Thursday, 17 December 2009

"X-Moz: prefetch" and skewed page-hits

Earlier today I installed a WordPress plugin recommend for tracking the popularity of posts. The plugin is unsurprisingly named "Recently Popular". After installing the plugin I ran some quick tests and found that I was getting extra hits recorded. I spent a bit of time back-tracking to find the source and after systematically disabling all other plugins and page elements found that it was firing in wp_head() in the page header.

After some more digging, I noticed that the extra hit was for the chronologically next published post and that the problem occurred in both WordPress and WordPressMU. This wasn't making a lot of sense so I decided to try a different browser - more of a sanity test than anything. That's when I found it didn't occur in Chrome, or Opera - just Firefox 3.5.6 that I'd upgraded to a few hours earlier.

I fired up the Live HTTP Headers add-on and checked out the requests Firefox was making. It was definitely making both post requests. I took a closer look at the second request and noticed the extra header "X-Moz: prefetch".

A quick search for X-Moz: prefetch turns up Mozilla's Link prefetching FAQ which gives a good description of what is happening and why. WordPress creates a tag similar to the following when wp_head() is executed:

<link rel='next' title='The Next Post' href='http://your_domain/year/month/day/the_next_post/' />

I am unaware of anyway to disable the prefetch hints. You could edit your header.php and remove the wp_head() statement, but many plugins rely on the execution of this function so results could be unexpected and undesirable. The issue for me was not that the hint was published but that the prefetch hits were being counted as real post requests, as well as the actual request when I clicked through a second or two later. This would seriously skew the perceived popularity of posts.

My solution was to ensure that the Recently Popular plugin ignored post requests that passed the "X-Moz: prefetch" header. Depending on your server configuration, the method of checking the header exists may differ - apache_request_headers() (alias getallheaders()) is only supported when PHP is installed as an Apache module. Most servers should support checking for $_SERVER['HTTP_X_MOZ'].

I wonder how many other people will wonder why their page hit stats have mysteriously increased without any increase in ad impressions, etc.

I will contact the plugin author to suggest an update once I've published this post.

No comments:

Post a Comment