Currently deployment (or it could be called upgrade from 2.4) needs a checklist of items executed in the proper sequence for replication to start successfully.
Pinloki should be more aware that it is reading from the master for the first time, e.g. incoming replicants must be parked until enough data exists to allow them to start to read files. This change need not, and should not, make any explicit dependencies between "Writer", the component that reads from the primary and creates binlogs, and "Reader", the component that reads binlogs and serves replicants. This is important, as it allows a Reader to run in a separate process (for which some small changes to the Inventory class will be needed).
Make it robust, and allow re-initialization at any time, i.e. allow the entire pinloki state to be erased and re-created seamlessly and automatically on the flip of a switch. This re-initialization feature will save situations where a bug or some unforeseen situation breaks pinloki replication (there are no such known bugs or situations currently).