标签：#rebac

ReBAC (relationship-based access control) decides access based on how entities are connected to each other, rather than on a role attached to the user. Nowhere is it written that you can see that repository. You see it because a chain of relationships leads you there. It's a model I like, and I want to say that up front, because what follows isn't a criticism of ReBAC. I spent a few weeks integrating OpenFGA into a prototype to use it properly: declarative model, sixteen test scenarios, a hundred and twenty assertions running offline in two seconds with no database and no application. It worked well. I removed it anyway. Not because of a bug, and not because of check latency. I removed it because none of the tools I tried gives me a usable answer to the second question every application asks. The check is fast. The list isn't. "Can this user see this object?" resolves in milliseconds. The problem is that the first screen after login is almost always a list. And "which objects can this user see?" looks like the same question reversed, but it isn't: nowhere is it recorded which objects are reachable, which is the point of ReBAC seen from the other side. In the first case you hold two things and walk the graph from one to the other. In the second you hold only the user, and the set of possible answers is everything that exists in the system. Three routes, and where each one stops Filter afterwards. Normal query with its normal pagination, then you send the twenty-five ids to the service and drop the ones that don't pass. The result is correct, but the total at the bottom of the page is the one from before the filter, so it's a lie. And the user with access to a small slice gets three rows out of twenty-five. Filter first. You ask the service which objects the user holds that permission on and hand them to the database. Except the list arrives whole. There's no real pagination to draw twenty-five from. It ends up as an IN with thousands of identifiers. A local index. Yo