Skip site navigation (1)Skip section navigation (2)
Date:      Thu, 2 Sep 2021 09:32:55 +0100
From:      Matthew Seaman <matthew@FreeBSD.org>
To:        freebsd-questions@freebsd.org
Subject:   Re: BIND 'max-cache-size' Value on FreeBSD-13.0
Message-ID:  <57a5c2ef-87c1-dd1a-775e-acae89b0bbbc@FreeBSD.org>
In-Reply-To: <bef9fde7-b36a-ba87-471f-b512f6b33ba4@tinka.africa>
References:  <bef9fde7-b36a-ba87-471f-b512f6b33ba4@tinka.africa>

next in thread | previous in thread | raw e-mail | index | archive | help
On 02/09/2021 09:13, Mark Tinka wrote:
> Hi all.
> 
> Ever since we moved from BIND-9.11 to BIND-9.16, we've been experiencing 
> 'named' crashing after 24hrs - 36hrs on high-load resolver-only servers, 
> running on FreeBSD-13.0.
> 
> We found that the reason for this was due to BIND running out of swap 
> space.
> 
> An increase in swap space by creating a 4GB swap file did not help.
> 
> So we are now playing with the 'max-cache-size' value in BIND. The 
> system has 15GB of physical RAM. Limiting BIND to 13GB of memory does 
> not work; 'named' still crashes due to a lack of swap space.
> 
> We have then switched to % values, and it's still crashing for the same 
> reason at 90% and now 80%.
> 
> We are now testing 70%.
> 
> Anyone have some idea of how we can get this under control?
> 
> Is there a possibility that BIND is not properly understanding how much 
> physical RAM is available to FreeBSD, and just burns through it anyway, 
> tripping swap space in the process? I can't think of any reason why BIND 
> would keep burning RAM if it has been told to limit its demand to a 
> certain value or %.
> 
> All help appreciated. Thanks.

Hmmm.... unlike many big opensource groups, ISC has traditionally used 
FreeBSD extensively as a development platform, so they should be on top 
of the differences between FreeBSD and Linux with regard to memory 
management.

You've clearly got some sort of memory leak, which you are attributing 
to bind not managing its cache correctly.  I think that may possibly be 
a red-herring, and the leak is occurring in some other aspect of bind 
operation.  But what that might be I have no idea.

I suggest asking on the bind-users@lists.isc.org mailing list, as that's 
where the ISC devs and many bind specialists hang out.

	Cheers,

	Matthew




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?57a5c2ef-87c1-dd1a-775e-acae89b0bbbc>