Tuning an SQL query

Introduction

I had a complex SQL query that was used to extract data from my system and send it across via web request to an external system. The query was taking unduly longer time to run. I made few changes to the query structure and reduced it run time to 7 mins. A part of the main query that was actually causing the delay has been pasted in the sections below.

Problem Description

One of my external systems reported that one of the queries that run at my end have failed with the error below:

ORA-20001: Error - WebService Request Failed. ORA-29276: transfer timeout.

When I tried running the SELECT query in question at my end (on Toad), the query kept running after I aborted it after almost 90 minutes.

The central part of the original query is as below. This part was causing the entire query to slow down. The tables TEAM, TIMESHEET, TIMEENTRY and ASSIGNMENT are considerably huge tables with billions of records. The explain plan showed a full table access on ASSIGNMENT and TIMESHEET tables.

The code

Problem

Code:

SELECT TP.START STARTED ,
TP.FINISH FINISHED,
TP.ID PERIOD,
TM.PROJECTID PROJECTID,
TM.ID TEAMID,
TM.RESOURCEID RESOURCEID,
SUM(TE.ACT) ACT 
FROMTASK TK,
TEAM TM,
ASSIGNMENT ASM,
TIMEENTRY TE,
PERIOD TP,
(SELECTMAX(A.ID)ID,PERIODID,RESOURCEID
FROM TIMESHEET A, PERIOD B 
WHERE STATUS IN(1,3,4)AND ADJUSTMENT =0
ANDA.periodid = B.id
ANDTRUNC(B.START)>=TRUNC(NEXT_DAY(NVL(NULL,SYSDATE)-28,'MONDAY'))
GROUPBY PERIODID,RESOURCEID)TS 
WHERE TM.PROJECTID IN(SELECT PRJ.ID
FROM PROJECTS PRJ,
CU_PROJECT CU,
PRJ_ASSOC AS,
PRJ_UNITS OBS,
PRJ_TYPES TYPE
WHEREAS.TABLE_NAME ='PROJECTS'
ANDAS.RECORD_ID = PRJ.ID
ANDAS.UNIT_ID = OBS.ID
AND OBS.TYPE_ID=TYPE.ID
ANDTYPE.NAME='TEST1'
AND CU.ID=PRJ.ID
AND CU.PRT_CODE='TEST2')
AND TK.PROJECTID=TM.PROJECTID 
AND ASM.TASKID=TK.ID
AND ASM.RESOURCEID=TM.RESOURCEID
AND TE.TIMESHEETID=TS.ID
AND TS.PERIODID=TP.ID
AND TE.ASSIGNID=ASM.ID
ANDTRUNC(TP.START)>=TRUNC(NEXT_DAY(NVL(NULL,SYSDATE)-28,'MONDAY'))
GROUPBY TP.START,TM.ID,TM.RESOURCEID,TM.PROJECTID,TP.ID,TP.FINISH

I changed the query structure as below. This query took 7 mins to run, a considerable improvement from the old query:

Code:

 
WITH TS AS(SELECT/*+ materialize */MAX(A.ID)ID,PERIODID,RESOURCEID
FROM TIMESHEET A, PERIOD B 
WHERE STATUS IN(1,3,4)AND ADJUSTMENT =0
ANDA.periodid = B.id
ANDTRUNC(B.START)>=TRUNC(NEXT_DAY(NVL(NULL,SYSDATE)-28,'MONDAY'))
GROUPBY PERIODID,RESOURCEID)
SELECT TP.START STARTED ,
TP.FINISH FINISHED,
TP.ID PERIOD,
TM.PROJECTID PROJECTID,
TM.ID TEAMID,
TM.RESOURCEID RESOURCEID,
SUM(TE.ACT) ACT 
FROMTASK TK,
TEAM TM,
ASSIGNMENT ASM,
TIMEENTRY TE,
PERIOD TP,
(SELECT PRJ.ID
FROM PROJECTS PRJ,
CU_PROJECT CU,
PRJ_ASSOC AS,
PRJ_UNITS OBS,
PRJ_TYPES TYPE
WHEREAS.TABLE_NAME ='PROJECTS'
ANDAS.RECORD_ID = PRJ.ID
ANDAS.UNIT_ID = OBS.ID
AND OBS.TYPE_ID=TYPE.ID
ANDTYPE.NAME='TEST1'
AND CU.ID=PRJ.ID
AND CU.PRT_CODE='TEST2')S,
WHERE TK.PROJECTID=TM.PROJECTID 
AND ASM.TASKID=TK.ID
AND ASM.RESOURCEID=TM.RESOURCEID
AND S.ID= TM.PROJECTID
AND TE.TIMESHEETID=TS.ID
AND TS.RESOURCEID = TM.RESOURCEID
AND TS.PERIODID=TP.ID
AND TE.ASSIGNID=ASM.ID
AND TP.START>=TRUNC(NEXT_DAY(NVL(NULL,SYSDATE)-28,'MONDAY'))
GROUPBY TP.START,TM.ID,TM.RESOURCEID,TM.PROJECTID,TP.ID,TP.FINISH

Using the WITH clause with materialize hint considerably improved the run time of the sub-query with a GROUP by clause. Moreover, the second sub-query which was joined in a WHERE clause earlier, was used as a table in the new query and this also helped increasing the performance.

