I'm trying to get live migration without block copying working in a custom solution, but I think I need a better understanding of the migration process. In my environment, I have two iscsi targets exported which are then placed into a raid via mdadm on the compute node. I've tried using the already exposed hook on stop to disassemble and re-export storage, and I also made my own when the source hits perform3_done phase, but at that point the md device is still locked. It seems like there should be a time in a live migration where both instances are in a paused state and everything has been flushed to disk, I'm just not familiar enough with the code to determine where that is.

-Paul