对 "IN" 子句使用存储的查询结果命中所有分区
Using a Stored Query Result For an "IN" Clause Hits ALL Partitions
我希望能够使用存储的查询结果来缩小我用于后续查询的分区。 table 按年划分，然后按 mun_cty_co 划分。查询很简单，如下所示：
with counties as (
select distinct mun_cty_co
from municipal_boundaries_of_nj_3857
where st_intersects(
wkb_geometry,
ST_TileEnvelope(10, 298, 389)
)
)
select count(*)
from ard_accidents_geom_partition
where year = 2016
and mun_cty_co in (select mun_cty_co from counties)
这会导致扫描每个辅助分区索引的索引。但是，如果我如下硬编码查询中存储结果的结果，它只使用相关的 tables.
select count(*)
from ard_accidents_geom_partition
where year = 2016
and mun_cty_co in ('01','04','06','08','17')
为了（希望）进一步了解，我已经包含了下面的查询计划：
Finalize Aggregate (cost=38424.49..38424.50 rows=1 width=8) (actual time=2688.778..2787.861 rows=1 loops=1)
Buffers: shared hit=25303 read=9701
-> Gather (cost=38424.27..38424.48 rows=2 width=8) (actual time=2685.598..2787.842 rows=3 loops=1)
Workers Planned: 2
Workers Launched: 2
Buffers: shared hit=25303 read=9701
-> Partial Aggregate (cost=37424.27..37424.28 rows=1 width=8) (actual time=2524.317..2524.335 rows=1 loops=3)
Buffers: shared hit=25303 read=9701
-> Hash Join (cost=313.83..37411.15 rows=5248 width=0) (actual time=1718.983..2523.545 rows=11752 loops=3)
Hash Cond: ((ard_accidents_geom_partition.mun_cty_co)::text = (municipal_boundaries_of_nj_3857.mun_cty_co)::text)
Buffers: shared hit=25303 read=9701
-> Parallel Append (cost=0.00..36784.72 rows=116614 width=3) (actual time=2.294..2421.200 rows=93291 loops=3)
Buffers: shared hit=24755 read=9574
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_bergen ard_accidents_geom_partition_2 (cost=0.00..3772.85 rows=12388 width=3) (actual time=0.487..28.623 rows=29731 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared hit=3618
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_middlesex ard_accidents_geom_partition_12 (cost=0.00..3609.27 rows=11702 width=3) (actual time=0.371..23.792 rows=28084 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared hit=3463
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_essex ard_accidents_geom_partition_7 (cost=0.00..3586.76 rows=11900 width=3) (actual time=0.253..22.558 rows=28561 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared hit=3438
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_union ard_accidents_geom_partition_20 (cost=0.00..2621.31 rows=12105 width=3) (actual time=44.684..2015.981 rows=20578 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared read=2470
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_hudson ard_accidents_geom_partition_9 (cost=0.00..2621.04 rows=12484 width=3) (actual time=0.311..22.098 rows=21222 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared hit=2465
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_monmouth ard_accidents_geom_partition_13 (cost=0.00..2504.88 rows=11191 width=3) (actual time=0.024..19.360 rows=19024 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared hit=2365
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_passaic ard_accidents_geom_partition_16 (cost=0.00..2232.54 rows=10604 width=3) (actual time=19.069..1498.141 rows=18026 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared read=2100
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_ocean ard_accidents_geom_partition_15 (cost=0.00..1964.26 rows=8741 width=3) (actual time=10.490..470.290 rows=4953 loops=3)
Filter: (year = '2016'::numeric)
Buffers: shared read=1855
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_morris ard_accidents_geom_partition_14 (cost=0.00..1919.81 rows=8625 width=3) (actual time=0.387..255.318 rows=7331 loops=2)
Filter: (year = '2016'::numeric)
Buffers: shared hit=1072 read=740
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_camden ard_accidents_geom_partition_4 (cost=0.00..1817.93 rows=8074 width=3) (actual time=0.068..6.901 rows=13726 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared hit=1717
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_burlington ard_accidents_geom_partition_3 (cost=0.00..1758.49 rows=7639 width=3) (actual time=0.029..5.552 rows=12986 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared hit=1663
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_mercer ard_accidents_geom_partition_11 (cost=0.00..1650.51 rows=7401 width=3) (actual time=0.017..6.058 rows=12581 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared hit=1558
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_somerset ard_accidents_geom_partition_18 (cost=0.00..1446.36 rows=6349 width=3) (actual time=13.808..651.948 rows=10793 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared read=1367
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_atlantic ard_accidents_geom_partition_1 (cost=0.00..1073.80 rows=4624 width=3) (actual time=0.005..3.565 rows=7861 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared hit=1016
