PG Bug reporting form <noreply(at)postgresql(dot)org> writes:
> I have experienced a misbehaviour (IMHO) in PostgreSQL 17 that didn't occur
> in PostgreSQL 14.
What have you got work_mem set to in the two installations?
For me, both versions switch to the cheap hashed plan if work_mem
is large enough that the inner table is estimated to fit in it
(somewhere around 10MB-20MB for this specific example). If anything
the newer code seems to have a smaller threshold.
regards, tom lane