For correctness and performance reasons, secondaries do not update oplog visibility when write units of work commit, but rather at the end of applying a batch.

The batch completion logic for initial sync is separate and does not seem to contain an analogous clause.

This can result in hangs finishing up initial sync.