The paradox of event sourcing

The hardest part for me is knowing when to use this. It creates a lot of friction for a small application, but all applications start small. Moving to an event-sourced architecture when your application (and team) is no longer small feels like a big undertaking that could be hard to justify.

Dave Copeland, Event Sourcing in the Small

Once an application is big enough to need it, it’s already hard to introduce it. But, it’s too much trouble to start an application with this architecture. Maybe this is corollary to “most things are easy/workable on small teams/applications”?

A few problems that Dave ran into building a small event-sourced data model were in deriving the domain models (he called them projections) from the event data model. It’s possible that there’s a sweet balance point between rolling this kind of data flow behavior by hand and building an entire framework around capturing events that are transformed for various consumers to their specific domain model needs. I haven’t seen it yet.

I haven’t kept up with Datomic, but the interesting about it a few years ago was that it was sort of event sourcing as a database. Data producers store events to it (in a format that strongly resembled RDF triples). Consumers used data flow queries to define how to transform and scope that data to their needs. It also had a pretty sweet time-travel story. (I’m always a sucker for a good time-travel story.)

If well-considered boundaries and excellent operational tooling are the enabling factors of a services architecture, what are the enabling factors of an event-modeled architecture?