This is cool. Another alternative would be to use postgres savepoints, which are like transactions inside transactions, as a wrapper around each migration. Yo

This is cool. Another alternative would be to use postgres savepoints, which are like transactions inside transactions, as a wrapper around each migration. You can do the same thing - set lock_timeout and catch errors when those values are exceeded, and try the transaction again. Provide an option to run the handful of operations that can't be run inside a transaction as an escape hatch, and then you can retain the ability to run migrations inside transactions, which is usually a good thing.