[Multiple databases] How to access Heroku follower database and manage upgrades

rubyonrails-talk
1

1/ What is the proper way to let a single-DB Heroku Rails 7 app start accessing a follower DB after it has been provisioned in Heroku? I understand we can continue to get the config of the leader DB from the DATABASE_URL environment variable; as for the replica/follower database, should we specify its URL manually like below? One issue I see is that the environment variable name will change when upgrading the DB (see point 2 below).

# config/database.yml
production:
  primary:
    adapter: postgresql
    # no URL specified: use DATABASE_URL
  primary_replica:
    adapter: postgresql
    url: <%= ENV["HEROKU_POSTGRESQL_XXX_URL"] %>
    replica: true

Is there anything else we should know, e.g. should we start to use different Rails commands when running migrations, or specifying database users somewhere? (Not really, if I understand the guide correctly?)

2/ Once such a follower starts being accessed by the app, how to upgrade the leader and follower databases when a new Postgresql version becomes available? I understand that usually this is done by follower changeover. Should the procedure be like below?

  1. In Heroku, provision a first new follower with the latest Postgresql version (heroku addons:create)
  2. In Heroku, place the app in maintenance mode (heroku ps:scale worker=0 + heroku maintenance:on)
  3. In Heroku, promote A to become the primary (heroku pg:unfollow + heroku pg:promote)
  4. In Heroku, provision a second new follower with the latest Postgresql version (heroku addons:create)
  5. In Rails, update config/database.yml > primary_replica > url to use the new DB URL of this second follower, deploy the change
  6. In Heroku, disable maintenance mode (heroku ps:scale worker=1, heroku maintenance:off)
  7. After checking everything in the app works as expected, decommission the 2 previous databases (heroku addons:destroy)

We are interested in this to protect against writing to the DB for operations that should be read-only, such as when running reports (*) or simply displaying content, and to reduce the load on the primary DB. I suppose there can be a lot of edge cases in deciding which DB should be accessed as sometimes a DB might be modified even in GET requests (e.g. to store some analytics/logs), so at first we would be satisfied with always using the primary DB, and only accessing the follower when manually specified.

(*) As suggested in the Blazer gem for instance.