from Hacker News

This is actually a lot harder to do than you might imagine, we need to scan the whole index and use a hyperloglog (handily implemented for us in Redis) to count the URLs.