为什么在PostgreSQL上启用intarray扩展导致性能下降10倍？

2023年6月22日 160次阅读

这是我运行的日志.

随着intarray关闭,它运行在56毫秒.

然后我启用了intarray,并运行相同的查询,而且速度要慢得多.

在启用扩展程序后,是否需要重新索引表格或类似内容？

test_int=# explain analyze select * from tutor_topic tt1 where tt1.topic_id @> '{5,7,8,9}';
                                                      QUERY PLAN                                                       
-----------------------------------------------------------------------------------------------------------------------
 Seq Scan on tutor_topic tt1  (cost=0.00..29742.00 rows=1000 width=105) (actual time=24.903..1937.480 rows=68 loops=1)
   Filter: (topic_id @> '{5,7,8,9}'::integer[])
   Rows Removed by Filter: 999932
 Planning time: 0.084 ms
 Execution time: 1937.521 ms
(5 rows)

Time: 1938.000 ms
test_int=# DROP EXTENSION intarray;                                                                                                                              DROP EXTENSION
Time: 10.516 ms
test_int=# explain analyze select * from tutor_topic tt1 where tt1.topic_id @> '{5,7,8,9}';
                                                            QUERY PLAN                                                            
----------------------------------------------------------------------------------------------------------------------------------
 Bitmap Heap Scan on tutor_topic tt1  (cost=108.78..487.18 rows=100 width=105) (actual time=55.063..55.138 rows=68 loops=1)
   Recheck Cond: (topic_id @> '{5,7,8,9}'::integer[])
   Heap Blocks: exact=68
   ->  Bitmap Index Scan on message_rdtree_idx  (cost=0.00..108.75 rows=100 width=0) (actual time=55.047..55.047 rows=68 loops=1)
         Index Cond: (topic_id @> '{5,7,8,9}'::integer[])
 Planning time: 0.196 ms
 Execution time: 55.180 ms
(7 rows)

Time: 56.095 ms
test_int=#

这是tutor_topic表的模式;

CREATE TABLE IF NOT EXISTS tutor_topic                                               
(                                                                                    
      tutor_id              INT NOT NULL,                                            
      topic_id              INT[]                                                    
);

这些是指数：

ALTER TABLE tutor_topic ADD FOREIGN KEY (tutor_id) REFERENCES tutor(tutor_id);       
CREATE INDEX ON tutor_topic (tutor_id);                                              
CREATE INDEX message_rdtree_idx ON tutor_topic USING GIN (topic_id)   

 Schema |           Name            | Type  | Owner  |    Table    
--------+---------------------------+-------+--------+-------------
 public | message_rdtree_idx        | index | xxxxxx | tutor_topic
 public | topic_pkey                | index | xxxxxx | topic
 public | tutor_pkey                | index | xxxxxx | tutor
 public | tutor_topic_tutor_id_idx  | index | xxxxxx | tutor_topic
 public | tutor_topic_tutor_id_idx1 | index | xxxxxx | tutor_topic

最佳答案 我想出了答案！

创建索引时,我需要从扩展名中指定函数：
CREATE INDEX message_rdtree_idx ON tutor_topic使用GIN(topic_id gin__int_ops);

现在表现匹配.从explain analyze调用来看,似乎根本没有使用索引.

点赞