Graphile Migrate

Tracking Changes

  • GM has an .gmrc file, which allows us to define hooks. We can therefore define a hook that will run pg_dump on the shadow db after every migration. If we track this file in git, then we can see a clear history of what each migration has done.

Development

  • In Development, aside from the main database, there is a shadow db which is used internally by graphile-migrate and is mainly for testing consistency of the migrations, among other minor tasks
  • the shadow database gets reset frequently
  • commit, uncommit, watch and reset are development-only commands.

Flow

  1. We write our new idempotent migration in current.sql
    • any seed data we have can be placed at the bottom, to be removed prior to committing.
  2. when ready, we run commit
    • this should only be done immediately before the branch is merged into master (see README###Collaboration). This it to ensure commits are linear.
    • all sql in current.sql is removed and catalogued into committed/
    • the shadow database is dropped, and recreated by running all migrations. This is to ensure that the migration works without a hitch.

Pitfalls

  • when dropping tables, schemas, and functions, we should use CASCADE
  • use idempotent commands whenever possible.
    • ex. DROP ____ IF EXISTS
    • ex. CREATE OR REPLACE FUNCTION
    • The initial migrations don't need to be idempotent if this migration starts off by dropping all schemas (which implicitly drops all tables attached to those schemas)

Backlinks