larvinloy commented :
@MichelDiz We can’t use bulk loader cause its a live cluster. We cant use live loader (with set operations) cause we cant afford duplicate nodes in a data-set of that size. Also, having to store an xid mapping somewhere else adds unnecessary additional complexity.
The idea of leasing all the uids was that, we’d hash our ids to some uid in the entire uid set( 2⁶⁴). That would only work if we have access to all 2⁶⁴ uids at once.
This is why upsert support in the live loader would be so useful to us.