In 3.2 replication introduced the journal listener, which is used to tell the replication subsystem when a particular log op has been persisted locally.

Without it, read concern majority is calculating the journalled committed optime without taking into account primary's contribution.

Because of this, the RocksDB storage engine needs to support the journal writer.

Note that this task might become unnecessary due to SERVER-22575. However, until either one of these two issues is fixed, sharding and replication tests with RocksDB will continue failing.