Re: Http Accept filters (accf_http)

On Tue, Apr 22, 2008 at 10:34:14PM +0200, Ivan Voras wrote:
> Scott Long wrote:
> >Poul-Henning Kamp wrote:
> >>In message <480E307B.901_at_quis.cx>, Jille writes:
> >>>Hello,
> >>>
> >>>I've read about accf_http(9) some time ago, and I was wondering about 
> >>>it's performance.
> >>>Does it increase performance on all workloads ?
> >>>(I'm intrested in the improvements for a PHP-apache-webserver with 
> >>>about 50 request/second average.)
> >>
> >>I doubt you will see measurable performance difference from using
> >>request filters at such low traffic.
> >>
> >
> >The accept filters do reduce service latency and probably have a small
> >benefit in CPU utilization.  50 requests/sec is probably enough to see
> >a benefit for something like PHP or PERL.  It definitely won't hurt, and
> >even if there's no measurable benefit now, it'll help prepare you for
> >scaling in the future.
> 
> Does anyone know why accf_accept is disabled by default in the ports' 
> stock Apache 2.2 (it's disabled in the default config files)? I thought 
> it was because it was dangerous or flawed for some reason, though (at 
> least for light loads comparable to those of OP) it seems to work fine.

There's not technical reason actually. It's an "opt-in" feature ;-)
In the early 2.2.x times, httpd used to print a warning when accf_http 
is disabled. It was, of course, just a matter of loglevel. apache ran 
perfectly fine, but a warning got printed. Some users started 
complaining about how my port was broken, sometimes in very rude 
manner.
So I decided to explicilty disable AcceptFilter unless 
apache22_http_accept_enable is set to "YES" in /etc/rc.conf.

clem