Hi Will,

Thanks for the info. Query execution plans are hard to predict, and can be very sensitive to the data in the tables, the version of the database, etc.

I've seen some other cases where switching from a subselect to a join greatly improves Postgres performance, so I made a change so that the server will generate SQL similar to your revised query. It will be part of 16.1, but I'm attaching the patch here in case you want to try it against 15.3 on a development machine.

Thanks,
Josh