More readable code by Nimesh

Code:

SELECT TP.START STARTED
     , TP.FINISH FINISHED
     , TP.ID PERIOD
     , TM.PROJECTID PROJECTID
     , TM.ID TEAMID
     , TM.RESOURCEID RESOURCEID
     , SUM(TE.ACT) ACT
  FROM TASK TK
     , TEAM TM
     , ASSIGNMENT ASM
     , TIMEENTRY TE
     , PERIOD TP
     , (SELECT MAX(A.ID) ID
             , PERIODID
             , RESOURCEID
          FROM TIMESHEET A
             , PERIOD B
         WHERE STATUS IN (1,3,4)
           AND ADJUSTMENT = 0
           AND A.periodid = B.id
           AND TRUNC(B.START) >= TRUNC(NEXT_DAY(NVL(NULL,SYSDATE)-28,'MONDAY'))
         GROUP BY PERIODID
             , RESOURCEID
       ) TS
 WHERE TM.PROJECTID IN (SELECT PRJ.ID
                          FROM PROJECTS PRJ
                             , CU_PROJECT CU
                             , PRJ_ASSOC AS
                             , PRJ_UNITS OBS
                             , PRJ_TYPES TYPE
                         WHERE AS.TABLE_NAME = 'PROJECTS'
                           AND AS.RECORD_ID = PRJ.ID
                           AND AS.UNIT_ID = OBS.ID
                           AND OBS.TYPE_ID=TYPE.ID
                           AND TYPE.NAME='TEST1'
                           AND CU.ID=PRJ.ID
                           AND CU.PRT_CODE='TEST2'
                       )
   AND TK.PROJECTID = TM.PROJECTID 
   AND ASM.TASKID = TK.ID
   AND ASM.RESOURCEID = TM.RESOURCEID
   AND TE.TIMESHEETID = TS.ID
   AND TS.PERIODID = TP.ID
   AND TE.ASSIGNID = ASM.ID
   AND TRUNC(TP.START) >= TRUNC(NEXT_DAY(NVL(NULL,SYSDATE)-28,'MONDAY'))
 GROUP BY TP.START
     , TM.ID
     , TM.RESOURCEID
     , TM.PROJECTID
     , TP.ID
     , TP.FINISH

Sameer

 

thanks dear admin

 

Nice work

A kind request to format the codes in a more readable format, from next time.

-Nimesh

 

how to make it more readable?

 

Indentation?

 

I follow this kind of Indentation, which is little different from what other use
And when writing the codes, use fixed with font so the indentation will appear properly.

Code:

SELECT TP.START STARTED
     , TP.FINISH FINISHED
     , TP.ID PERIOD
     , TM.PROJECTID PROJECTID
     , TM.ID TEAMID
     , TM.RESOURCEID RESOURCEID
     , SUM(TE.ACT) ACT
  FROM TASK TK
     , TEAM TM
     , ASSIGNMENT ASM
     , TIMEENTRY TE
     , PERIOD TP
     , (SELECT MAX(A.ID) ID
             , PERIODID
             , RESOURCEID
          FROM TIMESHEET A
             , PERIOD B
         WHERE STATUS IN (1,3,4)
           AND ADJUSTMENT = 0
           AND A.periodid = B.id
           AND TRUNC(B.START) >= TRUNC(NEXT_DAY(NVL(NULL,SYSDATE)-28,'MONDAY'))
         GROUP BY PERIODID
             , RESOURCEID
       ) TS
 WHERE TM.PROJECTID IN (SELECT PRJ.ID
                          FROM PROJECTS PRJ
                             , CU_PROJECT CU
                             , PRJ_ASSOC AS
                             , PRJ_UNITS OBS
                             , PRJ_TYPES TYPE
                         WHERE AS.TABLE_NAME = 'PROJECTS'
                           AND AS.RECORD_ID = PRJ.ID
                           AND AS.UNIT_ID = OBS.ID
                           AND OBS.TYPE_ID=TYPE.ID
                           AND TYPE.NAME='TEST1'
                           AND CU.ID=PRJ.ID
                           AND CU.PRT_CODE='TEST2'
                       )
   AND TK.PROJECTID = TM.PROJECTID 
   AND ASM.TASKID = TK.ID
   AND ASM.RESOURCEID = TM.RESOURCEID
   AND TE.TIMESHEETID = TS.ID
   AND TS.PERIODID = TP.ID
   AND TE.ASSIGNID = ASM.ID
   AND TRUNC(TP.START) >= TRUNC(NEXT_DAY(NVL(NULL,SYSDATE)-28,'MONDAY'))
 GROUP BY TP.START
     , TM.ID
     , TM.RESOURCEID
     , TM.PROJECTID
     , TP.ID
     , TP.FINISH

Did you use the same query exactly, as I found lot of syntactical errors? :crazy:

 

You can make some more changes (which I don't prefer )
like this,

Code:

SELECT PRJ.ID
  FROM PROJECTS   PRJ
     , CU_PROJECT CU
     , PRJ_ASSOC  AS
     , PRJ_UNITS  OBS
     , PRJ_TYPES  TYPE
 WHERE AS.TABLE_NAME = 'PROJECTS'
   AND AS.RECORD_ID  = PRJ.ID
   AND AS.UNIT_ID    = OBS.ID
   AND OBS.TYPE_ID   = TYPE.ID
   AND TYPE.NAME     = 'TEST1'
   AND CU.ID         = PRJ.ID
   AND CU.PRT_CODE   = 'TEST2'

 

hmm nw i understand...thx

 

I resolved the syntactical errors from 1st code, hopefully all

 

thank you nimesh

 

Any tips on how to use joins and unions effectively in SQL queries when it comes to play with multiple tables?

 

thanks to you too dear admin

 

Thanks to share this information.

 

Indentation?