Also... why is a node name required at all in the branch qualifier? If you simply used an empty bqual ID that consists only of the numeric resource # sequence like I illustrated, wouldn't you still be able to recover? The bqual node name information is not useful in identifying the transactions to recover as far as I can see, since in the end each transaction coordinator is going to rely on the transport layer to restore any connections and talk to the subordinate(s) to get their list of transactions, and ultimately the transaction ownership would be determined by the gtid (right?), so the bqual information doesn't appear to add anything at all and merely needs to be unique per branch. The worst I could imagine would be that the same node might list two branches as needing recovery... is this logic more subtle than I seem to think it is?