Database Migrations

Any changes made to the database models must include migrations that can modify the database state from one state to another.

Whilst we use Sequelize as our ORM layer, we do not use the migration tooling it provides - we have our own.

Creating migrations

Filename

A migration is provided as JavaScript in the directory forge/db/migrations. Its file name must follow the pattern:

YYYYMMDD-nn-description.js
  • YYYYMMDD - the date the migration is added
  • nn - a two digit number
  • description - a name for the migration

For example 20220204-01-add-billing.js.

This ensures the migrations have a natural order to be applied. The nn part of the name allows multiple migrations to be added on the same day but is kept in the right order.

Structure

The migration code should use the following layout:

module.exports = {
up: async (context) => {
// Apply the migration
},
down: async (context) => {
// Remove the migration
}
}

The up function applies to the migration. This can be to create new tables, add columns to existing ones - whatever is needed.

The down function reverses the migration. It should restore the database back to how it was prior to the migration.

The context argument is an instance of Sequelize.QueryInterface that can be used to perform operations on the database.

Applying migrations

Migrations are applied automatically at the start of the FlowFuse application. Down migrations are not yet supported.

Considerations when writing migrations

Whilst migrations give us the ability to make changes to the database, they must be used with great care. A failing migration will prevent the platform from starting and may require manual intervention. Everything should be done to avoid that from happening.

Certain types of migration need particular guidance and care over.

Adding constraints

If a migration is adding a new constraint to an existing table, you need to consider very carefully what impact that could have on an existing system with real data.

For example, adding a new 'unique' constraint where you cannot guarantee that constraint hasn't already been broken. What strategy will you use to guard against that or to help recover from it? What additional testing is needed for the migration to verify its behavior in those situations?

The preferred method to add a new unique constraint is by adding a new index to the database. This is because SQLite doesn't provide a way to alter columns that doesn't involve dropping the whole table and triggering any cascade triggers.