Migrations

Database migrations are handled by this package. All migrations should be created in separate files, following a starndard naming convetion

The migrations.go file defines an array of migrate functions that are called by the gormigrate helper. Each migration function should perform a specific migration.

Creating a new migration

Create a migration ID based on the time using the YYYYMMDDHHMM format. Example: August 21 2018 at 2:54pm would be 201808211454.

Your migration's name should be used in the file name and in the function name and should adequately represent the actions your migration is taking. If your migration is doing too much to fit in a name, you should consider creating multiple migrations.

Create a separate file in pkg/db/ following the naming schema in place: <migration_id>_<migration_name>.go. In the file, you'll create a function that returns a gormmigrate.Migration object with gormigrate.Migrate and gormigrate.Rollback functions defined.

Add the function you created in the separate file to the migrations list in pkg/db/migrations.go.

If necessary, write a test to verify the migration. See test/integration/migrations_test.go for examples.

Migration Rules

Migration IDs

Each migration has an ID that defines the order in which the migration is run.

IDs are numerical timestamps that must sort ascending. Use YYYYMMDDHHMM w/ 24 hour time for format. Example: August 21 2018 at 2:54pm would be 201808211454.

Migration IDs must be descending. If you create a migration, submit an MR, and another MR is merged before yours is able to be merged, you must update the ID to represent a date later than any previous migration.

Models in Migrations

Represent modesl inline with migrations to represent the evolution of the object over time.

For example, it is necessary to add a boolean field "hidden" to the "Account" model. This is how you would represent the account model in that migration:

type Account struct {
  Model
  Username string
  FirstName string
  LastName string
  Hidden boolean
}

err := tx.AutoMigrate(&Account{}).Error
if err != nil {
...

DO NOT IMPORT THE API PKG. When a migration imports the api pkg and uses models defined in it, the migration may work the first time it is run. The models in pkg/api are bound to change as the project grows. Eventually, the models could change so that the migration breaks, causing any new deployments to fail on your old shitty migration.

Record Deletions

If it is necessary to delete a record in a migration, be aware of a couple caveats around deleting wth gorm:

1. You must pass a record with an ID to gorm.Delete(&record), otherwise ALL RECORDS WILL BE DELETED

2. Gorm soft deletes by default. This means it sets the deleted_at field to a non-null value and any subsequent gorm calls will ignore it. If you are deleting a record that needs to be permanently deleted (like permissions), use gorm.Unscoped().Delete.

See the gorm documentation around deletions for more information

Migration tests

In most cases, it shouldn't be necessary to create a test for a migration. However, if the migration is manipulating records and poses a significant risk of completely borking up important data, a test should be written.

Tests are difficult to write for migrations and are likely to fail one day long after the migration has already run in production. After a migration is run in production, it is safe to delete the test from the integration test suite.

The test helper has a couple helpful functions for testing migrations. You can use h.CleanDB() to completely wipe the database clean, then h.MigrateTo(<migration_id>) to migrate to a specific migration ID. You should then be able to create whatever records in the database you need to test against and finally run h.MigrateDB() to run your created migration and all subsequent migrations.