9.5.0-P2 under Win XP: warnings about the 'files' limit in Event Log

[Jukka Pakkanen]

----- Original Message ----- 
From: "Danny Mayer" <mayer at gis.net>
To: "Jukka Pakkanen" <jukka.pakkanen at qnet.fi>
Cc: <bind-users at isc.org>
Sent: Sunday, August 10, 2008 9:22 PM
Subject: Re: 9.5.0-P2 under Win XP: warnings about the 'files' limit in 
Event Log


> Jukka Pakkanen wrote:
>> Danny Mayer <mayer at gis.net> kirjoitti :
>>
>>> Jukka Pakkanen wrote:
>>>> Danny Mayer <mayer at ntp.isc.org>
>>> kirjoitti :
>>>>> What is the error that you see? There are several possible errors that
>>>>> you might be seeing.
>>>>>
>>>>> Danny
>>>> When the 9.5.0-P2 crashes (freezes) there's no errors or anything in 
>>>> the
>>> event viewer. Only message is the above mentioned FD_SETSIZE when 
>>> restarting the
>>> service.
>>>> The named service keeps growing in memory size, and then randomly stops
>>> responding to anything. At that point the service stops growing as well. 
>>> But the
>>> service itself is still active (task manager lists it). We have set our 
>>> network
>>> monitor make a dns status query every minute, and if there's no response 
>>> to
>>> restart the service.
>>>> Usually the named crashes few times a day. One of our recursive servers
>>> crashed 4 hours ago and after restart named was about 28M of size. Now 
>>> after
>>> hour hours it's already 126 megs, 2195 handles and 9 threads. And this 
>>> is not
>>> very busy server.
>>>> Another observation, one of our not-so-busy caching-only servers was
>>> running 9.5.1b1, it run days without crashing, but was eating memory. 
>>> After 5
>>> days of running named was using over 2 gigs of RAM and the server was 
>>> running
>>> out of memory. Now installed 9.5.0-P2 to this server 4 hours ago, and it 
>>> has
>>> crashed 8 times by now... named just stops responding and we need to 
>>> restart the
>>> service.
>>>
>>> When you look at the number process how many handles and threads do you 
>>> have? You need to add those columns to task manager via the view->select 
>>> columns... menu. They are not there by default.
>>
>> In this recursive server (2xPIII, W2KSP4, 1G RAM) which crashed 5 hours 
>> ago named now has named using 138M of RAM, 2325 handles and 9 threads.
>>
>> The caching-only server (2xXeon dualcore, W2K3, 4G) where I switched from 
>> 9.5.1b1 to 9.5.0-P2 today, just crashed again. After restart the named 
>> used 28M RAM, 270 handles and 14 threads. Now after only 5 minutes it is 
>> 35 megs, 690 handles and 14 threads and both the memory consumption and 
>> handles are growing every cycle.
>
> And the logs do not have any error messages? It is likely to be in the 
> application event log unless you have reconfigured your logs.

Actually there are, in W2K server around the time named freezes we always 
get:
Windows cannot unload your registry file. If you have a roaming profile, 
your settings are not replicated. Contact your administrator.

DETAIL - Access is denied. , Build number ((2195)).



And in the W2K3 server we get the following error, when named stops 
responding:



Faulting application named.exe, version 0.0.0.0, faulting module ntdll.dll, 
version 5.2.3790.3959, fault address 0x0002a36b.

And just noticed that in this W2K3 server the named process actually exits 
when this happens, and there are both named.exe.hdmp and named.exe.mdmp 
files. In the W2K server the service doesn't exit when it stops responding, 
and there are no those files. You want me to email files to you?