Issues with BIND 9.5.0-P2

[Sotiris Tsimbonis]

On 08/05/2008 08:51 PM, Milan Jurik wrote:
> Hi,
> 
> V út, 05. 08. 2008 v 15:12, Sotiris Tsimbonis píše:
>> On 08/05/2008 03:53 PM, Sotiris Tsimbonis wrote:
>>> On 08/05/2008 02:41 PM, Milan Jurik wrote:
>>>> Hi,
>>>>
>>>> V út, 05. 08. 2008 v 13:31, Sotiris Tsimbonis píše:
>>>>> On 08/05/2008 11:21 AM, Elias wrote:
>>>>>> Hi all,
>>>>>> I've tried running 9.5.0-P2 on a Solaris 10 box but it is not running OK as queries fail (gives a SERVFAIL error) intermittently. Trace logging shows a 'socket: too many open file descriptors' error message and this does not happen when running 9.5.1b1.
>>>>>>
>>>>> Unfortunately, same 'too many open file descriptor' errors here, when I 
>>>>> downgraded from 9.4.3b2 to 9.4.2-P2 (Solaris 9 and 10, with unlimited 
>>>>> open files).
>>>>>
>>>> Did you try to define ISC_SOCKET_FDSETSIZE with value > 1024 for your
>>>> compilation?
>>> Initially no, I compiled 9.4.2-P2 using the values detected by 
>>> ./configure. I have now recompiled with ISC_SOCKET_FDSETSIZE=4096 and 
>>> will deploy the new binary shortly..
>>>
>> Bad luck with 4096.. A few seconds after startup it crashes..
>>
>> 05-Aug-2008 16:04:30.423 general: socket.c:2394: fatal error:
>> 05-Aug-2008 16:04:30.423 general: select() failed: Invalid argument
>> 05-Aug-2008 16:04:30.423 general: exiting (due to fatal error in library)
>>
>> I reverted back to 9.4.2-P2 for the time being, but if 'too many open 
>> file descriptors' show up too often, I'll probably move to 9.4.3b2 again..
> 
> Mea culpa, bad reading and c&p problem. Yes, Solaris allows to set
> FD_SETSIZE directly during compilation, so, "Linux" specific
> ISC_SOCKET_FDSETSIZE isn't the correct value to set.
> 
>>From the source code, -P2 improved UDP file descriptors handling, but
> the amount of file descriptors for select() arrays is still high. So
> setting FD_SETSIZE for some higher values (depending on your load) in
> CFLAGS is still valid for -P2 releases.
> 

Thanks for this, but we decided to try 9.4.3b2 again, with the patch 
posted at http://marc.info/?l=bind-users&m=121796486117750&w=2
Mainly because there is a notable difference in the system's load 
between the -P2 release (avg load 3) and b2 (avg load 2)..

Cheers,
Sotiris.