Grzegorz Jaśkiewicz wrote:
> you're joining on more than one key. That always hurts performance.
That's very clearly *not* the problem, as there is a plan which runs
in acceptable time but the optimizer is not choosing without being
coerced.
(1) Virtually every query we run joins on multi-column keys, yet we
have good performance except for this one query.
(2) We're talking about a performance regression due to a new release
picking a newly available plan which it wrongly estimates to be an
order of magnitude faster, when it's actually more than five times
slower.
-Kevin