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_gloucester ard_accidents_geom_partition_8 (cost=0.00..1048.90 rows=4552 width=3) (actual time=0.034..3.300 rows=7738 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared hit=992
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_cumberland ard_accidents_geom_partition_6 (cost=0.00..564.20 rows=2496 width=3) (actual time=0.038..1.754 rows=4243 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared hit=533
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_hunterdon ard_accidents_geom_partition_10 (cost=0.00..514.00 rows=2320 width=3) (actual time=0.031..1.745 rows=3944 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared hit=485
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_warren ard_accidents_geom_partition_21 (cost=0.00..444.85 rows=1988 width=3) (actual time=4.744..512.568 rows=3380 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared read=420
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_sussex ard_accidents_geom_partition_19 (cost=0.00..426.77 rows=1902 width=3) (actual time=10.438..258.951 rows=3233 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared read=403
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_cape_may ard_accidents_geom_partition_5 (cost=0.00..391.72 rows=1738 width=3) (actual time=0.120..1.417 rows=2954 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared hit=370
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_salem ard_accidents_geom_partition_17 (cost=0.00..231.40 rows=992 width=3) (actual time=6.074..237.824 rows=1687 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared read=219
-> Hash (cost=313.72..313.72 rows=9 width=3) (actual time=89.302..89.304 rows=5 loops=3)
Buckets: 1024 Batches: 1 Memory Usage: 9kB
Buffers: shared hit=520 read=127
-> Unique (cost=313.57..313.63 rows=9 width=3) (actual time=56.201..56.218 rows=5 loops=3)
Buffers: shared hit=520 read=127
-> Sort (cost=313.57..313.60 rows=12 width=3) (actual time=56.200..56.203 rows=20 loops=3)
Sort Key: municipal_boundaries_of_nj_3857.mun_cty_co
Sort Method: quicksort Memory: 25kB
Buffers: shared hit=520 read=127
Worker 0: Sort Method: quicksort Memory: 25kB
Worker 1: Sort Method: quicksort Memory: 25kB
-> Index Scan using municipal_boundaries_of_nj_3857_wkb_geometry_idx on municipal_boundaries_of_nj_3857 (cost=0.14..313.35 rows=12 width=3) (actual time=19.338..55.244 rows=20 loops=3)
Index Cond: (wkb_geometry && '0103000020110F00000100000005000000B66B2B14C3F25FC1CC3F3AA2A8365241B66B2B14C3F25FC1EE37C592E05C52419473A0238BCC5FC1EE37C592E05C52419473A0238BCC5FC1CC3F3AA2A8365241B66B2B14C3F25FC1CC3F3AA2A8365241'::geometry)
Filter: st_intersects(wkb_geometry, '0103000020110F00000100000005000000B66B2B14C3F25FC1CC3F3AA2A8365241B66B2B14C3F25FC1EE37C592E05C52419473A0238BCC5FC1EE37C592E05C52419473A0238BCC5FC1CC3F3AA2A8365241B66B2B14C3F25FC1CC3F3AA2A8365241'::geometry)
Buffers: shared hit=498 read=127
Planning:
Buffers: shared hit=357
Planning Time: 72.631 ms
Execution Time: 2808.158 ms
为了比较，我还包括了我希望获得的计划，给定硬编码查询。或者至少对我来说似乎更直观......
Aggregate (cost=5402.44..5402.45 rows=1 width=8) (actual time=25.161..25.164 rows=1 loops=1)
Buffers: shared hit=4477
-> Append (cost=0.00..5314.31 rows=35255 width=0) (actual time=0.025..23.289 rows=35255 loops=1)
Buffers: shared hit=4477
-> Seq Scan on ard_accidents_geom_partition_2016_atlantic ard_accidents_geom_partition_1 (cost=0.00..1163.39 rows=7861 width=0) (actual time=0.023..4.500 rows=7861 loops=1)
Filter: ((year = '2016'::numeric) AND ((mun_cty_co)::text = ANY ('{01,04,06,08,17}'::text[])))
Buffers: shared hit=1016
-> Seq Scan on ard_accidents_geom_partition_2016_camden ard_accidents_geom_partition_2 (cost=0.00..1974.36 rows=13726 width=0) (actual time=0.121..7.977 rows=13726 loops=1)
Filter: ((year = '2016'::numeric) AND ((mun_cty_co)::text = ANY ('{01,04,06,08,17}'::text[])))
Buffers: shared hit=1717
-> Seq Scan on ard_accidents_geom_partition_2016_cumberland ard_accidents_geom_partition_3 (cost=0.00..612.56 rows=4243 width=0) (actual time=0.072..2.527 rows=4243 loops=1)
Filter: ((year = '2016'::numeric) AND ((mun_cty_co)::text = ANY ('{01,04,06,08,17}'::text[])))
Buffers: shared hit=533
-> Seq Scan on ard_accidents_geom_partition_2016_gloucester ard_accidents_geom_partition_4 (cost=0.00..1137.09 rows=7738 width=0) (actual time=0.052..4.867 rows=7738 loops=1)
Filter: ((year = '2016'::numeric) AND ((mun_cty_co)::text = ANY ('{01,04,06,08,17}'::text[])))
Buffers: shared hit=992
-> Seq Scan on ard_accidents_geom_partition_2016_salem ard_accidents_geom_partition_5 (cost=0.00..250.63 rows=1687 width=0) (actual time=0.050..1.146 rows=1687 loops=1)
Filter: ((year = '2016'::numeric) AND ((mun_cty_co)::text = ANY ('{01,04,06,08,17}'::text[])))
Buffers: shared hit=219
Planning:
Buffers: shared hit=51
Planning Time: 1.832 ms
Execution Time: 25.254 ms
我根据 Laurenz 的建议添加了年份，mun_cty_co 索引，现在它正在使用索引扫描。但是，它似乎仍在检查所有分区。
