rndc reconfig causing long timeouts

[Dave Clark]

I would be interested in learning if the BIND developers are making any
plans towards a version of BIND that asyncronously replies to queries while
reconfiging/reloading.  I don't personally have this problem, but I do know
of another person who does.  As far as I was able to learn, attempts to use
multithreading was not a solution even when using multiple CPUs.  Correct me
if I am wrong.

It would be ideal if there was some way to use rndc to have BIND add or
remove a single zone, but I have not thought of a practical implementation
for this, so I have not made a formal feature suggestion.

Dave

----- Original Message ----- 
From: "Tavis Gustafson" <tavis at hq.newdream.net>
To: <bind-users at isc.org>
Sent: Friday, September 30, 2005 11:04 PM
Subject: Re: rndc reconfig causing long timeouts


> On Fri, Sep 30, 2005 at 09:20:25PM -0400, Kevin Darcy wrote:
> > Tavis Gustafson wrote:
> >
> > >Hello. We are using bind 9.2.3 on linux kernel 2.4.24 .
> > >Our main zone file is 16MB.  We are experiencing 30 second lookup
> > >timeouts every time we execute an rndc reconfig.  We do this only when
> > >updating new or removed zones (sometimes doing it every few minutes).
> > >Is this beahaviour on par with others' like installations? And is there
> > >another way to tell bind about new zones besides the reconfig?
> > >When running rndc reconfig, named uses 35% of our system's main memory.
> > >During the timeouts named pegs the processor at 99% while disk activity
> > >remains relatively calm.
> > >
> > My understanding is that rndc reconfig *only* looks at and (re-)parses
> > the named.conf file, so the size of your zone file is irrelevant, and it
> > would explain why you're not seeing any significant disk I/O. More
> > important than the size of a zone or all zones would be _how_many_ zones
> > you have, and perhaps how complex those zones are (e.g. lots of
> > zone-level access restrictions, etc.). If you're doing reconfigs every
> > few minutes sometimes, I would expect that you have a *lot* of zones.
> > Maybe you should split those among multiple machines.
> >
>
> Yes, we have about 171,000 zones, all with multiple cnames like ftp,
> mail, www, etc, and sub-domains.  I am now thinking that the best way to
solve these
> timeouts is to make our ns1 machine not be the master on which we run
> the rndc reconfig but run the reconfig on a non public master and
> have it update our 3 public slave namservers.  What is the term for this
> type of
> setup and is there a good methodology for migrating from what we have now
(3
> nameservers, ns1 being master).
>
> -tavis
>
>
>
>