failed rebalance followed by failover leads to very large data loss

After loading about 6 million docs into a 3 node cluster, I attempted to add 2 more nodes and rebalance, but it hung and I stopped it after ~20 minutes. The cluster however showed that the nodes were added so I failed over 2 of the original nodes and the dataloss was 72% (6million to 700k docs). Seems rather high, though I suspect this was due to a failure to distribute items to nodes that were first rebalanced in.

If this behavior is expected, I wonder if it will be possible to also warn user about how much data will be lost.

Real bug is probably with rebalance. diags attached.