Fixing the expected failures on starting processes with farms could have allowed us to catch this. I added some new tests with mocks to check that the correct get_default_farm is called, and opened https://redmine.teklia.com/issues/7025 to deal with all the other expected failures.