Finalize Aggregate (cost=7166.30..7166.31 rows=1 width=8) (actual time=93.193..107.194 rows=1 loops=1)
Buffers: shared hit=1202
-> Gather (cost=7166.19..7166.30 rows=1 width=8) (actual time=92.957..107.186 rows=2 loops=1)
Workers Planned: 1
Workers Launched: 1
Buffers: shared hit=1202
-> Partial Aggregate (cost=6166.19..6166.20 rows=1 width=8) (actual time=52.912..52.923 rows=1 loops=2)
Buffers: shared hit=1202
-> Hash Join (cost=314.12..6147.67 rows=7408 width=0) (actual time=8.761..51.825 rows=17628 loops=2)
Hash Cond: ((ard_accidents_geom_partition.mun_cty_co)::text = (municipal_boundaries_of_nj_3857.mun_cty_co)::text)
Buffers: shared hit=1202
-> Parallel Append (cost=0.29..5392.52 rows=164635 width=3) (actual time=0.283..33.851 rows=139937 loops=2)
Buffers: shared hit=820
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_bergen_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_bergen ard_accidents_geom_partition_2 (cost=0.29..482.16 rows=17489 width=3) (actual time=0.568..3.314 rows=14866 loops=2)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=86
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_essex_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_essex ard_accidents_geom_partition_7 (cost=0.29..463.50 rows=16801 width=3) (actual time=0.633..5.470 rows=28561 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=81
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_middlesex_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_middlesex ard_accidents_geom_partition_12 (cost=0.29..455.12 rows=16520 width=3) (actual time=0.463..5.309 rows=28084 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=79
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_hudson_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_hudson ard_accidents_geom_partition_9 (cost=0.29..344.29 rows=12484 width=3) (actual time=0.538..4.285 rows=21222 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=60
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_union_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_union ard_accidents_geom_partition_20 (cost=0.29..334.67 rows=12105 width=3) (actual time=0.042..3.395 rows=20578 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=59
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_monmouth_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_monmouth ard_accidents_geom_partition_13 (cost=0.29..308.87 rows=11191 width=3) (actual time=0.043..3.021 rows=19024 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=54
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_passaic_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_passaic ard_accidents_geom_partition_16 (cost=0.29..293.52 rows=10604 width=3) (actual time=0.039..3.040 rows=18026 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=52
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_ocean_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_ocean ard_accidents_geom_partition_15 (cost=0.29..242.15 rows=8741 width=3) (actual time=0.040..2.471 rows=14860 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=43
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_morris_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_morris ard_accidents_geom_partition_14 (cost=0.29..239.50 rows=8625 width=3) (actual time=0.045..2.300 rows=14662 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=43
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_camden_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_camden ard_accidents_geom_partition_4 (cost=0.29..223.97 rows=8074 width=3) (actual time=0.041..2.426 rows=13726 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=40
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_burlingto_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_burlington ard_accidents_geom_partition_3 (cost=0.29..212.07 rows=7639 width=3) (actual time=0.041..2.137 rows=12986 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=38
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_mercer_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_mercer ard_accidents_geom_partition_11 (cost=0.29..205.65 rows=7401 width=3) (actual time=0.042..2.033 rows=12581 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=37
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_somerset_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_somerset ard_accidents_geom_partition_18 (cost=0.29..176.72 rows=6349 width=3) (actual time=0.041..1.842 rows=10793 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=32
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_atlantic_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_atlantic ard_accidents_geom_partition_1 (cost=0.28..129.48 rows=4624 width=3) (actual time=0.040..1.400 rows=7861 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=24
