优化PostgreSql查询，以获取根据多组by's找到并需要有限行数的记录总数

Question

在这个查询中，我试图根据给定的过滤器获取记录的总数，并试图在一个查询中获取分页所需的有限数量的数据。在前面的两个步骤中，我首先根据Optimising Select number of rows in PostgreSql for multiple group by fields计算记录数，然后根据页面大小和所需的偏移量在不同的查询中获得所需的行数。

EXPLAIN select count (1) OVER () AS numrows , lower(column1) as column1, column2, column3, column4, column5, column6, column7, column8, sum(column9) as column9, sum(column10) as column10 
from tableName tablename 
where 
    column8 in (SOME 50-100 sets) 
    and column_date >= '2013-09-01' 
    and column_date < '2013-09-30' 
    group by lower(column1), column2, column3, column4, column5, column6, column7, column8 
    ORDER BY column9 desc 
    LIMIT 1000 OFFSET 0

XN Limit  (cost=1000134721702.61..1000134721702.63 rows=1000 width=67)
  ->  XN Merge  (cost=1000134721702.61..1000135118514.84 rows=158724893 width=67)
        Merge Key: sum(column9)
        ->  XN Network  (cost=1000134721702.61..1000135118514.84 rows=158724893 width=67)
              Send to leader
              ->  XN Sort  (cost=1000134721702.61..1000135118514.84 rows=158724893 width=67)
                    Sort Key: sum(column9)
                    ->  XN Window  (cost=107149638.61..113101822.10 rows=158724893 width=67)
                          ->  XN Network  (cost=107149638.61..108340075.31 rows=158724893 width=67)
                                Send to slice 0
                                ->  XN HashAggregate  (cost=107149638.61..108340075.31 rows=158724893 width=67)
                                      ->  XN Seq Scan on tableName tablename  (cost=0.00..67468415.44 rows=1587248927 width=67)
                                            Filter: ((column_date < '2013-09-30'::date) AND (column_date >= '2013-09-01'::date) AND (column8 = ANY ('{SOME 50-100 sets}'::integer[])))

版本细节

PostgreSQL 8.0.2 on i686-pc-linux-gnu, compiled by GCC gcc (GCC) 3.4.2 20041017 (Red Hat 3.4.2-6.fc3), Redshift 1.0.666 

谢谢

Answer 1

对于分页记录返回，没有优化红移。它的目的是最大限度地提高分析查询的性能，这些查询涉及许多记录，但只返回一个很小的输出。

如果您打算对结果进行分页，我强烈建议首先将SELECT的输出发送到临时表。然后，您可以做一个简单的(和快速！)在该表上计数(*)并对其分页，而不强制重新执行查询。