Thread by @mathiasverraes

I should probably clarify this one before people start calling me names :-)

https://twitter.com/mathiasverraes/status/1431555431193878531?s=20

I probably used the term monolith a bit liberally there. But imagine you have a large chunk of your system that is very complex, because it has to deal with something with a lot of variability, in this case, different ways of doing things in different countries.

Naive strategic design here is to separate by domain (billing, invoicing, debt collection). But Bounded Contexts don't need to align with subdomains. That separation is still complex because each context needs to know country-specific things.

Instead, it could be a context for "Belgium billing, invoicing, debt collection", another for "Germany billing, invoicing, debt collection" etc.

How to get there? The constraints on our strategy are that everything is tightly coupled, that we risk breaking everything. How do we limit risk? For example, by not solving for all countries, but just for one.

We copy the whole thing, involve the Belgium experts, get rid of everything that doesn't apply to Belgium, clean it up, create better separation, set up CI/CD for Belgium, get it in prod.

Now we do it again for Germany. We start noticing that some things are in fact reusable, say invoice pdf generation, and most of debt collection. Let's extract that from the Belgium BC, reuse it in the Germany BC.

By the third or fourth time, we're a lot more comfortable, we have discovered where the variability is, we can do other countries a lot faster now.

Or, the whole idea failed with Belgium, but at least none of the other countries were affected.

It may sound like a lot of work, and it is. But what's the alternative? Nobody dares to touch the original monolith, so it just gets worse, and the three previous attempts at rebuilding from scratch have failed.

Duplicating everything is dangerous, if you think of it as an upfront design decision. But it is not if you think of it as a disposable, reversible experiment.