I’m not yet sure.
Personally I’d like to eliminate the need for UnionFS altogether. The “binary install” basically did the trick, but it’s rarely, if ever, used to actually keep several versions on the same system. The only real use for it is rollback after unsuccessfull upgrade. I still believe this problem can be solved at package level. While image install solved the rollback problem, it creates new problems, like the need for reboot on any upgrade and difficulties with installing backwards-compatible package updates, thus making it harder to provide e.g. security bugfixes.
On the other hand this change hardly can be backwards compatible. My original plan was to reimplement vyatta with different design decisions, but everyone else insisted on gradual change, and this summer changes also made it necessary to keep the original product alive if we still want it.
So might be we’ll be making fixes and non-radical changes to the original design while working on “nextgen” at the same time. If we can find a way to do that gradually, even better.
I’m going to see if we can replace it with AuFS. If it works, of course I would like to upgrade the kernel and bring VXLAN and other nice things into original design.