Not sure, but do you see prepared transactions in "pg_prepared_xacts"?
No, the -1 in the virtualtransaction (https://www.postgresql.org/docs/11/view-pg-locks.html) for pg_prepared_xacts was another clue I saw! But, it seems more or less a dead end as I have nothing in pg_prepared_xacts.
Thanks for the idea, though.
I still need to put more effort into Tom's idea about SIReadLock hanging out after the transaction, but some evidence pointing in this direction is that I've reduced the number of db connections and found that the '-1/0' locks will eventually go away! I interpret this as the db needing to find time when no overlapping read/write transactions are present. This doesn't seem completely correct, as I don't have any long lived transactions running while these locks are hanging out. Confusion still remains, for sure.
