Bret Taylor from FriendFeed presented today at San Francisco MySQL meetup on how FriendFeed uses MySQL. If you’ve read Bret’s blog post previously, the presentation itself wasn’t news, but the Q&A session was pretty good. A few interesting takeaways:
- For replication FriendFeed relies on MySQL replication, and it works quite well
- They did try document storage systems, including Couch DB, but none proved to be mature enough for production deployment
- They’re aware that a write might take longer than usual in this two-phase commit scenario, but they don’t need to present the data in real-time right away – a delay of a few seconds is acceptable for the users
- Some users tax the existing index system by following 30,000 friends on FriendFeed
- Failover is done manually most of the time – from Bret’s experience with Google and FriendFeed, a babysitter script for automatic failover tends to introduce issues
- There are some new sorting parameters ready to be pushed out for FriendFeed, and for each new sorting parameter FriendFeed creates a new set of indexes – there’s never ORDER BY in any of the queries
Hi, you have made a good summary of Bret’s presentation and blog post. But i can not still understand how they skip the order by in their queries. As far as I know InnoDB do not has reversed primary key index order, so how do they retrieve the data reversed ordered without order by?
Thanks in advance for your reply