This was originally posted on my own blog here
Recently at work I've spent some time working on a scaffold project that we'll be using for API projects that we'll be building.
For a number of reasons we've decided to use Python as our backend language and Flask as our API framework. One of the things I love most about Flask is that it's very unopinionated and let's you build what you want, pretty much how you want.
One of the features I've been working on is an audit log. For data integrity purposes we decided to use an enum field for the
event type value in both the code itself and also in the database. Like many Flask applications, we're using SQLAlchemy as an ORM and Flask-Migrate to automatically create Alembic migrations. Using a SQLAlchemy model with the field type set to an enum equal to the enum used in the code, I had expected Flask-Migrate to automatically create a new migration any time we added values to the EventType enum class, however this is not the case.
After some searching around, I discovered that this is a known issue with Alembic and that migrations for enum changes have to be created manually. I've got a sample migration that I created manually below as well as some steps so you can see how to handle adding or removing a value from an enum in Flask. Note that this migration is specifically written to work with PostgreSQL as that is the database engine that we use.
- Make changes to the Enum in the relevant model
Create an empty migration file
flask db revision -m 'Add Logout_Success to AuditEvent'
Populate the new migration with code to create the changes, note you will need to add values for the existing and new options ensuring to keep the revision and down_revision numbers that already exist in the new migration file
""" Add Logout_Success to AuditEvent Revision ID: 08720b8a9d11 Revises: 810eac468f83 Create Date: 2020-03-25 12:19:09.432635 """ from alembic import op import sqlalchemy as sa # revision identifiers, used by Alembic. revision = '08720b8a9d11' down_revision = '810eac468f83' branch_labels = None depends_on = None # Enum 'type' for PostgreSQL enum_name = 'auditevent' # Set temporary enum 'type' for PostgreSQL tmp_enum_name = 'tmp_' + enum_name # Options for Enum old_options = ('LOGIN_SUCCESS', 'LOGIN_FAIL') new_options = sorted(old_options + ('LOGOUT_SUCCESS',)) # Create enum fields old_type = sa.Enum(*old_options, name=enum_name) new_type = sa.Enum(*new_options, name=enum_name) def upgrade(): # Rename current enum type to tmp_ op.execute('ALTER TYPE ' + enum_name + ' RENAME TO ' + tmp_enum_name) # Create new enum type in db new_type.create(op.get_bind()) # Update column to use new enum type op.execute('ALTER TABLE audit ALTER COLUMN event_type TYPE ' + enum_name + ' USING event_type::text::' + enum_name) # Drop old enum type op.execute('DROP TYPE ' + tmp_enum_name) def downgrade(): # Instantiate db query audit = sa.sql.table('audit', sa.Column('event_type', new_type, nullable=False)) # Convert LOGOUT_SUCCESS to LOGIN_SUCCESS (this is just a sample so may not make sense) op.execute(audit.update().where(audit.c.event_type == u'LOGOUT_SUCCESS').values(event_type='LOGIN_SUCCESS')) # Rename enum type to tmp_ op.execute('ALTER TYPE ' + enum_name + ' RENAME TO ' + tmp_enum_name) # Create enum type using old values old_type.create(op.get_bind()) # Set enum type as type for event_type column op.execute('ALTER TABLE audit ALTER COLUMN event_type TYPE ' + enum_name + ' USING event_type::text::' + enum_name) # Drop temp enum type op.execute('DROP TYPE ' + tmp_enum_name)
Hopefully this will be of use to someone.