The scheduler logic (implicitly throttle writes) was originally added as a tactical fix because the materialization lag flow control wasn't effective at the time. However, now that Haiyin has fixed the shim bug (SERVER-121319), I believe it's time to remove this logic. Even if removing it makes flow control worse, it would give us an early opportunity to evaluate and improve our flow control story.

The current logic is particularly concerning as it can completely block oplog replication, prevent the no-op oplog entry from replicating during materialization lag, and ultimately cause unnecessary failovers.

Additionally, when we remove this logic, it would be valuable to run all the high-value workloads and document any performance impact. Drafting a comparison of throughput with and without this logic would help us better understand its overall effect.