Migrate the schema

What runs in this step

Schema migration reconstructs your database structure on the destination before any row is copied. There is no pg_dump in the browser, so the schema is rebuilt from queries against information_schema and pg_catalog: extensions, types, tables, constraints, indexes, sequences, functions, triggers, and RLS policies. See schema replay for how the DDL is derived.

DDL dependency order

Objects are created in an order that satisfies their dependencies:

extensions → types → tables (no FK) → FK constraints
→ indexes → functions → triggers → RLS policies → publications

Foreign keys are added after every table exists, so a posts.author_id → users.id reference does not fail because users was created later. Enum and composite types are created before the tables that use them.

Gap-fill

Replay covers the objects introspection can see. A second pass diffs the destination against the source and fills gaps — a missing DEFAULT, an ungranted EXECUTE on a function, a sequence not yet owned by its column. Each statement is idempotent, so re-running the step after a failure does not double-create objects.

Common failures

• type "..." does not exist — a table referenced an enum or composite type that was not created first. Re-run the step; the ordered pass creates types before tables.
• relation "..." already exists — the object is already on the destination from a previous attempt. Safe to ignore, or drop the partial schema and start clean.
• permission denied for schema ... — you are connected with a role that cannot create in that schema. Connect as postgres (the direct connection string, not a restricted role).

Each generated statement maps to a migration file you can inspect before it executes.

Verify

After the step completes, confirm the table count matches the source:

select count(*)
from information_schema.tables
where table_schema = 'public';

Compare against the same query on the source. RLS policies:

select schemaname, tablename, policyname
from pg_policies
where schemaname = 'public';

Next

With the structure in place, move on to migrating data — rows are inserted in FK-safe order and sequences are resynced afterward.