我有一个按周分区的表。 start_time上有一个索引 当我运行这个 -
EXPLAIN SELECT COUNT(uniq_id) FROM call_fact WHERE CAST(start_time AS DATE) = '2018-01-23' AND source_key=2;
它会扫描每个分区表，而不是转到正确的子表。查询优化器是否应该足够聪明才能选择正确的子表？有没有办法强制postgres去正确的子表？
EXPLAIN(analyze, buffers) SELECT COUNT(uniq_id) FROM call_fact WHERE CAST(start_time AS DATE) = '2018-01-23' AND source_key=2;
QUERY PLAN
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Aggregate (cost=649579.52..649579.53 rows=1 width=26) (actual time=845.998..845.998 rows=1 loops=1)
Buffers: shared hit=34839 read=1312
-> Append (cost=0.12..647969.68 rows=643937 width=26) (actual time=175.416..792.387 rows=457658 loops=1)
Buffers: shared hit=34839 read=1312
-> Index Scan using call_fact_cast_start_time_idx on call_fact (cost=0.12..8.14 rows=1 width=26) (actual time=0.006..0.006 rows=0 loops=1)
Index Cond: ((start_time)::date = '2018-01-23'::date)
Filter: (source_key = 2)
Buffers: shared hit=1
-> Bitmap Heap Scan on call_fact_2017w42 (cost=259.52..43804.31 rows=13604 width=26) (actual time=16.957..16.957 rows=0 loops=1)
Recheck Cond: ((start_time)::date = '2018-01-23'::date)
Filter: (source_key = 2)
Buffers: shared read=3
-> Bitmap Index Scan on call_fact_2017w42_cast_start_time_idx (cost=0.00..256.12 rows=13825 width=0) (actual time=16.954..16.954 rows=0 loops=1)
Index Cond: ((start_time)::date = '2018-01-23'::date)
Buffers: shared read=3
-> Bitmap Heap Scan on call_fact_2017w43 (cost=258.00..43186.24 rows=13425 width=26) (actual time=14.656..14.656 rows=0 loops=1)
Recheck Cond: ((start_time)::date = '2018-01-23'::date)
Filter: (source_key = 2)
Buffers: shared read=3
-> Bitmap Index Scan on call_fact_2017w43_cast_start_time_idx (cost=0.00..254.64 rows=13628 width=0) (actual time=14.653..14.653 rows=0 loops=1)
Index Cond: ((start_time)::date = '2018-01-23'::date)
Buffers: shared read=3
-> Bitmap Heap Scan on call_fact_2017w44 (cost=272.45..45811.42 rows=14264 width=26) (actual time=5.930..5.930 rows=0 loops=1)
Recheck Cond: ((start_time)::date = '2018-01-23'::date)
Filter: (source_key = 2)
Buffers: shared read=3
-> Bitmap Index Scan on call_fact_2017w44_cast_start_time_idx (cost=0.00..268.88 rows=14460 width=0) (actual time=5.921..5.921 rows=0 loops=1)
Index Cond: ((start_time)::date = '2018-01-23'::date)
Buffers: shared read=3
-> Bitmap Heap Scan on call_fact_2017w45 (cost=258.80..43538.05 rows=13534 width=26) (actual time=10.631..10.631 rows=0 loops=1)
Recheck Cond: ((start_time)::date = '2018-01-23'::date)
Filter: (source_key = 2)
Buffers: shared read=3
-> Bitmap Index Scan on call_fact_2017w45_cast_start_time_idx (cost=0.00..255.41 rows=13731 width=0) (actual time=10.630..10.630 rows=0 loops=1)
Index Cond: ((start_time)::date = '2018-01-23'::date)
Buffers: shared read=3
-> Bitmap Heap Scan on call_fact_2017w46 (cost=258.38..43396.98 rows=13479 width=26) (actual time=14.289..14.289 rows=0 loops=1)
Recheck Cond: ((start_time)::date = '2018-01-23'::date)
Filter: (source_key = 2)
Buffers: shared read=3
-> Bitmap Index Scan on call_fact_2017w46_cast_start_time_idx (cost=0.00..255.01 rows=13678 width=0) (actual time=14.281..14.281 rows=0 loops=1)
