Corey Taylor <corey.taylor.fl@gmail.com> writes:
> I found after testing other situations, that the psql command would always
> finish as expected after canceling the first query that ran too long. I
> was able to reproduce this scenario with psql and pgadmin4 with various
> combinations.
Well, that's just weird.
It's well known that the second run of a query can be much faster due
to having fully-populated caches to draw on, but you seem to have a
case that may go beyond that. Maybe check for waiting on a lock?
It'd be useful to look in pg_stat_activity and/or top(1) while the
initial query is running, to see if it seems to be eating CPU or
is blocked on some condition. (I forget how thorough the
wait_event coverage is in 9.6, but it does at least have those
columns.)
Can you create a self-contained test case that acts like this?
regards, tom lane