Under severe memory pressure, the current replicated size and count flush can lag behind the tail of the oplog significantly. When a restart happens, initialization of the in-memory size and count needs to scan the oplog entries since the last flush to reconstruct the accurate values. This will block the server from starting up for an extended period of time.

This ticket is to guard the logic of oplog scanning behind a feature flag. Currently, the correctness testing of replicated size and count is run with all feature flags enabled, so the test coverage of the in-memory read path will be maintained. The performance testing and production do not have the extra feature flag enabled, so the potential regression will not affect them.