On Wed, 4 May 2022 at 00:21, Benjamin Coutu wrote:
> Thanks David, using extended statistics for both (and only for both) tables
> solved this problem.
Oh, whoops. I did get that backwards. The estimate used by the
Memoize costing code is from the outer side of the join, which is the
extdataemp
> I'd say it's a pretty different problem. The cache hit ratio
> discussion on that thread talks about underestimating the hit ratio.
> That particular problem could only lead to Memoize plans *not* being
> chosen when they maybe should be. Not the other way around, which is
> your case.
>
> creat
On Tue, 3 May 2022 at 23:05, Benjamin Coutu wrote:
> -> Memoize (cost=0.58..0.67 rows=1 width=16) (actual time=0.002..0.002
> rows=0 loops=4067215)
> Cache Key: e2.field, e2.index
> Cache Mode: logical
> Hits: 0 Misses: 4067215 Evictions: 3228355 Overflows: 0 Memor
