Mmmm well that’s a granularity argument isn’t it? What is a service for you? Is it the same as mine? For me a business-level service might well be built out of a bunch of smaller services (how small those services are might be worth further discussion). Does that fit with your view?

“I’d also submit that designing services database-first (while not explicitly stated in the post, but is implied)…..”

Hmmm I wasn’t meaning to imply any such thing. Clearly schema is dictated by what you want to do and is design driven. It would be instructive for me (always looking to improve my communications) if you could point out what I said that caused you to see such an implication?

“not the least due to the duplication of schema elements (or master data definitions if you will) across databases.”

Hmmmm, do you believe strongly in normalized databases? There are lots of real world cases where we duplicate elements for speed but in this case I’m not advocating that behaviour. In line with my comment re: services underlying business services above, I would expect some service to own a chunk of data and that chunk of data would be accessed by other services via the owning services APIs (whatever form they take).

i.e. We’re divorcing the how of our storage of the data which will require one schema from the how we pass the data between services which can be another schema. Thus there’s no reason why a service couldn’t internally use a flat file system for storage of data.

There’s lots more to discuss here but I’d like to get a better grip on what you’re thinking first….

“We end up with a system of lots of discrete services each wrapping up their own data storage.”

My experience has been slightly different. I find the result to contain relatively few services, each one being of a business-level granularity. I’d also submit that designing services database-first (while not explicitly stated in the post, but is implied) has a high probability of leading us to a poor business-level service decomposition, not the least due to the duplication of schema elements (or master data definitions if you will) across databases.

What are your thoughts?