That technique used for migrations will fail with a real production app where you want to avoid downtime. The process outlined is:
1. deploy new code to prod
2. run migration on prod via a one-off dyno instance
However, in step 1 you are pushing code that relies on new tables/fields that don't yet exist, causing errors and likely downtime for the app.
To avoid this, you can instead have Alembic generate SQL from the migration and then apply that SQL to the prod DB directly prior to deploying the updated code.