- For the change_stream_latency tests and its derivatives, run only a single configuration: 1KB documents, ~24 worker threads, 1 collection, 1 change stream. Keep the pre-image variants, consider similarly splitting the updateLookup test into its own variant. Consider getting rid of the findOne worker since that just adds load with no direct impact on the test.
- Consider getting rid of the change_stream_throughput tests entirely. These don't measure anything to do with change streams itself, but instead measure the impact of increasing numbers of change streams on an ongoing read/write workload. One benefit of having these tests around is that they can be correlated against the corresponding change_stream_latency tests; if write throughput improves, change streams latency regresses. But we could get the same information by simply recording write throughput in the change_stream_latency tests.
- Consider extending the change_stream_listen_throughput test, which is currently insert-only, with mixed-workload, updateLookup and pre-image variants. This doesn't have to happen under this ticket, since it would involve adding new tests rather than reworking the current ones.