We'd like to have some kind of a stress test for oplog visibility. One goal of this is to be able to identify any cases where we have delayed visibility updates.

We aren't sure exactly what existing workloads might reveal issues with delayed visibility updates, and so as part of this process we should experiment by adding some artificial delays to today's visibility tracking information and then running our existing performance tests to see which of them reveal issues that result from visibility delays. From there, we can figure out what a oplog visibility-specific test for this should look like.