Shared Database Integration
also called Integration Database
Two or more applications reading and writing the same database directly — the most damaging integration pattern and the hardest to unwind.
It is chosen because it is the fastest thing to do: the data is already there, no API is required, and a join across both systems' tables works immediately.
What it costs is everything a boundary would have provided. The schema becomes an undocumented public contract, so no table can be changed without coordinating every application that reads it. Neither application can enforce its invariants, because the other can write directly. Deployment is coupled, since a schema change must land with changes to every consumer. And there is no observability of who reads what, so nobody can determine the blast radius of a change.
The practical unwinding path: identify readers using database audit or query logs, publish an API or a CDC-derived event feed covering their needs, migrate them one at a time behind that, then revoke direct access — which is the step that must actually happen, or the pattern returns.