Currently, slaveDelay is handled in a few places in sync_tail.cpp, and batches end after the first op that can't be applied yet. In addition to other issues, this means that ops that can be applied must wait for ops that can't. The plan for this ticket is to leave ops that can't be applied yet in BGSync's BlockingQueue and not adding them to a batch.
- is related to
SERVER-21118 A slaveDelay'd member in sleep gets stuck in shutdown