BIND -P2 update plans (Was: Re: The BIND scandal)

[Doug Barton]

Thomas Rasmussen wrote:
> I've posted to the bind-users list to say this, but to confirm here: On 
> 7-STABLE from a few weeks ago on a couple of busy recursive servers, 
> this patch made an extreme positive difference. I was having problems 
> with constant timeouts, very slow recursive lookups when they did work, 
> and frequent errors about too many open files or somesuch in messages 
> (regardless of kern.maxfiles and FD_SETSIZE settings), all of this 
> disappeared when I applied P2. Number of successful queries almost 
> doubled the minute I restarted with the -P2 patch applied, no more 
> slowness or timeouts.

That's good news even taking your change to fd_setsize into account.

> This is the bind9.4 port by the way, 9.5 had even more weird errors and 
> behaviour. I've since seen various sources claiming that 9.5 isn't ready 
> for primetime on busy resolvers, so I'll wait for a while before moving 
> on to 9.5.

Yeah, if you don't have time to help debug the problems then sticking 
with 9.4 is a good decision. OTOH they can use all the help they can 
get. :)

> For the record, I have compiled dns/bind94 with
> 
>        make CFLAGS="-DFD_SETSIZE=65000" install clean
> 
> to avoid "too many open file descriptors" errors, but with this setting 
> (and increasing kern.maxfiles with sysctl) everything seems to be 
> running nicely. -P2 might have removed the need for increasing 
> FD_SETSIZE but this works, and for now I'll leave it at that.

I can certainly understand not wanting to change something that's 
working, but I would like to get at least a couple of users to confirm 
that -P2 works out of the box before I import them. I don't mind 
adding a "big fd_setsize" knob to the ports and the base, but I want 
to be sure it's needed first.

Doug

-- 

     This .signature sanitized for your protection