
>> But I was receiving WARN: messages like you, because of small buffer
>> size in pmacctd daemon.
>>
>> You can try to set this options:
>> plugin_pipe_size: 10240000
>> plugin_buffer_size: 102400
>>
>> Hope this helps.
>>
>> 2010/5/5 Yavetskiy Yuriy <[email protected]>:
>>   
>>> Hello.
>>>
>>> Packets pass from one witch to another pass through few links in LACP (802.3
>>> ad). There are linux boxes in bridge, one per link. Pmacct with nfprobe
>>> plugin works on each linux bridge. Nfprobe exports data to collector on
>>> other linux box (nfacctd).
>>> First, I see errors/warnings on sensor:
>>> Expiring orphan fragment ip_src=x.x.x.x ip_dst=y.y.y.y
>>> Its easy to explaine - TCP session are brocken because lacp balance traffic
>>> between links, and packets from one session pass through diffrent linux
>>> boxes.
>>> Question: how to disable connection traking of tcp sessions.
>>> Also I have a lot of warning on nfacctd:
>>> WARN: expecting flow '2206474' but received '2206496' collector=y.y.y.y:2100
>>> agent=x.x.x.x:0
>>> I see such warning in this list, it was connected with ipv6, but I don't use
>>> ipv6 yet.
>>> Can this warnings on collector be related to LACP?
>>>
>>>
>>> --
>>> WBR
>>> Yavetskiy Yuriy
>>> ULTI-RIPE
>>>
>>>
>>> _______________________________________________
>>> pmacct-discussion mailing list
>>> http://www.pmacct.net/#mailinglists
>>>
>>>     
>>
>>
>>
>>   
>
>
> -- 
> WBR
> Yavetskiy Yuriy
> ULTI-RIPE
>
>
> _______________________________________________
> pmacct-discussion mailing list
> http://www.pmacct.net/#mailinglists

_______________________________________________
pmacct-discussion mailing list
http://www.pmacct.net/#mailinglists

Reply via email to