Tag: sysadmin

I run multiple Zope instances on one server. Each Zope instance listens on a different port (localhost:100xx). Historically I’ve just used Apache as a front end which forwards requests to the Zope instance.

Unfortunately there are periods of the year when one site gets a deluge of requests (for example; when hosting a school site, if it snows overnight, all the parents will check the site in the morning at around about 8am).

Zope is not particularly quick on it’s own – Apache’s “ab” reports that a dual core server with plenty of RAM can manage about 7-14 requests per second – which isn’t that many when you consider each page on a Plone site will have a large number of dependencies (css/js/png’s etc).

Varnish is a reverse HTTP proxy – meaning it sits in-front of the real web server, caching content.

So, as I’m using Debian Lenny….

1. apt-get install -t lenny-backports varnish
2. Edit /etc/varnish/default.vcl
3. Edit Apache virtual hosts to route requests through varnish (rather than directly to Zope)
4. I didn’t need to change /etc/default/varnish.

In my case there are a number of Zope instances on the same server, but I only wanted to have one instance of varnish running. This is possible – but it requires me to look at the URL requested to determine which Zope instance to route through to.

So, for example, SiteA runs on a Zope instance on localhost:10021/sites/sitea. My original Apache configuration would contain something like :

<IfModule mod_rewrite.c>

   RewriteEngine on

   RewriteRule ^/(.*) http://127.0.0.1:10021/VirtualHostBase/http/www.sitea.com:80/sites/sitea/VirtualHostRoot/$1 [L,P]

 </IfModule>

To use varnish, I’ll firstly need to tell Varnish how to recognise requests for sitea (and other sites), so it can forward a cache miss to the right place, and then reconfigure Apache – so it sends requests into varnish and not directly to Zope.

So, firstly, in Varnish’s configuration (/etc/varnish/default.vcl), we need to define the different backend server’s we want varnish to proxy / cache. In my case they’re on the same server –

backend zope1 {
   .host = "127.0.0.1";
   .port = "10021";
}
backend zope2 {
   .host = "127.0.0.1";
   .port = "10022";
}
Then, in the 'sub vcl_recv' section, use logic like :

if ( req.url ~ "/sites/sitea/VirtualHostRoot") {

   set req.backend = zope1;
}
if ( req.url ~ "/siteb/VirtualHostRoot") {
    set req.backend = zope2;
}

With the above in place, I can now just tell Apache to rewrite Sitea to :

RewriteRule ^/(.*) http://127.0.0.1:6081/VirtualHostBase/http/www.sitea.com:80/sites/sitea/VirtualHostRoot/$1 [L,P]

Instead….. and now we’ll find that our site is much quicker 🙂 (This assumes your varnish listens on localhost:6081).

There are a few additional snippets I found – in the vcl_fetch { … } block, I’ve told Varnish to always cache items for 30 seconds, and to also overwrite the default Server header given out by Apache etc, namely :

sub vcl_fetch {

    # ..... <snip> <snip>

    # force minimum ttl for objects

    if (obj.ttl < 30s) {

        set obj.ttl = 30s;

    }

    # ... <snip> <snip>

    unset obj.http.Server;

    set obj.http.Server = "Apache/2 Varnish";

    return (deliver);

}
I'm happy anyway. :)
Use 'varnishlog', 'varnishtop' and 'varnishhist' to monitor varnish.