PostgreSQL中的IN语句性能(以及一般情况下)

Question

我知道之前可能已经提到了这个问题,但是我找不到SO的搜索.

让我说我有TABLE1和TABLE2,我应该如何期望这样的查询的性能:

SELECT * FROM TABLE1 WHERE id IN SUBQUERY_ON_TABLE2;

随着TABLE1和TABLE2中的行数增加而下降,而id是TABLE1上的主键.

是的,我知道使用IN是一个n00b错误,但TABLE2与其他多个表有一个通用关系(django泛型关系),所以我想不出另一种过滤数据的方法.在TABLE1和TABLE2中的什么(aproximate)ammount,我应该注意到性能问题吗？性能会根据行数线性地,指数地降低吗？

Answer 1

当子查询返回的记录数很少,并且主查询返回的结果行数也很小时,您只需对每个查询进行快速索引查找.随着返回数据的百分比增加,最终两者中的每一个都将切换到使用顺序扫描而不是索引扫描,以一次吞下整个表而不是将它拼凑在一起.这不是一个简单的性能下降,无论是线性的还是指数的; 随着计划类型的变化,存在重大的不连续性.并且发生这些行的行数取决于表的大小,因此对您来说也没有有用的经验法则.您应该构建一个类似于我在下面进行的模拟,看看您自己的数据集上发生了什么,以了解曲线的样子.

以下是使用装载Dell Store 2数据库的PostgreSQL 9.0数据库的工作原理示例.一旦子查询返回了1000行,它就会对主表进行全表扫描.一旦子查询考虑了10,000条记录,那么它也会变成全表扫描.这些都运行了两次,所以你看到了缓存的性能.基于缓存与未缓存状态的性能变化如何完全是一个完整的主题:

dellstore2=# EXPLAIN ANALYZE SELECT * FROM customers WHERE customerid IN 
  (SELECT customerid FROM orders WHERE orderid<2);
Nested Loop  (cost=8.27..16.56 rows=1 width=268) (actual time=0.051..0.060 rows=1 loops=1)
  ->  HashAggregate  (cost=8.27..8.28 rows=1 width=4) (actual time=0.028..0.030 rows=1 loops=1)
        ->  Index Scan using orders_pkey on orders  (cost=0.00..8.27 rows=1 width=4) (actual time=0.011..0.015 rows=1 loops=1)
              Index Cond: (orderid < 2)
  ->  Index Scan using customers_pkey on customers  (cost=0.00..8.27 rows=1 width=268) (actual time=0.013..0.016 rows=1 loops=1)
        Index Cond: (customers.customerid = orders.customerid)
Total runtime: 0.191 ms

dellstore2=# EXPLAIN ANALYZE SELECT * FROM customers WHERE customerid IN 
  (SELECT customerid FROM orders WHERE orderid<100);
Nested Loop  (cost=10.25..443.14 rows=100 width=268) (actual time=0.488..2.591 rows=98 loops=1)
  ->  HashAggregate  (cost=10.25..11.00 rows=75 width=4) (actual time=0.464..0.661 rows=98 loops=1)
        ->  Index Scan using orders_pkey on orders  (cost=0.00..10.00 rows=100 width=4) (actual time=0.019..0.218 rows=99 loops=1)
              Index Cond: (orderid < 100)
  ->  Index Scan using customers_pkey on customers  (cost=0.00..5.75 rows=1 width=268) (actual time=0.009..0.011 rows=1 loops=98)
        Index Cond: (customers.customerid = orders.customerid)
Total runtime: 2.868 ms

dellstore2=# EXPLAIN ANALYZE SELECT * FROM customers WHERE customerid IN 
  (SELECT customerid FROM orders WHERE orderid<1000);
Hash Semi Join  (cost=54.25..800.13 rows=1000 width=268) (actual time=4.574..80.319 rows=978 loops=1)
  Hash Cond: (customers.customerid = orders.customerid)
  ->  Seq Scan on customers  (cost=0.00..676.00 rows=20000 width=268) (actual time=0.007..33.665 rows=20000 loops=1)
  ->  Hash  (cost=41.75..41.75 rows=1000 width=4) (actual time=4.502..4.502 rows=999 loops=1)
        Buckets: 1024  Batches: 1  Memory Usage: 24kB
        ->  Index Scan using orders_pkey on orders  (cost=0.00..41.75 rows=1000 width=4) (actual time=0.056..2.487 rows=999 loops=1)
              Index Cond: (orderid < 1000)
Total runtime: 82.024 ms

dellstore2=# EXPLAIN ANALYZE SELECT * FROM customers WHERE customerid IN 
  (SELECT customerid FROM orders WHERE orderid<10000);
Hash Join  (cost=443.68..1444.68 rows=8996 width=268) (actual time=79.576..157.159 rows=7895 loops=1)
  Hash Cond: (customers.customerid = orders.customerid)
  ->  Seq Scan on customers  (cost=0.00..676.00 rows=20000 width=268) (actual time=0.007..27.085 rows=20000 loops=1)
  ->  Hash  (cost=349.97..349.97 rows=7497 width=4) (actual time=79.532..79.532 rows=7895 loops=1)
        Buckets: 1024  Batches: 1  Memory Usage: 186kB
        ->  HashAggregate  (cost=275.00..349.97 rows=7497 width=4) (actual time=45.130..62.227 rows=7895 loops=1)
              ->  Seq Scan on orders  (cost=0.00..250.00 rows=10000 width=4) (actual time=0.008..20.979 rows=9999 loops=1)
                    Filter: (orderid < 10000)
Total runtime: 167.882 ms