Just saw this post tweeted so here’s an update. I have the ultimate “drop everything” script that now uses the Inspector to pull everything in – so far I’ve tested with Postgresql and MS-SQL. Though currently its doing just all the tables and related objects – I notice Pavel’s script does views and routines too. http://www.sqlalchemy.org/trac/wiki/UsageRecipes/DropEverything

Method declaration:

CREATE OR REPLACE FUNCTION drop_db()
RETURNS void AS
$BODY$
DECLARE
objname character varying;

BEGIN

FOR objname IN SELECT table_name FROM information_schema.views WHERE table_schema = ‘public’ LOOP
EXECUTE ‘DROP VIEW ‘ || objname ;
END LOOP;

FOR objname IN SELECT table_name FROM information_schema.tables WHERE table_schema = ‘public’ LOOP
EXECUTE ‘DROP TABLE ‘ || objname || ‘ CASCADE ‘;
END LOOP;

FOR objname IN SELECT sequence_name FROM information_schema.sequences WHERE sequence_schema = ‘public’ LOOP
EXECUTE ‘DROP SEQUENCE ‘ || objname || ‘ CASCADE ‘;
END LOOP;

FOR objname IN SELECT routine_name FROM information_schema.routines WHERE routine_schema = ‘public’ LOOP
CONTINUE WHEN objname = ‘drop_db’;
EXECUTE ‘DROP FUNCTION ‘ || objname || ‘() ‘ ;
END LOOP;

END;
$BODY$
LANGUAGE ‘plpgsql’ VOLATILE
COST 100;

sqlalchemy drops tables in order of dependency, where by “dependency” we mean foreign key dependencies..you can declare these explicitly using the ForeignKey construct. But table reflection already uses the PG system tables to get this information automatically (we used to use information_schema but it proved to be too slow and quirky). this feature can be used to drop all tables in the correct order like so:

engine = create_engine(“posgres://user:pass@host/db”)
meta = MetaData(engine)
meta.reflect()
meta.drop_all()

however, it currently doesn’t cover sequences except for those which were created implicitly using SERIAL columns…we might add support for reflection of external sequences eventually.

- mike