How do you deal with test record leaks?

I've written about test leaks before, they are a pernicious, hard to debug thing that must be avoided at all costs.

Today I got a weird failure on CI and while investigating, it became apparent that some other spec is leaving records in the test DB, working outside DB transaction.

I know of two common ways to sidestep transactionality:

1. Creating records in a before(:all) block. Sometimes this may be necessary to save on repeat setup time, but look into TestProf's before_all helper for a safer approach.
2. Creating records in inline code outside example context:

describe "#some_method" do
  # the two users fabricated in this hash will leak.
  {
    regular: FactoryBot.create(:user), 
    elevated: FactoryBot.create(:super_user),
  }.each_pair do |permission_level, record| 
    it "works for '#{#{permission_level}}'" do
      # some asserts ...
    end
  end
end

To avoid cases like this in the future, I recommend setting up a simple hook that checks the test DB has no records of popular models used in your app.
Place in a file location that gets loaded by spec helper, usually spec/support/hooks/:

class RecordsLeftInTestDBError < StandardError; end

RSpec.configure do |config|
  config.after(:suite) do
    counts = {
      some_records: SomeModel.count,      
    }

    counts.values.sum.positive? &&
      # I'd prefer `abort` to `raise`, but for some reason
      # it  does not fail the RSpec run when called from an
      # after-suite hook like this.
      raise(
        RecordsLeftInTestDBError.new(
          "Leak detected!\n" \
          "Records in the test DB after running the suite:\n" \
          "#{counts}",
        ),
      )
  end
end