Version should be written after every migration

[sqlalchemy-bot]

Migrated issue, originally created by kaukas (@kaukas)

Consider two migrations: A and B; A succeeds while B fails due to some error. After database migration you get A fully applied and B partially applied or not applied at all.

However, the migrations table remains empty. Once I fix B I should be able to rerun just B but A is executed as well.

[sqlalchemy-bot]

Michael Bayer (@zzzeek) wrote:

Consider two migrations: A and B; A succeeds while B fails due to some error. After database migration you get A fully applied and B partially applied or not applied at all.

this is because your database doesn't support transactional ddl.

However, the migrations table remains empty. Once I fix B I should be able to rerun just B but A is executed as well.

alembic does do this for when transactional DDL isn't on, on line 212 of migrations.py:

        change(**kw)
        if not self.impl.transactional_ddl:
            self._update_current_rev(prev_rev, rev)
        prev_rev = rev

so then I checked what env.py does, by default it's this:

try:
    with context.begin_transaction():
        context.run_migrations()
finally:
    connection.close()

so that's the problem right? wrong, begin_transaction() also checks transactional DDL, and doesn't actually run a transaction:

def begin_transaction(self):
    if not self.is_transactional_ddl():
        @contextmanager
        def do_nothing():
            yield
        return do_nothing()
    elif self.is_offline_mode():
        @contextmanager
        def begin_commit():
            self.get_context().impl.emit_begin()
            yield
            self.get_context().impl.emit_commit()
        return begin_commit()
    else:
        return self.get_bind().begin()

so I am not seeing how your error is occurring, unless your tranasctional_ddl flag is incorrectly set. The version table is written after each migration and the connection is in autocommit mode when transactional_ddl is false. I need a lot more detail here including database in use as well as some log output.

[sqlalchemy-bot]

Iuri de Silvio (@iurisilvio) wrote:

I had this issue with sqlite too.

[sqlalchemy-bot]

Michael Bayer (@zzzeek) wrote:

the flag was not set properly for SQLite (note http://bugs.python.org/issue10740) or MySQL, fixed in 3d1ed96

[sqlalchemy-bot]

Changes by Michael Bayer (@zzzeek):

• changed status to closed