On Wed, 2022-11-23 at 15:38 +0200, Danny Shemesh wrote:
> -> Limit (cost=0.56..24.17 rows=1 width=16) (actual time=2657.167..2657.167 rows=1 loops=39)
> Buffers: shared hit=323452 read=887661
> I/O Timings: read=2369932.536
> -> Index Only Scan using idx_hashes on refs refs_1 (cost=0.56..265306.68 rows=11238 width=16) (actual
time=2657.162..2657.162rows=1 loops=39)
> Index Cond: ((tid = '13371337-1337-1337-1337-133713371337'::uuid) AND (pidh > cte_1.pidh))
> Filter: (tidh = ANY ('{13391339-1339-1339-1339-133913391339}'::uuid[])) <<<<<<<<<<<<<<<- Note this
line
> Rows Removed by Filter: 346024
> Heap Fetches: 1506359
> Buffers: shared hit=323452 read=887661
> I/O Timings: read=2369932.536
PostgreSQL thinks that there are enough such rows that it is cheaper to use the index
that supports the ORDER BY. I don't know why there is a difference between = ANY
and = here, but you can use an expression like "ORDER BY pidh + 0" to avoid that.
Yours,
Laurenz Albe