hm… I think I’ve had uname deploys roll back before (most of my deploys have used uname). Although a rollback based on git alone wouldn’t be possible, all the deployed files are kept around in S3 I believe.
I’ve had an issue previously where all my deploys would fail because of an out-of-memory error, and the roll-back would also fail (so I was stuck in down-time). I eventually found that I could get deploys to work again simply by terminating the compute instance in the ec2 console, waiting for a new instance to be auto-started, and then re-deploying right away. For a while I had to do that every time, but it mysteriously stopped being an issue a week or so ago.