• Status: Solved
  • Priority: Medium
  • Security: Public
  • Views: 610
  • Last Modified:

Merge performance

Hi, i have a merge statement which taking lots of time to execute, below is the sample of it

OldTable --> Column (ColA, ColB, ColC, ColD, ColE, Value)
               - Index on ColA until ColE
NewTable --> Column (ColA, ColB, ColC, ColD, ColE, Type, Value)
              - Index on ColA until ColE

-- Merge Newtable into Oldtable

MERGE INTO OldTable Old
USING (SELECT ColA, ColB, ColC, ColD, ColE, Type, value FROM NewTable) New      
ON (Old.ColA = New.ColA AND Old.ColB = New.ColB AND Old.ColC = New.ColC And Old.ColD = New.ColD AND (Old.ColE = New.ColE OR (Old.ColE IS NULL AND New.ColE IS NULL)))

WHEN MATCHED THEN
   UPDATE SET Old.value =
         (CASE WHEN New.Type = 1 THEN Old.value + New.value
                  WHEN New.Type = 2 THEN
                             (CASE WHEN New.Value < Old.Value THEN New.Value END)
                  ELSE  Old.Value
        END)
WHEN NOT MATCHED THEN                
   INSERT (Old.ColA, Old.ColB, Old.ColC, Old.ColD, Old.ColE, Old.Value)
   VALUES(New.ColA, New.ColB, New.ColC, New.ColD, New.ColE, New.value);

Anyone know how to optimize it? Thanks
0
yuching
Asked:
yuching
  • 4
  • 2
  • 2
1 Solution
 
sujith80Commented:
What indexes do you have on OldTable ?
Try to have a composite index on (colA, colB, colC, colD).
0
 
yuchingAuthor Commented:
i have a composite index on (colA, ColB, ColC, ColD, ColE) on old and new table.
0
 
yuchingAuthor Commented:
But i dont have an index on Newtable. Type , is this will affect the performance as i used it in case when?
And the order of the colA, ColB, ColC --> ColE for the index is not based on the most granulate one. In this case, ColC is more unique as compare to ColA, should i put colC as 1st order of the composite index, how it will affect the performance?
0
Never miss a deadline with monday.com

The revolutionary project management tool is here!   Plan visually with a single glance and make sure your projects get done.

 
sujith80Commented:
Analyze the oldTable.
Post the execution plan.
0
 
yuchingAuthor Commented:
Execution Plan
----------------------------------------------------------
   0      MERGE STATEMENT Optimizer=CHOOSE (Cost=24515 Card=8168 Bytes
          =4761944)

   1    0   MERGE OF 'NEWTABLE'
   2    1     VIEW
   3    2       NESTED LOOPS (OUTER) (Cost=24515 Card=8168 Bytes=2352384)
   4    3         TABLE ACCESS (FULL) OF 'OLDTABLE' (Cost=11 Card=8168 Bytes=1200696)
   5    3         VIEW (Cost=3 Card=1 Bytes=141)
   6    5           TABLE ACCESS (BY INDEX ROWID) OF 'NEWTABLE'  (Cost=3 Card=1 Bytes=59)
   7    6             INDEX (RANGE SCAN) OF 'INDEX_U01' (UNIQUE) (Cost=2 Card=1)
0
 
yuchingAuthor Commented:
hi, my resolution

MERGE INTO OldTable Old
USING (SELECT ColA, ColB, ColC, ColD, ColE, Type, value FROM NewTable) New      
ON (Old.ColA = New.ColA AND Old.ColB = New.ColB AND Old.ColC = New.ColC And Old.ColD = New.ColD AND
---Replace this line
--(Old.ColE = New.ColE OR (Old.ColE IS NULL AND New.ColE IS NULL)))
(NVL(Old.ColE, '') = NVL(New.ColE, '') ))

WHEN MATCHED THEN
   UPDATE SET Old.value =
         (CASE WHEN New.Type = 1 THEN Old.value + New.value
                  WHEN New.Type = 2 THEN
                             (CASE WHEN New.Value < Old.Value THEN New.Value END)
                  ELSE  Old.Value
        END)
WHEN NOT MATCHED THEN                
   INSERT (Old.ColA, Old.ColB, Old.ColC, Old.ColD, Old.ColE, Old.Value)
   VALUES(New.ColA, New.ColB, New.ColC, New.ColD, New.ColE, New.value);
0
 
modus_operandiCommented:
Closed, 500 points refunded.
modus_operandi
EE Moderator
0
 
modus_operandiCommented:
Question PAQ'ed with refund of points.
modus_operandi
EE Moderator
0

Featured Post

Never miss a deadline with monday.com

The revolutionary project management tool is here!   Plan visually with a single glance and make sure your projects get done.

  • 4
  • 2
  • 2
Tackle projects and never again get stuck behind a technical roadblock.
Join Now