Question

我的用例是，我需要对字段进行文本搜索，然后按照与文本搜索无关的其他列进行排序，但我似乎无法创建处理这两者的索引。

创建表格：

create table file (
id bigint,
path character varying(2048),
peers bigint,
text_search tsvector
);

要测试的一些指数：

create index idx_file_text_search_1 on file using gin (text_search);
create index idx_file_text_search_2 on file using gin (peers, text_search);
create index idx_file_peers on file using btree (peers desc);

这是我的主要问题：

explain analyze 
select * 
from file_fast 
where text_search @@ to_tsquery('whatever') 
order by peers desc 
limit 10;

然而它只使用同行指数：

 Limit  (cost=0.43..20870.27 rows=10 width=316) (actual time=2507.304..9016.220 rows=10 loops=1)
   ->  Index Scan using idx_file_peers on file  (cost=0.43..18286146.09 rows=8762 width=316) (actual time=2507.301..9016.205 rows=10 loops=1)
         Filter: (text_search @@ to_tsquery('ole'::text))
         Rows Removed by Filter: 497504
 Planning time: 0.399 ms
 Execution time: 9016.265 ms
(6 rows)

当我在没有订单的情况下尝试它时，似乎使用文本搜索索引：

-------------------------------------------------------------------------------------------------------------------------------------------------
 Limit  (cost=104.15..143.54 rows=10 width=316) (actual time=76.949..76.977 rows=10 loops=1)
   ->  Bitmap Heap Scan on file  (cost=104.15..34612.36 rows=8762 width=316) (actual time=76.946..76.970 rows=10 loops=1)
         Recheck Cond: (text_search @@ to_tsquery('ole'::text))
         Heap Blocks: exact=10
         ->  Bitmap Index Scan on idx_file_text_search_1 (cost=0.00..101.96 rows=8762 width=0) (actual time=76.802..76.802 rows=515 loops=1)
               Index Cond: (text_search @@ to_tsquery('ole'::text))
 Planning time: 0.376 ms
 Execution time: 175.775 ms
(8 rows)

postgres是否真的缺少能够进行文本搜索的索引，并对另一个字段进行排序？

Answer 1

不知道你是否可以改进索引，但如果第二个查询更快，你可以拆分查询

with cte as (
    select * 
    from file_fast 
    where text_search @@ to_tsquery('whatever') 
)
SELECT *
FROM cte
order by peers desc 
limit 10;

Postgres文本搜索顺序/排序

1 个答案: