BIND 9.4.1-P1 freezes randomly on FreeBSD 6.1

[Mark Andrews]

> 
> On Sep 4, 2007, at 2:35 PM, Craig Cocca wrote:
> 
> >
> > On Aug 30, 2007, at 1:22 PM, Craig Cocca wrote:
> >
> >>
> >> On Aug 26, 2007, at 6:17 PM, Mark Andrews wrote:
> >>
> >>>
> >>>> Recently, we've updated our internal nameservers from BIND 9.3.2 to
> >>>> 9.3.4-P1, and then to BIND 9.4.1-P1.  We're running this on FreeBSD
> >>>> 6.1 on a P4 2.8 GHz with 1GB of RAM.  We're experiencing a problem
> >>>> where named will be running normally, at 10-14% CPU load and  
> >>>> right at
> >>>> the 515M memory cap that we set for it, and then it will suddenly
> >>>> stop doing recursive lookups.  What is interesting here is that the
> >>>> box will continue to do lookups for domains for which we have
> >>>> authority, but lookups of outside domains just flat out stop  
> >>>> until we
> >>>> restart named.
> >>>>
> >>>> I am curious if anyone else has experienced a similar problem with
> >>>> BIND 9.4.1-P1, and if so, if there is a fix for this.
> >>>>
> >>>> Thanks,
> >>>>
> >>>> --Craig
> >>>
> >>> 	This is symptomatic of named hitting the datasize limit.
> >>>
> >>> 	Which config option did you use?  The cache size should
> >>> 	be limited by max-cache-size which is a soft limit and
> >>> 	should be less than datasize.
> >>>
> >>> 	Note FreeBSD sets datasize to 512MB by default and you need
> >>> 	to reboot to reconfigure this upwards.  See the FreeBSD
> >>> 	handbook for how to change this.
> >>>
> >>> 	Mark
> >>> -- Mark Andrews, ISC
> >>> 1 Seymour St., Dundas Valley, NSW 2117, Australia
> >>> PHONE: +61 2 9871 4742                 INTERNET:  
> >>> Mark_Andrews at isc.org
> >>>
> >>>
> >>>
> >>
> >>
> >
> > Even after following Mark's suggestions above, we are still seeing  
> > the exact same problem where named becomes unresponsive to  
> > recursive lookups at random intervals.  I'm starting to think that  
> > this might be a memory leak in BIND 9.4.1-P1, and I would like to  
> > know what kind of debug output would be helpful here to help you  
> > (and us) solve this issue and get BIND into a stable operating mode.
> >
> > Thanks,
> > Craig Cocca
> >
> >
> >
> 
> Following  up on my earlier post, here are some of the stats of the  
> box that we're running BIND 9.4.1-P1 on:
> 
> P4-2.8 GHz
> 1 GB memory
> ~500 queries/second
> named runs as a recursive lookup nameserver at 10-15% CPU load in a  
> 515 MB footprint.
> 
> I would be happy to provide any additional debugging information that  
> someone might need to help steer me in the right directions.
> 
> Best,
> Craig Cocca

	max-cache-size controls the amount of memory the *cache* uses.
	It does not control all the rest of memory named uses.

	You are *still* running out of memory, increase the systems
	max datatsize from 512MB and/or decrease max-cache-size,
	max-cache-ttl.

	If you increase maxdsiz you will also need to set datasize in
	named.conf or change the default (kern.dfldsiz)

	e.g.
	/boot/loader.conf
	#kern.maxdsiz="1073741824"      # Set the max data size

	/boot/defaults/loader.conf
	#kern.dfldsiz=""		# Set the initial data size limit
	#kern.maxdsiz=""                # Set the max data size

	Mark
-- 
Mark Andrews, ISC
1 Seymour St., Dundas Valley, NSW 2117, Australia
PHONE: +61 2 9871 4742                 INTERNET: Mark_Andrews at isc.org