Well, to be fair maybe they didn't have enough notice to spin up additional instances and get replication completed beforehand but I highly doubt that considering how high profile this AMA is. The US President doesn't probably drop in on people / or websites :P

Yeah GAE might work better but they're a bit out of reach mostly due to the technologies that you are required to use. Any migration at that scale is a seriously massive matter.

I guess the funny thing is that you can't plan to grow that big. Usually it just happens suddenly (facebook/reddit style) and then you've got to make some tough architectural calls.

It's funny how folks are like "That's a problem I'd like to have..". I'm not sure they're very happy with this problem at this very moment.. sigh

I don't know about websites, but the US President drops in on lots of places (restaurants for instance) as a matter of security. From what I understand with interviews, it's often not uncommon for the interviewer to get very little notice either at times.

AWS Elastic Beanstalk will do the same thing.

Beanstalk has nothing to do with the database, which is the point of contention. Spinning up more app servers only helps with serving signed in users.

You can configure beanstalk to scale your database out as well.

Scaling out database is nothing at all like spinning up stateless app servers.

Well there is no magical database solution where you can do that and I have to make decisions on what I can buy today not on what may be magically developed in the future.

In this respect, application partitioning with a deployable schema is not a bad solution for what I do.

AFAIK Reddit aren't using it. If anyone can clarify I'd love to hear more.