1,假设有一个表 结构和索引如下:
create table test1(ID BIGINT,name varchar(10),age int,addr varchar(128));
插入测试数据:
insert into test1 select i,'name'||i,100,substr(md5(random()::text),1,30) from generate_series(1,1000000) as t(i);
-- 创建组合索引,ID + NAME
create index idx_test_id_name on test1(id ,name);
analyze test1;
2,分页查询语句
-- sql语句:
select id,name from test1 where id between 1000 and 10000
and name in ('name1','name2','name3')
order by id desc limit 24;
-- 执行计划显示
explain (analyze,buffers)
select id,name from test1 where id between 1000 and 10000
and name in ('name1','name2','name3')
order by id desc limit 24;
Limit (cost=0.42..10879 rows=1 width=51) (actual time=35.174..35.174 rows=0 loops=1)
Buffers:shared hit = 39 read=93
--> Index Scan using idx_test1_id_name on test1(cost=0.42..10879.37 row=1 width=51)(actual time=35.169..35.169 rows=0 loops=1)
Index Cond:((id >=1000 ) and (id <=10000)) Filter:((name)::text = ANY('{name1,name2}'::text[]))
Rows Removed by Filter:9001
Buffers:shared hit=39 read = 93
Planning time : 2.183ms
Execution time : 35.311ms
-- 2个疑问:
第一个问题: Filter:((name)::text = ANY('{name1,name2}'::text[]))
这步为什么不能在索引的块中过滤,Pg走的是回表过滤,这样会造成效率低下。
仅仅是根据idx_test1_id_name的前导列范围扫描id,直接回表读取heap page过滤
而且我只选中了id和name两列都在索引中的,理论上不需要回表嘛, 这点在oracle数据库中确认没有回表过程
第二个问题: postgresql的执行计划很难看出是否真正的回表了。上面的回表过滤我是通过查看 pg_buffercache插件确认的。