Index Cond: ((start_time)::date = '2018-01-23'::date)
Buffers: shared read=3
-> Bitmap Heap Scan on call_fact_2017w47 (cost=203.86..34194.57 rows=10556 width=26) (actual time=13.386..13.386 rows=0 loops=1)
Recheck Cond: ((start_time)::date = '2018-01-23'::date)
Filter: (source_key = 2)
Buffers: shared read=3
-> Bitmap Index Scan on call_fact_2017w47_cast_start_time_idx (cost=0.00..201.22 rows=10772 width=0) (actual time=13.384..13.384 rows=0 loops=1)
Index Cond: ((start_time)::date = '2018-01-23'::date)
Buffers: shared read=3
-> Bitmap Heap Scan on call_fact_2017w48 (cost=279.31..47044.41 rows=14662 width=26) (actual time=10.149..10.149 rows=0 loops=1)
Recheck Cond: ((start_time)::date = '2018-01-23'::date)
Filter: (source_key = 2)
Buffers: shared read=3
-> Bitmap Index Scan on call_fact_2017w48_cast_start_time_idx (cost=0.00..275.64 rows=14828 width=0) (actual time=10.143..10.143 rows=0 loops=1)
Index Cond: ((start_time)::date = '2018-01-23'::date)
Buffers: shared read=3
-> Bitmap Heap Scan on call_fact_2017w49 (cost=270.62..45091.94 rows=14064 width=26) (actual time=8.687..8.687 rows=0 loops=1)
Recheck Cond: ((start_time)::date = '2018-01-23'::date)
Filter: (source_key = 2)
Buffers: shared read=3
-> Bitmap Index Scan on call_fact_2017w49_cast_start_time_idx (cost=0.00..267.10 rows=14223 width=0) (actual time=8.684..8.684 rows=0 loops=1)
Index Cond: ((start_time)::date = '2018-01-23'::date)
Buffers: shared read=3
-> Bitmap Heap Scan on call_fact_2017w50 (cost=251.05..41960.30 rows=13073 width=26) (actual time=18.181..18.181 rows=0 loops=1)
Recheck Cond: ((start_time)::date = '2018-01-23'::date)
Filter: (source_key = 2)
Buffers: shared read=3
-> Bitmap Index Scan on call_fact_2017w50_cast_start_time_idx (cost=0.00..247.78 rows=13247 width=0) (actual time=18.175..18.175 rows=0 loops=1)
Index Cond: ((start_time)::date = '2018-01-23'::date)
Buffers: shared read=3
-> Bitmap Heap Scan on call_fact_2017w51 (cost=238.23..39983.46 rows=12434 width=26) (actual time=10.728..10.728 rows=0 loops=1)
Recheck Cond: ((start_time)::date = '2018-01-23'::date)
Filter: (source_key = 2)
Buffers: shared read=3
-> Bitmap Index Scan on call_fact_2017w51_cast_start_time_idx (cost=0.00..235.12 rows=12625 width=0) (actual time=10.726..10.726 rows=0 loops=1)
Index Cond: ((start_time)::date = '2018-01-23'::date)
Buffers: shared read=3
-> Bitmap Heap Scan on call_fact_2017w52 (cost=231.37..38822.25 rows=12069 width=26) (actual time=5.907..5.907 rows=0 loops=1)
Recheck Cond: ((start_time)::date = '2018-01-23'::date)
Filter: (source_key = 2)
Buffers: shared read=3
-> Bitmap Index Scan on call_fact_2017w52_cast_start_time_idx (cost=0.00..228.36 rows=12257 width=0) (actual time=5.905..5.905 rows=0 loops=1)
Index Cond: ((start_time)::date = '2018-01-23'::date)
Buffers: shared read=3
-> Bitmap Heap Scan on call_fact_2018w01 (cost=272.97..46008.85 rows=14313 width=26) (actual time=13.909..13.909 rows=0 loops=1)
Recheck Cond: ((start_time)::date = '2018-01-23'::date)
Filter: (source_key = 2)
Buffers: shared read=3
