[yummybear]

One of our problems in scaling is not starting new instances, but rather to ensure that new instances have the latest version of the platform.

Also, we are running a fantasy soccer game with trading windows, so the server load near the trading window closure time (the last hour or so) is orders of magnitude higher than average load. Ideally we would scale up when the load demands it, but with the platform versioning issue unresolved we need to run our platform at "full scale" almost constantly (or scale manually which is a really bad idea when someone forgets to scale back up again for the weekend).

We're running in a WISC setup.

[mnutt]

What if every time you did a deploy, afterwards you kicked off a job that wrote a snapshot and brought up instances based off that. Then at any point you'd only ever be at most a deploy behind.