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_glouceste_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_gloucester ard_accidents_geom_partition_8 (cost=0.28..127.84 rows=4552 width=3) (actual time=0.037..1.279 rows=7738 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=24
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_cumberlan_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_cumberland ard_accidents_geom_partition_6 (cost=0.28..71.06 rows=2496 width=3) (actual time=0.038..0.699 rows=4243 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=14
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_hunterdon_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_hunterdon ard_accidents_geom_partition_10 (cost=0.28..66.06 rows=2320 width=3) (actual time=0.040..0.674 rows=3944 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=13
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_warren_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_warren ard_accidents_geom_partition_21 (cost=0.28..57.51 rows=1988 width=3) (actual time=0.040..0.550 rows=3380 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=12
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_sussex_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_sussex ard_accidents_geom_partition_19 (cost=0.28..54.55 rows=1902 width=3) (actual time=0.035..0.553 rows=3233 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=11
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_cape_may_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_cape_may ard_accidents_geom_partition_5 (cost=0.28..50.81 rows=1738 width=3) (actual time=0.040..0.519 rows=2954 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=11
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_salem_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_salem ard_accidents_geom_partition_17 (cost=0.28..29.85 rows=992 width=3) (actual time=0.050..0.307 rows=1687 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=7
-> Hash (cost=313.72..313.72 rows=9 width=3) (actual time=3.386..3.388 rows=5 loops=2)
Buckets: 1024 Batches: 1 Memory Usage: 9kB
Buffers: shared hit=368
-> Unique (cost=313.57..313.63 rows=9 width=3) (actual time=3.375..3.383 rows=5 loops=2)
Buffers: shared hit=368
-> Sort (cost=313.57..313.60 rows=12 width=3) (actual time=3.374..3.375 rows=20 loops=2)
Sort Key: municipal_boundaries_of_nj_3857.mun_cty_co
Sort Method: quicksort Memory: 25kB
Buffers: shared hit=368
Worker 0: Sort Method: quicksort Memory: 25kB
-> Index Scan using municipal_boundaries_of_nj_3857_wkb_geometry_idx on municipal_boundaries_of_nj_3857 (cost=0.14..313.35 rows=12 width=3) (actual time=0.839..3.295 rows=20 loops=2)
Index Cond: (wkb_geometry && '0103000020110F00000100000005000000B66B2B14C3F25FC1CC3F3AA2A8365241B66B2B14C3F25FC1EE37C592E05C52419473A0238BCC5FC1EE37C592E05C52419473A0238BCC5FC1CC3F3AA2A8365241B66B2B14C3F25FC1CC3F3AA2A8365241'::geometry)
Filter: st_intersects(wkb_geometry, '0103000020110F00000100000005000000B66B2B14C3F25FC1CC3F3AA2A8365241B66B2B14C3F25FC1EE37C592E05C52419473A0238BCC5FC1EE37C592E05C52419473A0238BCC5FC1CC3F3AA2A8365241B66B2B14C3F25FC1CC3F3AA2A8365241'::geometry)
Buffers: shared hit=357
Planning Time: 4.253 ms
Execution Time: 107.782 ms
这仍然没有像许多人描述的那样工作，当我将子查询结果限制为单个值时，我继续调查并发现了以下内容。
with counties as (
(
select mun_cty_co
from county_boundaries_of_nj_3857
where st_intersects(
wkb_geometry,
ST_TileEnvelope(10, 298, 389)
)
limit 1
))
select count(*)
from ard_accidents_geom_partition
where year = 2016
and mun_cty_co = (select * from counties)
结果反映了一些人提到的（从未执行过）。如何使用多个值实现这一目标？
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_atlantic_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_atlantic ard_accidents_geom_partition_1 (cost=0.28..149.13 rows=4624 width=0) (actual time=0.080..1.676 rows=7861 loops=1)
Index Cond: ((year = '2016'::numeric) AND (mun_cty_co = ([=17=])::text))
Heap Fetches: 0
Buffers: shared hit=24
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_salem_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_salem ard_accidents_geom_partition_17 (cost=0.28..34.07 rows=992 width=0) (never executed)
Index Cond: ((year = '2016'::numeric) AND (mun_cty_co = ([=17=])::text))
Heap Fetches: 0
因为第二种情况优化器已经可以确定需要扫描哪些分区，而第一种情况无法提前确定。
此答案假定您的 PostgreSQL 版本为 v11 或更低版本，因为 v12 在查询执行时引入了分区修剪。
如果您的 PostgreSQL 版本是 v12 或更高版本，如果优化器将您的查询转换为与分区 table内侧：在这种情况下，跳过的分区的 EXPLAIN (ANALYZE) 将如下所示：
-> Seq Scan on ard_accidents_geom_partition_p42 tab_2 (cost=0.00..10000 rows=1 width=48) (never executed)
Filter: (mun_cty_co = 42)
在您的情况下，您应该在 ard_accidents_geom_partition(year, mun_cty_co) 上有一个索引，以鼓励可以受益的嵌套循环连接。
我希望能够使用存储的查询结果来缩小我用于后续查询的分区。 table 按年划分，然后按 mun_cty_co 划分。查询很简单，如下所示：
with counties as (
select distinct mun_cty_co
from municipal_boundaries_of_nj_3857
where st_intersects(
wkb_geometry,
ST_TileEnvelope(10, 298, 389)
)
)
select count(*)
from ard_accidents_geom_partition
where year = 2016
and mun_cty_co in (select mun_cty_co from counties)
这会导致扫描每个辅助分区索引的索引。但是，如果我如下硬编码查询中存储结果的结果，它只使用相关的 tables.
