Hi Thomas,
pg_controldata -D pgdata
returns
Date/time type storage: 64-bit integers
Regarding the parallel execution on two same-server databases, I can also reproduce the problem running sequentially (same-thread) across the two same-server databases.
(Note: in both parallel and sequential execution the multi-db same-query inserts are both completed before the multi-db same-query updates get executed)
What is interesting is that in either threading case the incorrect timestamp values are written consistently in both databases, which is a down-vote for random corruption.
