I debated filing this against database_cleaner-active_record, but after digging in a bit I think it would need a more comprehensive fix.

I have a Rails app running in a Docker stack, where under test, the database is configured the old-fashioned way in database.yml rather than with ENV['DATABASE_URL'] -- or rather, both are supported, but if DATABASE_URL isn't set, it falls back to a hard-coded value.

test:
  url: <%= ENV['DATABASE_URL'] ||'postgres://root:root@db/framework-test?pool=5' %>

I wanted a safeguard that would work both with the Docker stack in CI, or for a developer setting up a test database locally or on some other database server, so I assumed I had to set url_allowlist. My first attempt was this lambda:

  DatabaseCleaner.url_allowlist = [
    ->(url) { URI.parse(url).path.end_with?('framework-test') }
  ]

This blows up, though, because the URL that gets passed here is ENV['DATABASE_URL'], and URI.parse(nil) fails.

My next concern was that DatabaseCleaner would truncate the database at ENV['DATABASE_URL'] (if it exists) rather than the one ActiveRecord is actually using in tests. Thankfully database_cleaner-active_record is smarter than that, and it's using the ActiveRecord connection.

But it seems like if that's how it actually works, Safeguard should somehow hook into that, rather than relying on ENV['DATABASE_URL']. Maybe instead of running once before all cleaners, it should be run per-cleaner, and the cleaner implementations should be responsible for providing the URL they're actually using?