Follow-up from "Integration tests for GitLab Geo"

The following discussions from !3294 (merged) should be addressed:

• @grzesiek started a discussion: (+3 comments)

  It is far from perfect, this should be probably handled by replicate-geo-database command, but this workaround seems to be the most boring solution for this problem.

• @grzesiek started a discussion: (+2 comments)

  Heads up: we are not testing geo setup with SSL!

• @nick.thomas started a discussion: (+6 comments)

  It's not obvious at all because postgres replication is fast and this is slow, but there's a race here. It might become more obvious when we start asserting on the repository contents (since repository sync is much slower).

  I suspect we can handle it at that point, though. Something like the AJAX wait_for_requests where we sleep until the secondary reports that it has synchronized everything.

• @nick.thomas started a discussion: (+3 comments)

  How do we verify that this action succeeded? What happens if the HTTP request kicked off by click_button returns a 500 error?