[Pdns-dev] virtual memory issue and gmysql

[Stephen Manchester]

This would correlate with what I've seen as well.  We are not that  
busy at all, and even though I have our MySQL connection timeout set  
for 8 hours, I'm sure a single connection could easily not get used  
for that long of a time.  Not sure if pdns cycles the connections  
regularly, or just uses the first available one, but if the later,  
then that could explain it easily.  Still haven't looked at the code.

Just thought I'd drop the note.


On Oct 2, 2007, at 9:47 PM, Patrick Domack wrote:

> I did more testing, and actually watched the log file this time and  
> the log shows this for every extra 10megs of memory allocated:
>
> Oct  3 00:31:08 blade2 pdns[13541]: Backend error: Failed to  
> execute mysql_query, perhaps connection died? Err=1: MySQL server  
> has gone away
> Oct  3 00:31:08 blade2 pdns[13541]: gmysql Connection succesful
> Oct  3 00:31:08 blade2 pdns[13541]: gmysql Connection succesful
> Oct  3 00:31:13 blade2 pdns[13541]: Backend error: Failed to  
> execute mysql_query, perhaps connection died? Err=1: MySQL server  
> has gone away
> Oct  3 00:31:18 blade2 pdns[13541]: gmysql Connection succesful
> Oct  3 00:31:18 blade2 pdns[13541]: gmysql Connection succesful
> Oct  3 00:33:18 blade2 pdns[13541]: Backend error: Failed to  
> execute mysql_query, perhaps connection died? Err=1: MySQL server  
> has gone away
> Oct  3 00:33:18 blade2 pdns[13541]: gmysql Connection succesful
> Oct  3 00:33:18 blade2 pdns[13541]: gmysql Connection succesful
>
> This is with a request rate of one every 29 seconds, and a mysql  
> timeout of 60 seconds.
>
> This isn't an issue with busy servers, cause it never has a 60sec  
> idle time between requests. I don't have this issue with 7 of my  
> production server, just the 3 I setup and are in development  
> testing, cause they aren't high load yet.
> But this means a high load server, that has a temp routing issue or  
> other problem, will also start to notice this issue.
>
> Quoting bert hubert <bert.hubert at netherlabs.nl>:
>
>> On Mon, Oct 01, 2007 at 03:30:24AM -0400, Patrick Domack wrote:
>>> I have seen reports of this in the users maillist.
>>
>> Indeed - but quite a number of very large installations do not see a
>> problem. So it has to be something site specific.
>>
>>> the rpm package for pdns-static version 2.9.21
>>> on redhat el5 x86_64
>>
>> Can you count if the chunks of 10240KB it loses correspond to  
>> certain errors
>> in the log perhaps?
>>
>> And does your patch fully resolve the issue?
>>
>> Thanks.
>>
>>
>> --
>> http://www.PowerDNS.com      Open source, database driven DNS  
>> Software
>> http://netherlabs.nl              Open and Closed source services
>>
>
>
>
> _______________________________________________
> Pdns-dev mailing list
> Pdns-dev at mailman.powerdns.com
> http://mailman.powerdns.com/mailman/listinfo/pdns-dev