Hacker News Reader

Comment by dragonwriter | Hacker News Reader

> PostgreSQL Row Level Auth is horrible to use when you scale up compared to implementing auth on the app level.

In what way?

When you have RBAC based on multiple different conditions, e.g. "does user have XYZ?" you will have to architect your entire app around putting these claims into JWT so the GraphQL engine can properly map those onto RLS. In a traditional app auth is on the app level and defined in code, not a mix of code, SQL, JWT claims and Hasura/Supabase configs.

> When you have RBAC based on multiple different conditions, e.g. "does user have XYZ?" you will have to architect your entire app around putting these claims into JWT so the GraphQL engine can properly map those onto RLS.

If it's RBAC, why am I not just giving creating per-user postgres roles inheriting from appropriate roles for the relevant features?

> In a traditional app auth is on the app level and defined in code, not a mix of code, SQL, JWT claims and Hasura/Supabase configs.

Doing authz via DB-side permissions attached to per user DB accounts is an equally, and more well-established, traditional app approach.

I just wrote an article on this (adding custom claims to the JWT) for Supabase projects: https://dev.to/supabase/supabase-custom-claims-34l2

You don't necessarily have to put those claims into a JWT (other than the user ID). You can just store the "XYZ" that the user has into a table.

That is definitely true (and probably easier than doing custom claims), but the reason I use claims is that it saves doing a join against that extra table inside your RLS policies, which can sometimes have a big performance impact. Those claims are also available on the client side, too, saving a round trip to the server if you need to check the claims.

Reply on news.ycombinator.com