0

RLS primer: open until you close it

Securityused in ~95% of apps

You just spent three concepts making DevLog's posts table refuse bad data — the wrong type, a missing value, a duplicate. None of that protects a single row from the wrong person. The instant you create a table in the public schema, it is a plain Postgres table with no notion of ‘this row belongs to alice.’ Nothing about CREATE TABLE turns that on for you, and — contrary to what a lot of tutorials imply — nothing about writing the SQL yourself turns it on either.

Core idea. A table you create with raw SQL starts with row-level security OFF: every row is visible to anyone the Data API lets reach the table at all, is_public column or no is_public column. Turning on ROW LEVEL SECURITY flips that to the opposite extreme — zero policies means zero access, for everyone except a handful of roles that bypass it entirely. There is no safe middle state until you write an actual policy, which is a different concept (52+). This one is only about knowing which of the two walls you are standing against right now.
Playground, this isn't a screenshot
-- you just ran: create table posts …
Aalice's appselect * from posts
#1Why I switched to Postgres
#2Draft: my RLS notesprivate
#3Ships on Fridays
Bbob's appselect * from posts
#1Why I switched to Postgres
#2Draft: my RLS notesprivate
#3Ships on Fridays
Bob is reading Alice's private draft. This is the leak.

The table is wide open. The same query hands every row to both identities — Bob is reading Alice's private draft. Nothing is stopping him.

One query. Two identities. Different answers. No where clause chooses who sees what — the policy does, inside Postgres. You do not yet understand howthat clause works; that is Tiers 9–10. This concept exists to make you want the answer.
↑ play a little first, then dig in