A mirror currently needs to be in place at time zero (t0) otherwise all messages are not mirrored.

one approach is to sync journals before starting a mirror, but this can be difficult to achieve and coordinate.

having the option to start streaming the journals, and then continue to stream new messages would be ideal. There is always the possibility that a mirror is behind in this case, but good metrics and reporting can help here.

In this mode,  a mirror can get in sync and preserve order.

It may be that order is not vital,  and possibly two mirrors could be in play. One that just deals with history up to a point (and stops). And another that deals with new events. 

This could allow any large backlog to be mirrorred async with new messages. Trading off order and mirroring latency.

At any point, a mirror will need to resume from where it was, meaning that some state management will be implicit but can hopefully be confined to the snf, possibly leaving one pending message while in catch up mode.