如何在Netezza

Question

我在netezza中有一个100亿行FACT表，我想在一个查询中执行ROW_NUMBER（），MAX（）和SUM（）。当我做同样的查询执行超过3个小时。是否有任何方法可以提高查询性能。表分布在作为分区子句（COLA，COLB，COLC，COLD）一部分的4列上。

样本示例

SUM(STR_QTY) OVER (
                PARTITION BY 
                 COLA
                ,COLB
                ,COLC
                ,COLD
                ) AS SLS_RTRN_QTY
,
SUM(STR_QTY_1) OVER (
                PARTITION BY 
                 COLA
                ,COLB
                ,COLC
                ,COLD
                ) AS VAL_QTY
,MIN(ITM_FST_DT) OVER (
                PARTITION BY COLA
                ,COLB
                ) AS FIRST_DT
,MAX(ITM_LST_DT) OVER (
                PARTITION BY PARTITION BY COLA
                ,COLB
                ) AS LAST_DT

编辑1：原始查询

SELECT a.*
FROM (
    SELECT F.DT_KEY AS DT_KEY
        ,F.COL_KEY AS COL_KEY
        ,F.PCK_ITM_KEY AS PCK_ITM_KEY
        ,F.COLC AS COLC
        ,F.COLD AS COLD
        ,F.COLA AS COLA
        ,F.COLB AS COLB
        ,F.COLC AS COLC
        ,F.SH_QTY AS SH_QTY
        ,SUM(F.SLS_QTY) OVER (
            PARTITION BY F.COLD
            ,F.COLA
            ,F.COLB
            ,F.COLC
            ) AS SLS_QTY
        ,SUM(F.SLS_RTRN_QTY) OVER (
            PARTITION BY F.COLD
            ,F.COLA
            ,F.COLB
            ,F.COLC
            ) AS SLS_RTRN_QTY
        ,SUM(F.PCHSE_QTY) OVER (
            PARTITION BY F.COLD
            ,F.COLA
            ,F.COLB
            ,F.COLC
            ) AS PCHSE_QTY

        ,MAX(F.LST_ML_DT) OVER (
            PARTITION BY F.COLA
            ,F.COLC
            ) AS LST_ML_DT
        ,F.LST_MODFD_DTTM AS LST_MODFD_DTTM
        ,ROW_NUMBER() OVER (
            PARTITION BY F.COLD
            ,F.COLA
            ,F.COLB
            ,F.COLC
            ,F.COLE ORDER BY F.DT_KEY DESC
            ) AS RNK
    FROM FCT_ITEM F
    ) a
WHERE a.RNK = 1;

Answer 1

此查询将导致整个表在COLA和COLB上重新分发。如果分发列的集合是分区列的子集，那么您将无法获得昂贵的重新分配。

作为一般规则，在分发条款中使用尽可能少的列，同时仍保持相当均匀的分布。 如果单独使用COLA或COLB进行均匀分配，那么请选择其中之一。