select count(*)
from ard_accidents_geom_partition
where year = 2016
and mun_cty_co in ('01','04','06','08','17')
为了（希望）进一步了解，我已经包含了下面的查询计划：
Finalize Aggregate (cost=38424.49..38424.50 rows=1 width=8) (actual time=2688.778..2787.861 rows=1 loops=1)
Buffers: shared hit=25303 read=9701
-> Gather (cost=38424.27..38424.48 rows=2 width=8) (actual time=2685.598..2787.842 rows=3 loops=1)
Workers Planned: 2
Workers Launched: 2
Buffers: shared hit=25303 read=9701
-> Partial Aggregate (cost=37424.27..37424.28 rows=1 width=8) (actual time=2524.317..2524.335 rows=1 loops=3)
Buffers: shared hit=25303 read=9701
-> Hash Join (cost=313.83..37411.15 rows=5248 width=0) (actual time=1718.983..2523.545 rows=11752 loops=3)
Hash Cond: ((ard_accidents_geom_partition.mun_cty_co)::text = (municipal_boundaries_of_nj_3857.mun_cty_co)::text)
Buffers: shared hit=25303 read=9701
-> Parallel Append (cost=0.00..36784.72 rows=116614 width=3) (actual time=2.294..2421.200 rows=93291 loops=3)
Buffers: shared hit=24755 read=9574
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_bergen ard_accidents_geom_partition_2 (cost=0.00..3772.85 rows=12388 width=3) (actual time=0.487..28.623 rows=29731 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared hit=3618
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_middlesex ard_accidents_geom_partition_12 (cost=0.00..3609.27 rows=11702 width=3) (actual time=0.371..23.792 rows=28084 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared hit=3463
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_essex ard_accidents_geom_partition_7 (cost=0.00..3586.76 rows=11900 width=3) (actual time=0.253..22.558 rows=28561 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared hit=3438
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_union ard_accidents_geom_partition_20 (cost=0.00..2621.31 rows=12105 width=3) (actual time=44.684..2015.981 rows=20578 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared read=2470
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_hudson ard_accidents_geom_partition_9 (cost=0.00..2621.04 rows=12484 width=3) (actual time=0.311..22.098 rows=21222 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared hit=2465
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_monmouth ard_accidents_geom_partition_13 (cost=0.00..2504.88 rows=11191 width=3) (actual time=0.024..19.360 rows=19024 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared hit=2365
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_passaic ard_accidents_geom_partition_16 (cost=0.00..2232.54 rows=10604 width=3) (actual time=19.069..1498.141 rows=18026 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared read=2100
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_ocean ard_accidents_geom_partition_15 (cost=0.00..1964.26 rows=8741 width=3) (actual time=10.490..470.290 rows=4953 loops=3)
Filter: (year = '2016'::numeric)
Buffers: shared read=1855
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_morris ard_accidents_geom_partition_14 (cost=0.00..1919.81 rows=8625 width=3) (actual time=0.387..255.318 rows=7331 loops=2)
Filter: (year = '2016'::numeric)
Buffers: shared hit=1072 read=740
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_camden ard_accidents_geom_partition_4 (cost=0.00..1817.93 rows=8074 width=3) (actual time=0.068..6.901 rows=13726 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared hit=1717
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_burlington ard_accidents_geom_partition_3 (cost=0.00..1758.49 rows=7639 width=3) (actual time=0.029..5.552 rows=12986 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared hit=1663
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_mercer ard_accidents_geom_partition_11 (cost=0.00..1650.51 rows=7401 width=3) (actual time=0.017..6.058 rows=12581 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared hit=1558
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_somerset ard_accidents_geom_partition_18 (cost=0.00..1446.36 rows=6349 width=3) (actual time=13.808..651.948 rows=10793 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared read=1367
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_atlantic ard_accidents_geom_partition_1 (cost=0.00..1073.80 rows=4624 width=3) (actual time=0.005..3.565 rows=7861 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared hit=1016
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_gloucester ard_accidents_geom_partition_8 (cost=0.00..1048.90 rows=4552 width=3) (actual time=0.034..3.300 rows=7738 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared hit=992
