In the dim dark past of October 2005, spo0nman and teemus did a hack in office. It was a tiny bit of javascript magic combined with a lot of heavy PNG maps and individually marking each cube's position on the map. It was the beginning of so many other nights when we did hacks - even weekends. But that's irrelevant. What is relevant is that to avoid slashdotting my puny little desktop, we had to distribute our load across all the machines - someone asked me to explain how we did that, quite recently and it needs a quick mention outside at least.

RewriteMap: Introducing to you, a small bit of hidden magic inside the apache mod_rewrite module - something which gets just a passing mention in the manual. You can define a RewriteMap which is a program - a full fledged script in any language you please.

RewriteMap    lb      prg:/usr/local/bin/lb.pl

RewriteRule   ^/floors/(\?.*)?$ ${lb:$1}  [R]

The script is not invoked for every request, but an instance of the script is kept alive and feed one line at a time. And because spo0nman wrote it, it ended up being a perl script. Here's the entire perl code.

#!/usr/bin/perl -w

$| = 1;

my @urls = qw (http://m1 http://m2 ...)

$cnt = 0;
while () {
        $cnt = (($cnt+1) % 3);
        my $mirror = $urls[$cnt];
        print "$mirror\n";
}

Now, the beauty of this is not in this code. The perl real beauty that came out of this was something which generates and rewrites dynamically without restarting apache - with a db or just about any data source you can code up.

Imagine a script which watches your access log to accumulate statistics. Now combine that with a script which wget's frequently hit URLs into a local file. And then imagine a perl script which does a stat() on that file and does an internal redirect to static pages if the file is recent enough (blog archives *cough*).

Implementing something like this into S9Y would make a lot of sense - the gain of hitting a static file would be a LOT better than using something like APC. Would make sense for someone like lunatech who still gets heavy hits on his comment disallowed archives but still uses php to serve out the pages (at least, they seem static, yet send out a X-Blog: Serendipity header). And rather than statically rendering all the pages (like I do), a hack like this could let you do only pages which get n+ requests per day or something and clean up on a cron with stat.atime values.

Hope someone reads this blog and saves me all that work ;)

posted at: 07:44 | path: /hacks | permalink | Tags: public, apache, optimisations