Question

我一直在对56GB大小的表（789700760行）运行以下查询，并且在执行时遇到了瓶颈。从前面的一些例子中我发现，可能有一种方法可以“取消嵌套”INNER JOIN，以便查询对大型数据集的性能更好。特别是下面的查询花了7.651小时来完成MPP PostgreSQL部署的执行。

create table large_table as
select column1, column2, column3, column4, column5, column6
from
(
  select 
    a.column1, a.column2, a.start_time,
    rank() OVER( 
      PARTITION BY a.column2, a.column1 order by a.start_time DESC 
    ) as rank,
    last_value( a.column3) OVER (
      PARTITION BY a.column2, a.column1 order by a.start_time ASC
      RANGE BETWEEN unbounded preceding and unbounded following 
    ) as column3,
    a.column4, a.column5, a.column6
  from 
    (table2 s 
      INNER JOIN table3 t 
      ON s.column2=t.column2 and s.event_time > t.start_time 
    ) a
 ) b
 where rank =1;

问题1：有没有办法修改上面的sql代码以加快查询的整体执行时间？

Answer 1

您可以将last_value移动到外部子查询，这可能会为您带来一些性能提升。 last_value为每个开始时间最小的分区获取column3的值 - 确切地说rank = 1的位置：

select column1, column2,
       ast_value( a.column3) OVER (PARTITION BY column2, column1 order by start_time ASC
                                   RANGE BETWEEN unbounded preceding and unbounded following
                                  ) as column3,
       column4, column5, column6
from (select a.column1, a.column2, a.start_time,
             rank() OVER (PARTITION BY a.column2, a.column1 order by a.start_time DESC
                         ) as rank,
            a.column3, a.column4, a.column5, a.column6
      from (table2 s INNER JOIN
            table3 t
            ON s.column2 = t.column2 and s.event_time > t.start_time
           ) a
     ) b
where rank = 1

否则，您需要提供有关执行计划以及table2和table3的更多信息以获得更多帮助。

优化大型Postgresql表的嵌套连接窗口函数

1 个答案: