8 comments by 4 or more people

1. Mathew Mannion

   BlogBuilder will probably get a release tomorrow or the day after, which will make 3 in 3 weeks… ;)

   23 Aug 2005, 16:30

2. Chris May

   once a week, once a fortnight, one a month – it all looks pretty much the same from the perspective of someone doing 10 a day, I imagine.

   23 Aug 2005, 16:38

3. Mathew Mannion

   True, but is it really necessary to be rolling out a new version of flickr every half an hour? What does it acheive, and surely this level of iteration is risky

   23 Aug 2005, 16:54

4. Chris May

   That's exactly why I find the quote so interesting. Flickr is a big application, with a big user base and a high load. AFAICS it's not broken very often. What puzzles me is how they can have confidence that they haven't broken anything in half an hour. Even with the worlds greatest set of functional tests, I'd want some kind of manual tests to reassure me I hadn't missed anything.

   Of course, it could be that

   • the kind of changes they're making are of the trivial 'make this text more purple' kind; or
   • they have insanely well-separated components so that the scope for screw-ups is very limited; or
   • they really just release them to a test server once every half hour, and roll them out twice a week; or
   • it could just be that they have god-like coding powers and never make mistakes.

   Makes me regret not signing up for the London 'building flickr' event, so I could have found out a bit more.

   23 Aug 2005, 17:14

5. Mathew Mannion

   My money's on the test server one… :)

   23 Aug 2005, 18:37

6. John Dale

   My money's on the "no level 2 caching" theory coupled with the "server farm allowing for very easy roll-backs" theory so that frequent deploys are low risk, low cost to users and easily reversible.

   The more interesting question, I think, is what can they do that's worth rolling out with such frequency? It's not a huge team, and even with the best coders in the world, changes which are more than cosmetic just don't get written in that kind of timescale.

   23 Aug 2005, 22:24

7. Chris May

   Well, I've secured the last-but-one place on building flickr so I'll tell you on friday :-)

   23 Aug 2005, 22:31

8. They must not be using sessions, or have some sort of complicated session persistence outside of the container.

   I suppose if they had clever co-ordination from a frontside router and staged their deployments so there was always at least one running server, coupled with some clever session management you could transparently deploy apps, but twice an hour is just asking for trouble ;)

   24 Aug 2005, 10:19

Trackbacks

Trackback URL: http://blogs.warwick.ac.uk/chrismay/entry/release_once_every/trackback/

1. Deploy every 30 minutes: Redux

   So, it's true, they really do deploy (up to) every half an hour. Pretty cool; In answer to the two questions from the previous entry: Why do they do it? Because they can. Because, for them at least, 'Release early, R...

   Secret Plans and Clever Tricks - 25 Aug 2005, 21:32