All of these are solved problems in the large, when you’re on EC2 and have an army of developers to write serialization/deserialization plumbing, implement data migrations, and operate your Kafka clusters or whatever.

But how can this sort of system be built in the small?