-> Bitmap Index Scan on call_fact_2018w01_cast_start_time_idx (cost=0.00..269.39 rows=14528 width=0) (actual time=13.908..13.908 rows=0 loops=1)
Index Cond: ((start_time)::date = '2018-01-23'::date)
Buffers: shared read=3
-> Bitmap Heap Scan on call_fact_2018w02 (cost=259.65..43842.27 rows=13642 width=26) (actual time=12.807..12.807 rows=0 loops=1)
Recheck Cond: ((start_time)::date = '2018-01-23'::date)
Filter: (source_key = 2)
Buffers: shared read=3
-> Bitmap Index Scan on call_fact_2018w02_cast_start_time_idx (cost=0.00..256.24 rows=13841 width=0) (actual time=12.806..12.806 rows=0 loops=1)
Index Cond: ((start_time)::date = '2018-01-23'::date)
Buffers: shared read=3
-> Bitmap Heap Scan on call_fact_2018w03 (cost=245.87..41477.94 rows=12905 width=26) (actual time=10.540..10.540 rows=0 loops=1)
Recheck Cond: ((start_time)::date = '2018-01-23'::date)
Filter: (source_key = 2)
Buffers: shared read=3
-> Bitmap Index Scan on call_fact_2018w03_cast_start_time_idx (cost=0.00..242.64 rows=13095 width=0) (actual time=10.538..10.538 rows=0 loops=1)
Index Cond: ((start_time)::date = '2018-01-23'::date)
Buffers: shared read=3
-> Index Scan using call_fact_2018w04_cast_start_time_idx on call_fact_2018w04 (cost=0.43..49688.01 rows=457898 width=26) (actual time=8.644..562.016 rows=457658 loops=1)
Index Cond: ((start_time)::date = '2018-01-23'::date)
Filter: (source_key = 2)
Rows Removed by Filter: 5782
Buffers: shared hit=34822 read=1270
-> Index Scan using call_fact_2018w05_cast_start_time_idx on call_fact_2018w05 (cost=0.42..4.44 rows=1 width=26) (actual time=0.039..0.039 rows=0 loops=1)
Index Cond: ((start_time)::date = '2018-01-23'::date)
Filter: (source_key = 2)
Buffers: shared hit=3
-> Index Scan using call_fact_2018w06_cast_start_time_idx on call_fact_2018w06 (cost=0.14..8.16 rows=1 width=98) (actual time=0.004..0.004 rows=0 loops=1)
Index Cond: ((start_time)::date = '2018-01-23'::date)
Filter: (source_key = 2)
Buffers: shared hit=1
-> Index Scan using call_fact_2018w07_cast_start_time_idx on call_fact_2018w07 (cost=0.14..8.16 rows=1 width=98) (actual time=0.003..0.003 rows=0 loops=1)
Index Cond: ((start_time)::date = '2018-01-23'::date)
Filter: (source_key = 2)
Buffers: shared hit=1
-> Index Scan using call_fact_2018w08_cast_start_time_idx on call_fact_2018w08 (cost=0.14..8.16 rows=1 width=98) (actual time=0.003..0.003 rows=0 loops=1)
Index Cond: ((start_time)::date = '2018-01-23'::date)
Filter: (source_key = 2)
Buffers: shared hit=1
-> Index Scan using call_fact_2018w09_cast_start_time_idx on call_fact_2018w09 (cost=0.14..8.16 rows=1 width=98) (actual time=0.003..0.003 rows=0 loops=1)
Index Cond: ((start_time)::date = '2018-01-23'::date)
Filter: (source_key = 2)
Buffers: shared hit=1
-> Index Scan using call_fact_2018w10_cast_start_time_idx on call_fact_2018w10 (cost=0.14..8.16 rows=1 width=98) (actual time=0.003..0.003 rows=0 loops=1)
Index Cond: ((start_time)::date = '2018-01-23'::date)
Filter: (source_key = 2)
Buffers: shared hit=1
-> Index Scan using call_fact_2018w11_cast_start_time_idx on call_fact_2018w11 (cost=0.14..8.16 rows=1 width=98) (actual time=0.003..0.003 rows=0 loops=1)
