We have encountered workloads that maintain dirty cache usage just above the configured eviction_checkpoint_target in the steady state. When a checkpoint begins, the checkpoint scrub phase starts, and engages application threads to reduce cache usage down to eviction_checkpoint_target - the process of engaging application threads reduces the amount of throughput the application can maintain, causing a disruption to the application.

We should craft a standalone WiredTiger workload that recreates the symptoms, code up a fix that avoids the symptom, and add the workload to our automated testing.