bind-9.10.0-P2 memory leak?

Mike Hoskins (michoski) michoski at cisco.com
Fri Sep 12 18:07:10 UTC 2014


-----Original Message-----
From: Thomas Schulz <schulz at adi.com>
Date: Friday, September 12, 2014 at 11:47 AM
To: "bind-users at isc.org" <bind-users at isc.org>
Subject: Re: bind-9.10.0-P2 memory leak?

>> Mike Hoskins wrote:
>>
>> Do you guys have max-cache-size set?  I didn't see it in the
>>borderworlds
>> named.conf.  I've seen similar growth problems when testing 9.x before
>> setting that (experiment at the time just to see what would happen, and
>> confirmed this behavior).  Set sensible resource limits based on
>>available
>> resources.
>
>I am going to see what happens with max-cache-size set, but I am convinced
>that there is a bug in bind. My named has been running for 7.5 weeks now
>and has been steadily growing in size except for a 1.5 week pause after I
>did an rndc flush. The process size started out at 36 MB and is now up to
>584 MB. But when I do an rndc dumpdb -cache I get a file that is only 5 MB
>in size. Given the automatic cache cleaning, named should stabilize in
>size in less than 7.5 weeks.


Just to be clear, I tend to agree with the memory leak hypothesis at this
point...  Based on the described behavior and past experience I related, I
initially just did a search of your config looking for max-cache-size.
Sorry for that, was in training at the time and somewhat distracted.

However, your use case is obviously very different from mine as you are
not doing recursion (my test environment without max-cache-size was, and
getting hit with an almost endless stream of random real-world queries
from my queryfile).

That said, I wonder if it could be dlz related?  That's the only thing I
see "special" about your config.  Just trying to find possible clues,
since I have ran all 9.9.x versions over time in heavily loaded production
environments (authoritative and recursive) without seeing the unbounded
growth you mentioned below for 9.9.x.

I do have a lot of interest in the community getting to the bottom of
this, as we are just planning a large upgrade in one of our environments
which will move caching clusters serving 6-8k clients over to 9.10.1.


>
>> -----Original Message-----
>> From: Vinícius Ferrão <ferrao at if.ufrj.br>
>> Date: Tuesday, September 9, 2014 at 10:17 AM
>> To: Thomas Schulz <schulz at adi.com>
>> Cc: "bind-users at isc.org" <bind-users at isc.org>
>> Subject: Re: bind-9.10.0-P2 memory leak?
>> 
>>>I'm having the exactly same issue. Take a look at my post @ServerFault:
>>>http://serverfault.com/questions/616752/bind-9-10-constantly-killed-on-f
>>>re
>>>ebsd-10-0-with-out-of-swap-space
>>>
>>>Sent from my iPhone
>>>
>>>On 09/09/2014, at 11:15, "Thomas Schulz" <schulz at adi.com> wrote:
>>>
>>>>> Hello
>>>>> 
>>>>> I recently upgraded my authoritative nameservers to bind-9.10.0-P2
>>>>>and
>>>>> after a while one of them ended up using all its swap and the named
>>>>> process got killed. The other servers are seeing similar behaviour,
>>>>>but 
>>>>> I restarted named on all of them to postpone further crashes.
>>>>> 
>>>>> I am using rate-limiting as well DLZ with PostgreSQL. The server has
>>>>>two 
>>>>> views. The operating system is FreeBSD 8.4.
>>>>> 
>>>>> My configuration:
>>>>> http://borderworlds.dk/~xi/named-leak/named.conf
>>>>> 
>>>>> Log of the memory usage:
>>>>> http://borderworlds.dk/~xi/named-leak/named-mem-usage.log
>>>>> 
>>>>> As you can see, in less than a week, named has grown more than 900MB
>>>>>in 
>>>>> size.
>>>>> 
>>>>> Is anyone else experiencing something similar?
>>>>> 
>>>>> If I need to provide more information, I will be happy to do so.
>>>>> 
>>>>> -- 
>>>>> Christian Laursen
>>>> 
>>>> What version did you upgrade from? I am seeing bind 9.9.5 and 9.9.6
>>>> grow without any evidence that it will ever stop. See my mail to this
>>>> list with the subject "Re: Process size versus cache size." Mine is
>>>> growing slower than yours, but it is now up to 548 MB.
>>>> 
>>>> Tom Schulz
>>>> Applied Dynamics Intl.
>>>> schulz at adi.com
>
>Tom Schulz
>Applied Dynamics Intl.
>schulz at adi.com



More information about the bind-users mailing list