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_cumberland ard_accidents_geom_partition_6 (cost=0.00..564.20 rows=2496 width=3) (actual time=0.038..1.754 rows=4243 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared hit=533
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_hunterdon ard_accidents_geom_partition_10 (cost=0.00..514.00 rows=2320 width=3) (actual time=0.031..1.745 rows=3944 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared hit=485
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_warren ard_accidents_geom_partition_21 (cost=0.00..444.85 rows=1988 width=3) (actual time=4.744..512.568 rows=3380 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared read=420
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_sussex ard_accidents_geom_partition_19 (cost=0.00..426.77 rows=1902 width=3) (actual time=10.438..258.951 rows=3233 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared read=403
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_cape_may ard_accidents_geom_partition_5 (cost=0.00..391.72 rows=1738 width=3) (actual time=0.120..1.417 rows=2954 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared hit=370
-> Parallel Seq Scan on ard_accidents_geom_partition_2016_salem ard_accidents_geom_partition_17 (cost=0.00..231.40 rows=992 width=3) (actual time=6.074..237.824 rows=1687 loops=1)
Filter: (year = '2016'::numeric)
Buffers: shared read=219
-> Hash (cost=313.72..313.72 rows=9 width=3) (actual time=89.302..89.304 rows=5 loops=3)
Buckets: 1024 Batches: 1 Memory Usage: 9kB
Buffers: shared hit=520 read=127
-> Unique (cost=313.57..313.63 rows=9 width=3) (actual time=56.201..56.218 rows=5 loops=3)
Buffers: shared hit=520 read=127
-> Sort (cost=313.57..313.60 rows=12 width=3) (actual time=56.200..56.203 rows=20 loops=3)
Sort Key: municipal_boundaries_of_nj_3857.mun_cty_co
Sort Method: quicksort Memory: 25kB
Buffers: shared hit=520 read=127
Worker 0: Sort Method: quicksort Memory: 25kB
Worker 1: Sort Method: quicksort Memory: 25kB
-> Index Scan using municipal_boundaries_of_nj_3857_wkb_geometry_idx on municipal_boundaries_of_nj_3857 (cost=0.14..313.35 rows=12 width=3) (actual time=19.338..55.244 rows=20 loops=3)
Index Cond: (wkb_geometry && '0103000020110F00000100000005000000B66B2B14C3F25FC1CC3F3AA2A8365241B66B2B14C3F25FC1EE37C592E05C52419473A0238BCC5FC1EE37C592E05C52419473A0238BCC5FC1CC3F3AA2A8365241B66B2B14C3F25FC1CC3F3AA2A8365241'::geometry)
Filter: st_intersects(wkb_geometry, '0103000020110F00000100000005000000B66B2B14C3F25FC1CC3F3AA2A8365241B66B2B14C3F25FC1EE37C592E05C52419473A0238BCC5FC1EE37C592E05C52419473A0238BCC5FC1CC3F3AA2A8365241B66B2B14C3F25FC1CC3F3AA2A8365241'::geometry)
Buffers: shared hit=498 read=127
Planning:
Buffers: shared hit=357
Planning Time: 72.631 ms
Execution Time: 2808.158 ms
为了比较，我还包括了我希望获得的计划，给定硬编码查询。或者至少对我来说似乎更直观......
Aggregate (cost=5402.44..5402.45 rows=1 width=8) (actual time=25.161..25.164 rows=1 loops=1)
Buffers: shared hit=4477
-> Append (cost=0.00..5314.31 rows=35255 width=0) (actual time=0.025..23.289 rows=35255 loops=1)
Buffers: shared hit=4477
-> Seq Scan on ard_accidents_geom_partition_2016_atlantic ard_accidents_geom_partition_1 (cost=0.00..1163.39 rows=7861 width=0) (actual time=0.023..4.500 rows=7861 loops=1)
Filter: ((year = '2016'::numeric) AND ((mun_cty_co)::text = ANY ('{01,04,06,08,17}'::text[])))
Buffers: shared hit=1016
-> Seq Scan on ard_accidents_geom_partition_2016_camden ard_accidents_geom_partition_2 (cost=0.00..1974.36 rows=13726 width=0) (actual time=0.121..7.977 rows=13726 loops=1)
Filter: ((year = '2016'::numeric) AND ((mun_cty_co)::text = ANY ('{01,04,06,08,17}'::text[])))
Buffers: shared hit=1717
-> Seq Scan on ard_accidents_geom_partition_2016_cumberland ard_accidents_geom_partition_3 (cost=0.00..612.56 rows=4243 width=0) (actual time=0.072..2.527 rows=4243 loops=1)
Filter: ((year = '2016'::numeric) AND ((mun_cty_co)::text = ANY ('{01,04,06,08,17}'::text[])))
Buffers: shared hit=533
-> Seq Scan on ard_accidents_geom_partition_2016_gloucester ard_accidents_geom_partition_4 (cost=0.00..1137.09 rows=7738 width=0) (actual time=0.052..4.867 rows=7738 loops=1)
Filter: ((year = '2016'::numeric) AND ((mun_cty_co)::text = ANY ('{01,04,06,08,17}'::text[])))
Buffers: shared hit=992
-> Seq Scan on ard_accidents_geom_partition_2016_salem ard_accidents_geom_partition_5 (cost=0.00..250.63 rows=1687 width=0) (actual time=0.050..1.146 rows=1687 loops=1)
Filter: ((year = '2016'::numeric) AND ((mun_cty_co)::text = ANY ('{01,04,06,08,17}'::text[])))
Buffers: shared hit=219
Planning:
Buffers: shared hit=51
Planning Time: 1.832 ms
Execution Time: 25.254 ms
我根据 Laurenz 的建议添加了年份，mun_cty_co 索引，现在它正在使用索引扫描。但是，它似乎仍在检查所有分区。
Finalize Aggregate (cost=7166.30..7166.31 rows=1 width=8) (actual time=93.193..107.194 rows=1 loops=1)
Buffers: shared hit=1202
