First hand: Scaling under load spikes with Rackspace Cloud Server

I've run TweetStats on EC2 for the past year using the great scalr.net to provision new instances.

It's expensive as heck (the smallest instance scalr allows is an EC2 large), but it works great.

Scalr monitors one of several stats you can specify (RAM, CPU, queue size, HTTP response) and will spin up instances (you can specify a max) as necessary.

Granted the app has to be built to support such multiple app servers, which may be easier in some cases than others.

Sounds like the cloud server was rather inconvenient, couldn't handle immediate load, and required your manual intervention. Always tough to predict traffic spikes from Twitter. ;)

"It looked like a few hours of data was lost from the database for some reason"....eh, what? Was this the application's database? Losing a couple hours of data because of a traffic spike is not something I would find remotely acceptable.

Thought I'd give people a first hand look at dealing with this type of situation. Would really like to know how other people would have handled it.