Postgres：具有不同前导列的多列索引

Question

我有两列的两个索引：

create index idx_film_length_rating on film (length, rating);
create index idx_film_rating_length on film (rating, length);

当我执行时：

explain analyze select title, length, rating, replacement_cost, rental_rate
from film
where rating = 'G' and length between 60 and 70

postgres 认为 idx_film_rating_length 始终是最佳选择，并使用此索引。但是为什么如果没有第二个索引 idx_film_rating_length，查询会变慢呢？根据我的理解，执行计划是一样的，提取块是一样的，它们应该是一样的。

只有一个索引的结果是：

"Bitmap Heap Scan on film  (cost=4.44..35.70 rows=13 width=34) (actual time=0.102..0.120 rows=18 loops=1)"
"  Recheck Cond: ((rating = 'G'::mpaa_rating) AND (length >= 60) AND (length <= 70))"
"  Heap Blocks: exact=14"
"  ->  Bitmap Index Scan on idx_film_rating_length  (cost=0.00..4.44 rows=13 width=0) (actual time=0.095..0.095 rows=18 loops=1)"
"        Index Cond: ((rating = 'G'::mpaa_rating) AND (length >= 60) AND (length <= 70))"
"Planning time: 0.316 ms"
"Execution time: 0.160 ms"

表中有两个索引的结果：

"Bitmap Heap Scan on film  (cost=4.44..35.70 rows=13 width=34) (actual time=0.030..0.041 rows=18 loops=1)"
"  Recheck Cond: ((rating = 'G'::mpaa_rating) AND (length >= 60) AND (length <= 70))"
"  Heap Blocks: exact=14"
"  ->  Bitmap Index Scan on idx_film_rating_length  (cost=0.00..4.44 rows=13 width=0) (actual time=0.024..0.024 rows=18 loops=1)"
"        Index Cond: ((rating = 'G'::mpaa_rating) AND (length >= 60) AND (length <= 70))"
"Planning time: 0.199 ms"
"Execution time: 0.065 ms"

您可以看到虽然计划相同，但第二个更快。

======================================= @a_horse_with_no_name 推荐 添加缓冲区和细节后：

"Bitmap Heap Scan on film  (cost=4.44..35.70 rows=13 width=34) (actual time=0.692..0.716 rows=18 loops=1)"
"  Recheck Cond: ((rating = 'G'::mpaa_rating) AND (length >= 60) AND (length <= 70))"
"  Heap Blocks: exact=14"
"  Buffers: shared hit=14 read=2"
"  ->  Bitmap Index Scan on idx_film_cover  (cost=0.00..4.44 rows=13 width=0) (actual time=0.680..0.680 rows=18 loops=1)"
"        Index Cond: ((rating = 'G'::mpaa_rating) AND (length >= 60) AND (length <= 70))"
"        Buffers: shared read=2"
"Planning time: 1.773 ms"
"Execution time: 1.441 ms"

似乎问题可能与： " 缓冲区：共享命中=14 读取=2"？

Answer 1

你是对的，计划是一样的。

执行时间的差异只是巧合。或许更多的索引被第二次缓存了。

重复EXPLAIN 几次以获得统计相关的结果。