There is a known memory issue with the current usage of the CollectionCacheRecoverer.

In the event a thread sets it up and is then interrupted the CSR will remain with a recoverer in place. This recoverer will not be resolved until the next operation that encounters a shard version mismatch is handled.

As a result, all new oplog c entries will keep enqueueing to the recoverer for the drain stage. If enough of those are received then this might have an effect on memory usage. However, we'd need to receive potentially thousands of these entries before it becomes noticeable.