-> Gather (cost=7166.19..7166.30 rows=1 width=8) (actual time=92.957..107.186 rows=2 loops=1)
Workers Planned: 1
Workers Launched: 1
Buffers: shared hit=1202
-> Partial Aggregate (cost=6166.19..6166.20 rows=1 width=8) (actual time=52.912..52.923 rows=1 loops=2)
Buffers: shared hit=1202
-> Hash Join (cost=314.12..6147.67 rows=7408 width=0) (actual time=8.761..51.825 rows=17628 loops=2)
Hash Cond: ((ard_accidents_geom_partition.mun_cty_co)::text = (municipal_boundaries_of_nj_3857.mun_cty_co)::text)
Buffers: shared hit=1202
-> Parallel Append (cost=0.29..5392.52 rows=164635 width=3) (actual time=0.283..33.851 rows=139937 loops=2)
Buffers: shared hit=820
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_bergen_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_bergen ard_accidents_geom_partition_2 (cost=0.29..482.16 rows=17489 width=3) (actual time=0.568..3.314 rows=14866 loops=2)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=86
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_essex_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_essex ard_accidents_geom_partition_7 (cost=0.29..463.50 rows=16801 width=3) (actual time=0.633..5.470 rows=28561 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=81
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_middlesex_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_middlesex ard_accidents_geom_partition_12 (cost=0.29..455.12 rows=16520 width=3) (actual time=0.463..5.309 rows=28084 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=79
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_hudson_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_hudson ard_accidents_geom_partition_9 (cost=0.29..344.29 rows=12484 width=3) (actual time=0.538..4.285 rows=21222 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=60
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_union_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_union ard_accidents_geom_partition_20 (cost=0.29..334.67 rows=12105 width=3) (actual time=0.042..3.395 rows=20578 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=59
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_monmouth_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_monmouth ard_accidents_geom_partition_13 (cost=0.29..308.87 rows=11191 width=3) (actual time=0.043..3.021 rows=19024 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=54
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_passaic_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_passaic ard_accidents_geom_partition_16 (cost=0.29..293.52 rows=10604 width=3) (actual time=0.039..3.040 rows=18026 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=52
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_ocean_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_ocean ard_accidents_geom_partition_15 (cost=0.29..242.15 rows=8741 width=3) (actual time=0.040..2.471 rows=14860 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=43
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_morris_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_morris ard_accidents_geom_partition_14 (cost=0.29..239.50 rows=8625 width=3) (actual time=0.045..2.300 rows=14662 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=43
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_camden_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_camden ard_accidents_geom_partition_4 (cost=0.29..223.97 rows=8074 width=3) (actual time=0.041..2.426 rows=13726 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=40
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_burlingto_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_burlington ard_accidents_geom_partition_3 (cost=0.29..212.07 rows=7639 width=3) (actual time=0.041..2.137 rows=12986 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=38
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_mercer_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_mercer ard_accidents_geom_partition_11 (cost=0.29..205.65 rows=7401 width=3) (actual time=0.042..2.033 rows=12581 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=37
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_somerset_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_somerset ard_accidents_geom_partition_18 (cost=0.29..176.72 rows=6349 width=3) (actual time=0.041..1.842 rows=10793 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=32
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_atlantic_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_atlantic ard_accidents_geom_partition_1 (cost=0.28..129.48 rows=4624 width=3) (actual time=0.040..1.400 rows=7861 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=24
