We may have some init-order issues that are showing up as flaky runs, because we randomize the order of mongo initializers, per SERVER-40972.

The initializer order scrambling was done in order to highlight any missing dependency orderings. This may have been successful but without the ability to reproduce the exact order of a failing test, we would probably just call the problem a "flake" and ignore it. We should accept a pseudorandom seed so that failing tests can be re-run.

This could help with test automation and test deflaking, but it could even help with production debugging, since we still do that randomization in production builds.