Question

I’ve got Varnish (3.0.3) sitting as a load-balancer/static cache in front of two web servers. I’ve got a CDN set up using Original Pull method. If I grab a URL from an image on my site manually, drop in the CDN address, I can verify that original pull is working and the image is pulled to the CDN and served.

My application is fairly complex and I’m testing this CDN to see if it significantly speeds up the web app, so I don’t want to rewrite any of my php code to use the CDN images just yet.

What I’d like to do is set Varnish up to rewrite requests received for image files and pull them through the CDN instead of from the two Apache servers directly in my cluster.

I’ve read through the Varnish documentation and a couple howto’s online about doing something similar, but I just can’t get it to work properly and need a little help here.

Here are a couple different ways I tried doing this (edited for brevity):

sub vcl_recv {
  #if request is image, redirect to CDN
  if (req.url ~ "\.(gif|ico|jpg|jpeg|png)$") {
        set req.http.host = "cdn.domain.com/";
        error 750 req.http.host + req.url;

    }

 }
sub vcl_error {
  if (obj.status == 750) {
    set obj.status = 302;
    set obj.http.Location = obj.response;
    return(deliver);
  }
}

That didn’t work. It resulted in broken images everywhere, and anything that did show up was using the .webp extension, so it wasn’t being processed by the condition above.

So I tried this:

 backend cdn {
         .host = "cdn.domain.com";
         .port = "80";
}
sub vcl_recv {
  #if request is image, redirect to CDN
  if (req.url ~ "\.(gif|ico|jpg|jpeg|png)$") {

      set req.backend = cdn;
      return(lookup);
  }

}

This showed some images on the page, but when viewing their source, they looked to be coming from the Apache servers (the domain name wasn’t that of the CDN) and only about half the images were displaying…probably browser cache.

I’d love some input here, thanks guys.

Is there no way to use Varnish for this kind of redirect? Would I be better off setting nginx up in front of Varnish to rewrite requests to the cdn?

UPDATE:
Using both answers given below, I have the redirect working and an ACL in place to allow the CDN to pull images directly vs redirecting to itself. However, though I verified the ACL is allowing connection through by using my own external IP, the CDN isn’t pulling new images from the server. It gives a 502 error (odd<) instead of pulling the image from the local server to the CDN and serving it. This is what the block of my vcl_recv looks like now:

acl cdn {
     "ip.of.CDN";
}

sub vcl_recv {
  #if request is image, redirect to CDN
  if (req.url ~ "\.(gif|ico|jpg|jpeg|png)$") {
        if(!client.ip ~ cdn){
           error 750 "http://cdn.domain.com" + req.url;
        }
    }

 }
sub vcl_error {
  if (obj.status == 750) {
    set obj.status = 302;
    set obj.http.Location = obj.response;
    return(deliver);
  }
}

Answer 1

You can definitely do this with Varnish quite easily – no need to setup nginx or anything. Actually your first solution is very close to doing the trick. It just needs a few modifications.

sub vcl_recv {
  #if request is image, redirect to CDN
  if (req.url ~ "\.(gif|ico|jpg|jpeg|png)$") {
        error 750 "http://cdn.domain.com" + req.url;
    }
}
sub vcl_error {
  if (obj.status == 750) {
    set obj.status = 302;
    set obj.http.Location = obj.response;
    return(deliver);
  }
}

You forgot “http://” from your CDN URL, and you can omit the last slash from the host as all req.urls begin with /.

You also need to make sure that the vcl_error code is the first one that is run in vcl_error(). I.e. if you have multiple definitions of vcl_error, make sure that none of them get to deliver any output before the if (obj.status == 750) check is reached.

Bear in mind that this solution causes all client browsers to query your server first and then make another request to the CDN after the 302 redirect. This adds a significant delay to each image load, and is probably not the best way of determining if CDN improves your app performance.

Update: Regarding your problems with CDN showing 502 errors when trying to pull content from your origin. Relying on the remote IP address for determining the redirection is quite risky, as the CDN could very well use a number of servers to do the pull, and the addresses could change over time. That would make the VCL very laborious and error-prone to maintain.

Would it be possible setting up a unique virtual host for the CDN to use? For instance originpull.domain.com and setup the CDN so that it pulls content from that address instead of your primary www.domain.com address?

You could then modify the vcl_recv() as follows:

sub vcl_recv {
  #if request is image and request is not made from CDN, redirect to CDN
  if (req.http.host != "originpull.domain.com" &&
      req.url ~ "\.(gif|ico|jpg|jpeg|png)$") {
        error 750 "http://cdn.domain.com" + req.url;
    }
}

That would ensure that the requests from CDN will never be redirected.