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_glouceste_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_gloucester ard_accidents_geom_partition_8 (cost=0.28..127.84 rows=4552 width=3) (actual time=0.037..1.279 rows=7738 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=24
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_cumberlan_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_cumberland ard_accidents_geom_partition_6 (cost=0.28..71.06 rows=2496 width=3) (actual time=0.038..0.699 rows=4243 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=14
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_hunterdon_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_hunterdon ard_accidents_geom_partition_10 (cost=0.28..66.06 rows=2320 width=3) (actual time=0.040..0.674 rows=3944 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=13
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_warren_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_warren ard_accidents_geom_partition_21 (cost=0.28..57.51 rows=1988 width=3) (actual time=0.040..0.550 rows=3380 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=12
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_sussex_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_sussex ard_accidents_geom_partition_19 (cost=0.28..54.55 rows=1902 width=3) (actual time=0.035..0.553 rows=3233 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=11
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_cape_may_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_cape_may ard_accidents_geom_partition_5 (cost=0.28..50.81 rows=1738 width=3) (actual time=0.040..0.519 rows=2954 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=11
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_salem_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_salem ard_accidents_geom_partition_17 (cost=0.28..29.85 rows=992 width=3) (actual time=0.050..0.307 rows=1687 loops=1)
Index Cond: (year = '2016'::numeric)
Heap Fetches: 0
Buffers: shared hit=7
-> Hash (cost=313.72..313.72 rows=9 width=3) (actual time=3.386..3.388 rows=5 loops=2)
Buckets: 1024 Batches: 1 Memory Usage: 9kB
Buffers: shared hit=368
-> Unique (cost=313.57..313.63 rows=9 width=3) (actual time=3.375..3.383 rows=5 loops=2)
Buffers: shared hit=368
-> Sort (cost=313.57..313.60 rows=12 width=3) (actual time=3.374..3.375 rows=20 loops=2)
Sort Key: municipal_boundaries_of_nj_3857.mun_cty_co
Sort Method: quicksort Memory: 25kB
Buffers: shared hit=368
Worker 0: Sort Method: quicksort Memory: 25kB
-> Index Scan using municipal_boundaries_of_nj_3857_wkb_geometry_idx on municipal_boundaries_of_nj_3857 (cost=0.14..313.35 rows=12 width=3) (actual time=0.839..3.295 rows=20 loops=2)
Index Cond: (wkb_geometry && '0103000020110F00000100000005000000B66B2B14C3F25FC1CC3F3AA2A8365241B66B2B14C3F25FC1EE37C592E05C52419473A0238BCC5FC1EE37C592E05C52419473A0238BCC5FC1CC3F3AA2A8365241B66B2B14C3F25FC1CC3F3AA2A8365241'::geometry)
Filter: st_intersects(wkb_geometry, '0103000020110F00000100000005000000B66B2B14C3F25FC1CC3F3AA2A8365241B66B2B14C3F25FC1EE37C592E05C52419473A0238BCC5FC1EE37C592E05C52419473A0238BCC5FC1CC3F3AA2A8365241B66B2B14C3F25FC1CC3F3AA2A8365241'::geometry)
Buffers: shared hit=357
Planning Time: 4.253 ms
Execution Time: 107.782 ms
这仍然没有像许多人描述的那样工作，当我将子查询结果限制为单个值时，我继续调查并发现了以下内容。
with counties as (
(
select mun_cty_co
from county_boundaries_of_nj_3857
where st_intersects(
wkb_geometry,
ST_TileEnvelope(10, 298, 389)
)
limit 1
))
select count(*)
from ard_accidents_geom_partition
where year = 2016
and mun_cty_co = (select * from counties)
结果反映了一些人提到的（从未执行过）。如何使用多个值实现这一目标？
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_atlantic_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_atlantic ard_accidents_geom_partition_1 (cost=0.28..149.13 rows=4624 width=0) (actual time=0.080..1.676 rows=7861 loops=1)
Index Cond: ((year = '2016'::numeric) AND (mun_cty_co = ([=17=])::text))
Heap Fetches: 0
Buffers: shared hit=24
-> Parallel Index Only Scan using ard_accidents_geom_partition_2016_salem_year_mun_cty_co_idx on ard_accidents_geom_partition_2016_salem ard_accidents_geom_partition_17 (cost=0.28..34.07 rows=992 width=0) (never executed)
Index Cond: ((year = '2016'::numeric) AND (mun_cty_co = ([=17=])::text))
Heap Fetches: 0
因为第二种情况优化器已经可以确定需要扫描哪些分区，而第一种情况无法提前确定。
此答案假定您的 PostgreSQL 版本为 v11 或更低版本，因为 v12 在查询执行时引入了分区修剪。
如果您的 PostgreSQL 版本是 v12 或更高版本，如果优化器将您的查询转换为与分区 table内侧：在这种情况下，跳过的分区的 EXPLAIN (ANALYZE) 将如下所示：
-> Seq Scan on ard_accidents_geom_partition_p42 tab_2 (cost=0.00..10000 rows=1 width=48) (never executed)
Filter: (mun_cty_co = 42)
在您的情况下，您应该在 ard_accidents_geom_partition(year, mun_cty_co) 上有一个索引，以鼓励可以受益的嵌套循环连接。