Question

我正在使用以下查询。由于我试图自我加入的数据非常庞大，因此需要花费大量时间。有人可以指导我如何优化这个查询吗？

我也在考虑添加指数。我总共有19列。我每个表中有大约1000,000行一个月。有人可以建议解决这个问题的最佳方法吗？

解释计划：

OPERATION    OBJECT_NAME     CARDINALITY     COST 

 SELECT STATEMENT 
     5222342     34282 

 SORT 
     1     

 PX COORDINATOR 


 PX SEND 
 :TQ10000    1     

 SORT 
     1     

 PX BLOCK 
     18      466 

 TABLE ACCESS 
 SUCCESS_SIXMONTHS_JUL   18      466 

 Filter Predicates 

 AND 

 COLUMN14=:B1 

 COLUMN7=:B2 

 COLUMN13>=:B3 

 SORT 
     1     

 PX COORDINATOR 


 PX SEND 
 :TQ20000    1     

 SORT 
     1     

 PX BLOCK 
     18      466 

 TABLE ACCESS 
 SUCCESS_SIXMONTHS_JUL   18      466 

 Filter Predicates 

 AND 

 COLUMN14=:B1 

 COLUMN7=:B2 

 COLUMN13>=:B3 

 PX COORDINATOR 


 PX SEND 
 :TQ30001    5222342     34282 

 HASH 
     5222342     34282 

 PX RECEIVE 
     5222342     34282 

 PX SEND 
 :TQ30000    5222342     34282 

 HASH 
     5222342     34282 

 PX BLOCK 
     5222342     490 

 TABLE ACCESS 
 START_SIXMONTHS_JUL     5222342     490

SQL：

SELECT
  DISTINCT
  StMT.id1
  , TIMESTAMP_for_start_message
  , (SELECT MIN(TIMESTAMP_for_success_message)
     FROM SuccessMessageTable
     WHERE
       (id1 = StMT.id1)
       AND (someDate = StMT.someDate)
       AND (jobID = StMT.jobID)
       AND (TIMESTAMP_for_success_message >= StMT.TIMESTAMP_for_start_message)) TIMESTAMP_for_success_message
, (SELECT MIN(seconds_for_success_message)
     FROM SuccessMessageTable
     WHERE
       (id1 = StMT.id1)
       AND (someDate = StMT.someDate)
       AND (jobID = StMT.jobID)
       AND (TIMESTAMP_for_success_message >= StMT.TIMESTAMP_for_start_message)) seconds_for_success_message
  , StMT.someDate
  , StMT.jobID
FROM StartMessageTable StMT
ORDER BY id1, jobID, TIMESTAMP_for_start_message;

Answer 1

对于与性能调优相关的问题，您应该始终至少提出execution plan。

首先，您可以使用subquery factoring重写查询。如果您多次使用子查询，最好将其用作WITH子句。您不必多次重新定义相同的子查询。相反，我们只使用WITH子句中定义的查询名称，使查询更容易阅读。

例如，

WITH DATA AS(
SELECT MIN(SMT.TIMESTAMP_for_success_message)
     FROM SuccessMessageTable SMT, StartMessageTable StMT
     WHERE
       (SMT.id1 = StMT.id1)
       AND (SMT.someDate = StMT.someDate)
       AND (SMT.jobID = StMT.jobID)
       AND (SMT.TIMESTAMP_for_success_message >= StMT.TIMESTAMP_for_start_message)
)
SELECT ... FROM DATA A, table1 b, table2 c
...

例如，我修改了您的子查询以从SuccessMessageTable和StartMessageTable表中获取数据。此temporary结果集可用于join与其他表一起获取所需的行。

通过执行此操作，子查询结果集将被提取一次并解析为temporary table。因此，对子查询的重复引用可能更有效，因为可以从临时表中轻松检索数据，而不是被每个引用重复查询。

详细了解http://oracle-base.com/articles/misc/with-clause.php

修改

我认为以下查询应该可以正常使用 -

SELECT stmt.id, stmt.somedate, stmt.jobid, stmt.timestamp_for_start_message, min(smt.timestamp_for_success_message) timestamp_for_success_message, min(smt.seconds_for_success_message) seconds_for_success_message , FROM successmessagetable smt, startmessagetable stmt WHERE ( smt.id1 = stmt.id1) AND ( smt.somedate = stmt.somedate) AND ( smt.jobid = stmt.jobid) AND ( smt.timestamp_for_success_message >= stmt.timestamp_for_start_message) GROUP BY stmt.id, stmt.somedate, stmt.jobid, stmt.timestamp_for_start_message ORDER BY stmt.id1, stmt.jobid, smt.timestamp_for_start_message;

Answer 2

我认为这相当于你的要求。我不相信DISTINCT仍然是必要的。

SELECT DISTINCT
       M.ID1,
       M.Timestamp_for_start_message,
       MIN(S.Timestamp_for_success_message) Timestamp_for_success_message,
       MIN(S.Seconds_for_success_message) Seconds_for_success_message
       M.SomeDate,
       M.JobID
  FROM StartMessageTable M
  JOIN SuccessMessageTable S
    ON S.ID1 = M.ID1
   AND S.SomeDate = M.SomeDate
   AND S.JobID = M.JobID
   AND S.Timestamp_for_success_message >= M.Timestamp_for_start_message
 GROUP BY M.ID1, M.Timestamp_for_start_message, M.SomeDate, M.JobID
 ORDER BY M.ID1, M.JobID, M.Timestamp_for_start_message;

请注意，并不清楚＆＃39; Seconds_for_success_message＆＃39;值将来自＆Timehuamp_for_success_message＆＃39; Timestamp_for_success_message＆＃39;值。它可能会，但查询的结构并不能保证 - 但问题中的查询也是如此。

优化自我加入

2 个答案: