diff --git a/umbratrans_web/tests/test_postgres_repository.py b/umbratrans_web/tests/test_postgres_repository.py
index bfa2d23..63e2a2e 100644
--- a/umbratrans_web/tests/test_postgres_repository.py
+++ b/umbratrans_web/tests/test_postgres_repository.py
@@ -820,6 +820,12 @@ def _drop_tenant_schemas(dsn: str) -> None:
     schemas themselves would survive, and a hundred runs would leave a hundred
     schemas for organizations that no longer exist. Only reached when the guard
     above established the database held nothing to begin with.
+
+    One schema per transaction, and that is load-bearing rather than tidy. Dropping
+    several hundred schemas in a single transaction exhausts the lock table and
+    fails with "out of shared memory ... max_locks_per_transaction" - a message
+    that reads like a PostgreSQL defect and is not one. It is the same constraint
+    the 0005 rollback documents, learned here first.
     """
 
     pool = ConnectionPool(dsn, min_size=1, max_size=2, open=True, kwargs={"row_factory": dict_row})
@@ -830,6 +836,7 @@ def _drop_tenant_schemas(dsn: str) -> None:
             ).fetchall()
             for schema in schemas:
                 connection.execute(f'DROP SCHEMA IF EXISTS "{schema["nspname"]}" CASCADE')
+                connection.commit()
     finally:
         pool.close()