How 37 Signals Deploys Their Software

Written by



I had a great time last Saturday at Raleigh Ruby Camp and met a host of Ruby on Rails developers working on interesting things.

One of the discussions that occurred was lead by Mark Imbracio of 37 Signals. I’ve posted the audio below and have linked to some of the topics that came up in the discussion. The audio quality and production isn’t great (you may need to really turn your volume up), but I got quite a bit out of the discussion and you might as well.

Links

[audio:http://buzzwordcompliant.s3.amazonaws.com/mark-deployment-talk.mp3]

Download MP3

[update 10/23 - fixed broken link to AWS]

3 Responses to “How 37 Signals Deploys Their Software”

  1. Mark Imbriaco

    For the record, I said a couple of things that I realized after the fact were incorrect:

    1. InnoDB hot backup doesn’t block writes to the database while it’s running, but it is very I/O intensive so your slave can definitely backlog while it runs. I’ve seen better performance with LVM snapshots using mylvmbackup.

    2. We don’t run the Phusion patched Ruby, at least not completely. We use their backport of the security patches and some GC patches from another source (http://railsbench.rubyforge.org/svn/trunk/railsbench/ruby186gc.patch), but we don’t use their copy-on-write friendly garbage collector.

  2. John Meredith

    The “Amazon Web Services” link should be .com instead of .org

  3. Michael Buckbee

    Thanks John, I fixed it. I should have caught it myself, it’s not as if Amazon’s some entity that’s just giving away S3 and EC2 time.

    - Mike

Leave a Reply