Index Cond: ((start_time)::date = '2018-01-23'::date)
Filter: (source_key = 2)
Buffers: shared hit=1
-> Index Scan using call_fact_2018w12_cast_start_time_idx on call_fact_2018w12 (cost=0.14..8.16 rows=1 width=98) (actual time=0.002..0.002 rows=0 loops=1)
Index Cond: ((start_time)::date = '2018-01-23'::date)
Filter: (source_key = 2)
Buffers: shared hit=1
-> Index Scan using call_fact_2018w13_cast_start_time_idx on call_fact_2018w13 (cost=0.14..8.16 rows=1 width=98) (actual time=0.003..0.003 rows=0 loops=1)
Index Cond: ((start_time)::date = '2018-01-23'::date)
Filter: (source_key = 2)
Buffers: shared hit=1
-> Index Scan using call_fact_2018w14_cast_start_time_idx on call_fact_2018w14 (cost=0.14..8.16 rows=1 width=98) (actual time=0.002..0.002 rows=0 loops=1)
Index Cond: ((start_time)::date = '2018-01-23'::date)
Filter: (source_key = 2)
Buffers: shared hit=1
-> Index Scan using call_fact_2018w15_source_key_idx on call_fact_2018w15 (cost=0.14..8.16 rows=1 width=98) (actual time=0.003..0.003 rows=0 loops=1)
Index Cond: (source_key = 2)
Filter: ((start_time)::date = '2018-01-23'::date)
Buffers: shared hit=1
-> Index Scan using call_fact_2018w16_source_key_idx on call_fact_2018w16 (cost=0.14..8.16 rows=1 width=98) (actual time=0.003..0.003 rows=0 loops=1)
Index Cond: (source_key = 2)
Filter: ((start_time)::date = '2018-01-23'::date)
Buffers: shared hit=1
-> Index Scan using call_fact_2018w17_cast_start_time_idx on call_fact_2018w17 (cost=0.14..8.16 rows=1 width=98) (actual time=0.004..0.004 rows=0 loops=1)
Index Cond: ((start_time)::date = '2018-01-23'::date)
Filter: (source_key = 2)
Buffers: shared hit=1
-> Index Scan using call_fact_2018w18_cast_start_time_idx on call_fact_2018w18 (cost=0.14..8.16 rows=1 width=98) (actual time=0.005..0.005 rows=0 loops=1)
Index Cond: ((start_time)::date = '2018-01-23'::date)
Filter: (source_key = 2)
Buffers: shared hit=1
Planning time: 68.906 ms
Execution time: 846.648 ms
答案 0 :(得分：6)
start_time的类型是什么？我认为它是timestamp，而不是date。
表达式CAST(start_time AS DATE)阻止优化器在start_time上使用索引。通常，列上的任何类型的表达式都会阻止优化器使用索引。
编写此类查询的常用方法是比较原始列值，而不将其包装到具有闭合打开间隔的表达式中。
SELECT COUNT(uniq_id)
FROM call_fact
WHERE
source_key=2
AND start_time >= '2018-01-23'
AND start_time < '2018-01-24'
;
Postgres应该足够聪明，可以将2018-01-23类型的常量表达式date隐式转换为timestamp类型。 （而不是将start_time列的值转换为date类型。将哪种类型转换为由隐式类型转换的规则控制的类型。在SQL Server中，它被称为“类型优先级” “，但我在Postgres文档中找不到相关页面。
为了使其显式而不依赖于类型优先级和隐式规则，我将以这样的方式显式地编写日期/时间文字：timestamp类型与列的类型匹配。
SELECT COUNT(uniq_id)
FROM call_fact
WHERE
source_key=2
AND start_time >= '2018-01-23 00:00:00'
AND start_time < '2018-01-24 00:00:00'
;
当以这种方式编写查询时，不应该阻止优化器使用start_time上的索引。
答案 1 :(得分：0)
请参阅弗拉基米尔的答案，了解为何会发生这种情况。
您还可以使用准备好的查询将类型化值参数绑定到查询 - 然后可以在过滤器中使用：
PREPARE query(timestamp) as
select max(b)
from customer
where b > $1 - interval '1 months';
EXPLAIN ANALYZE EXECUTE query(current_date);
-- You should deallocate your statement when you are done using it.
DEALLOCATE query;
详细了解PREPARE https://www.postgresql.org/docs/current/static/sql-prepare.html