During analysis of HELP-96433, the high write-latency and low-availability on atlas-1-dev-infinite was attributed to the standby becoming CPU-saturated, falling behind on oplog application, and causing flow control to engage on the primary.

During load testing, the customer directed nearly all read traffic to the standby, overloading it and indirectly degrading write availability on the primary.

In DSC, existing flow control behavior allows standby overload from read traffic to translate into backpressure and primary-side flow control. In practice, this means a read-heavy workload directed at the standby can reduce write availability, even when the primary itself is not overloaded.

We should add a representative workload and evaluate whether any existing IWM overload protections would have mitigated this case. If not, we should explore read throttling on standby-side or another protective mechanism to prevent standby CPU saturation from cascading into primary write degradation.