Sql 重写此查询时需要帮助吗

我们每天都在生产中运行此查询 它进行了大量的连接，还使用了配置单元中的窗口函数

我们试图添加一些设置选项，但没有多大帮助

结构是这样的-

SELECT
        C.f1, C.f2, A.f2 ...
FROM (
    SELECT * FROM (
        SELECT T1.*, B.atid, B.a_id,
        ROW_NUMBER() OVER (PARTITION BY T1.wtid, B.atid ORDER BY T1.b_ts DESC) AS RANK_
        FROM T1 AS T1
        JOIN T5 ON T1.t_dt = T5.t_dt
        JOIN T2 B ON T1.wtid = B.wtid and T1.b_ts = B.b_ts
        LEFT OUTER JOIN (SELECT p_cd FROM T3 WHERE PV_TY_CD = 'ORIG_CD') PV
        ON T1.TYP = PV.p_cd
        WHERE T1.state not in ("INVALID")
        AND T1.evt_name NOT IN ('INACTIVE','DORMANT')
        AND ISNULL(PV.p_cd)
    ) T
    WHERE T.rank_ = 1
) A

JOIN (SELECT *, row_number() over (partition by ac_id order by b_ts desc) rank_  
      FROM T4
      WHERE event not in ('CT','UPD')
     ) AS C
  ON A.a_id = C.a_id
AND A.atid = C.ac_id
AND C.rank_ = 1
JOIN T6 ON C.t_dt = T6.t_dt

• 由于我不能忽略任何表（和联接），我的方法是使用聚合函数max用另一个联接替换窗口函数，但我无法重写它
• 此外，我不确定这是否一定会有助于提高绩效，因此任何指导都会对我们有所帮助

---

分析函数通常比使用select max的JOIN执行得更好，因为在分析函数的情况下，您只读取同一个表一次，并且行数计算通过分区进行并行化

尝试重新组合联接和筛选

加入

在where条件下，

ISNULL（PV.p_cd）

减少了T1中的一些行。 在这些条件下，情况也是如此：

WHERE T1.state not in ("INVALID")
        AND T1.evt_name NOT IN ('INACTIVE','DORMANT')

将此联接移动到子查询中，如果它过滤lo，这可能有助于在所有其他联接和行数（）之前减少T1中的数据集：

此外，第一行_数仅在T1和B表上计算：

PARTITION BY T1.wtid, B.atid ORDER BY T1.b_ts DESC

考虑在row_number filter之后连接T5表，如果此连接很重，并且row_number filter正在减少数据集，则再次在子查询中使用filter包装row_number，并连接使用T5过滤的子查询

(--filtered by row_number
select * from
(
 SELECT T1.*, B.atid, B.a_id,
        ROW_NUMBER() OVER (PARTITION BY T1.wtid, B.atid ORDER BY T1.b_ts DESC) AS RANK_
  from
    (select T1.* from T1 
                 left join (SELECT p_cd FROM T3 WHERE PV_TY_CD = 'ORIG_CD') PV 
                           ON T1.TYP = PV.p_cd 
     where T1.state not in ("INVALID")
            AND T1.evt_name NOT IN ('INACTIVE','DORMANT')
            AND ISNULL(PV.p_cd)
    ) as T1 JOIN T2 B ON T1.wtid = B.wtid and T1.b_ts = B.b_ts
) T WHERE T.rank_ = 1
) T --filtered
JOIN T5 ON T1.t_dt = T5.t_d  

这可能会有所帮助，具体取决于您的数据

---

也请阅读：这是：

如果您没有在Tez上运行它，请尝试。查询越复杂，Tez的改进就越多。如果无法改善查询，请调整并行性：

PARTITION BY T1.wtid, B.atid ORDER BY T1.b_ts DESC

(--filtered by row_number
select * from
(
 SELECT T1.*, B.atid, B.a_id,
        ROW_NUMBER() OVER (PARTITION BY T1.wtid, B.atid ORDER BY T1.b_ts DESC) AS RANK_
  from
    (select T1.* from T1 
                 left join (SELECT p_cd FROM T3 WHERE PV_TY_CD = 'ORIG_CD') PV 
                           ON T1.TYP = PV.p_cd 
     where T1.state not in ("INVALID")
            AND T1.evt_name NOT IN ('INACTIVE','DORMANT')
            AND ISNULL(PV.p_cd)
    ) as T1 JOIN T2 B ON T1.wtid = B.wtid and T1.b_ts = B.b_ts
) T WHERE T.rank_ = 1
) T --filtered
JOIN T5 ON T1.t_dt = T5.t_d