三元组索引 ORDER BY 优化

Question

我正在尝试实现搜索功能，经过一些调查（参见 this interesting read by Yorick Peterse at GitLab）我决定选择使用 pg_trgm 扩展的三元组方法。

我想要 return 10 个最相关的行。

以下是我针对具有 110868 行的 table 测试的几个查询（在 the doc 之后）：

SELECT name, similarity(name, 'search query') AS sml
FROM table
ORDER BY sml DESC, name;
Time: 701.814 ms

SELECT name, similarity(name, 'search query') AS sml
FROM table
WHERE name % 'search query'
ORDER BY sml DESC, name;
Time: 376.692 ms

SELECT name, similarity(name, 'search query') AS sml
FROM table
WHERE name % 'search query'
ORDER BY sml DESC, name LIMIT 10;
Time: 378.921 ms

使用 GiST 索引：

CREATE INDEX trigram_index ON table USING GIST (name gist_trgm_ops);
SELECT name, similarity(name, 'search query') AS sml
FROM table
WHERE name % 'search query'
ORDER BY sml DESC, name LIMIT 10;
Time: 36.877 ms

使用 GIN 索引：

CREATE INDEX trigram_index ON table USING GIN (name gin_trgm_ops);
SELECT name, similarity(name, 'search query') AS sml
FROM table WHERE name % 'search query'
ORDER BY sml DESC, name LIMIT 10;
Time: 18.992 ms

使用解释分析：

 Limit  (cost=632.37..632.39 rows=10 width=25) (actual time=22.202..22.204 rows=10 loops=1)
   ->  Sort  (cost=632.37..632.64 rows=111 width=25) (actual time=22.201..22.201 rows=10 loops=1)
         Sort Key: (similarity((name)::text, 'search query'::text)) DESC, name
         Sort Method: top-N heapsort  Memory: 26kB
         ->  Bitmap Heap Scan on table  (cost=208.86..629.97 rows=111 width=25) (actual time=6.900..22.157 rows=134 loops=1)
               Recheck Cond: ((name)::text % 'search query'::text)
               Rows Removed by Index Recheck: 2274
               Heap Blocks: exact=2257
               ->  Bitmap Index Scan on trigram_index  (cost=0.00..208.83 rows=111 width=0) (actual time=6.532..6.532 rows=2408 loops=1)
                     Index Cond: ((name)::text % 'World of Warcraft'::text)
 Planning time: 0.073 ms
 Execution time: 18.521 ms

使用 GIN 索引可以显着提高性能。然而，将结果限制为 10 行似乎没有任何影响。

还有没有考虑到的改进空间？我对利用我只需要整个 table.

这一事实的建议特别感兴趣

Answer 1

正如the documentation所说，GIN索引无助于优化ORDER BY子句：

A variant of the above query is
SELECT t, t <-> 'word' AS dist
FROM test_trgm
ORDER BY dist LIMIT 10;
This can be implemented quite efficiently by GiST indexes, but not by GIN indexes. It will usually beat the first formulation when only a small number of the closest matches is wanted.

另一方面，对于更大的 tables，GIN 索引通常比 GiST 索引表现更好。

所以我认为您应该同时尝试两者，并且只使用在实际大小的测试中速度更快的那个 table。

除了使用更多 RAM 来缓存数据外，我不认为您可以改进更多。

三元组索引 ORDER BY 优化

Trigram Index ORDER BY optimization

postgresql

database-indexes

postgresql-10

pg-trgm