At Trade Me we created scripts for all database changes. Stored procedure changes were included in the package and executed automatically against the database as the code changes were deployed.

Changes to table columns, indices, etc were also scripted, but were normally manually executed (given that the site would need to be taken offline to make these sort of changes that wasn’t really a constraint).

In most of those places, a deployment was a major multi-step (multiple chance of failure) process. Which meant… deployment happened infrequently, and was big (with greater chance of something being wrong). Which led to … lots of post-deployment analysis (whose fault was it?) meetings.

It takes a while to sort out a simple deploy process, but it is so worth it! But one thing I havent really seen done all that well, is synchronising database updates with code deployments. How does everyone do it?

I even use it on that project I’ve been working on lately.

1. build the app (it’s .NET) using NANT (one command)
2. build a deployment package (1-3 commands, one each for SYSTEST, UAT and LIVE)
3. hand it off to the production people, who actually do the deployment. Sadly, we have to do it this way, but for them it’s a case of unzip then run nant. Easy.

Was a bit of a mission to get the scripts written, but it’s paid off. What was a 1-2 day process can now be done 3-5x/day if we have to.

I even have Capistrano set up to deploy static html sites from Subversion. The latest version also handles Git repositories so it may be time to jump on the latest SCM bandwagon ;-)

I have also just added some Capistrano extensions so that I can now to ‘cap staging deploy’ and ‘cap production deploy’ to deploy to different servers for testing, etc. I have had too many problems in the past forgetting steps, not uploading files and generally screwing things up to not use automation.