Rewrite URL path in reversed proxy setup

ptemmerman · 11-17-2008, 08:54 AM

Hi all.

I have an Apache server that takes care of load balancing four Tomcat servers, using mod_proxy.
The four Tomcat servers are setup in two clusters:

cluster1: 2 Tomcats 5.5.7
cluster2: 2 Tomcats 5.5.27

Now, I have depending on the application that is being requested, the request should be send to cluster1 or cluster2.
I have this setup working using the <Location> tag.

Code:

<Proxy balancer://mycluster1>
        BalancerMember ajp://172.18.0.39:8009 route=vdebian1
        BalancerMember ajp://172.18.0.40:8009 route=vdebian2
</Proxy>

<Proxy balancer://mycluster2>
        BalancerMember ajp://172.18.0.78:8009 route=vdebian3
        BalancerMember ajp://172.18.0.84:8009 route=vdebian4
</Proxy>

<Location /SEDO>
       ProxyPass balancer://mycluster1 stickysession=JSESSIONID
       ProxyPassReverse /
</Location>

<Location /SEDO-NEW>
       ProxyPass balancer://mycluster2 stickysession=JSESSIONID
       ProxyPassReverse /
</Location>

This works perfectly as I expected. However, the problem I have is that some images or links on my webapp are specified relatively.
For example <img href="image.gif">, which results in errors because the image cannot be found on the root of my server, which is logical.

So, I thought I'd use the following:

Code:

<LocationMatch ^(.*\.gif)$>
        ProxyPassMatch balancer://mycluster1/$1
</LocationMatch>

However, the problem is that I do not know to which application the image belongs, thus I don't know either to which cluster I should proxy the request.
So I thought using mod_proxy_html to rewrite the URLs in the HTML files:

Code:

<Location /SEDO>
       ProxyPass balancer://mycluster1 stickysession=JSESSIONID
       ProxyPassReverse /
       SetOutputFilter proxy-html
       ProxyHTMLURLMap (.*\.gif) http://172.18.0.39:8080/$1 Rie
</Location>

This works, however, what I do not like is that I map the URLs to a hardcoded server. What if the Tomcat instance is down? Then my images won't be loaded.
As far as I tried, it is also impossible to specify a cluster path as URL.

Thus, my question folks, how can I solve this? I've been thinking about using Vhosts, but I don't know if that would be the right solution. Maybe there is something easier, such as altering a HTTP reponse header in order to know which webapp requested the image.

Any help will be kindly appreciated.

ptemmerman · 11-18-2008, 06:53 AM

This is solved.
In order to help other people with the same problem, I will describe the cause and give the solution.

PROBLEM:

When the browser sees a resource in a HTML file, such as <img href="image.gif", then it takes the current URL and removes the last part of the URL upto the slash.

Let's assume the URL is:
http://172.18.0.1/SEDO

In order to retrieve the relative image, the browser removes the last part of the URL, which would make it something like this:

http://172.18.0.1/image.gif

And this is where the problem resides. That particular image cannot be found at that location.

SOLUTION:
Rewrite all URLs to contain a trailing slash.

Code:

RewriteEngine On
RewriteRule ^/([^/\.]+)$ $1/ [R=301,L]

The URL http://172.18.0.1/SEDO will now be rewritten to http://172.18.0.1/SEDO/. When the browser sees the <img> tag with a relative path, it will remove the last part of the URL upto the slash, and insert the image name. This would be:

http://172.18.0.1/SEDO/image.gif

And tataa, it works!