Partition Record count

Question

Hi,
I have a table which was containing 2,68,86,632 rows. for performance issues i have partition the table using range partition using dbms_redefinition option
I have updated the table stats using
DBMS_STATS.GATHER_TABLE_STATS(USER, 'XXXXXXXXX', CASCADE => TRUE);
Once I completed the process successfully I have giving following SQL the check the row count
select /*+ NO_CPU_COSTING */  count(*) from XXXXXXXXX;
count(*)
26886632
when am querying from user_tab_partition its giving different result from above SQL
SELECT table_name,sum(num_rows) FROM user_tab_partitions where partition_name like 'YYYYYYYY%'
group by table_name
ORDER BY table_name;
TABLE_NAME         SUM(NUM_ROWS)
XXXXXXXXX          26892488
SQL query  from user_tab_partition shows me extra 5856 records then above SQL.
I don't known from where its comes.. Kindly help me trace the extra records where it comes from.
Thanks & Regards
Sami

Answer 1

Hi All,
Thanks all of you posting your suggestion and your valuable time
Finally I got the result my self where does those records came from actually its not extra records its sampling records under estimate method.
I have used
EXEC DBMS_STATS.GATHER_TABLE_STATS(USER, 'XXXXXXX', CASCADE => TRUE);which will work on table estimation by default
after googleing found difference between estimate vs compute
such as
Estimate:
Using the sample N rows fixes issues with regards to the sample scan of the first couple of blocks to get the est. number of rows -- you still have the issue that 14,694 rowids were randomly generated - only 10,694 of which were unique.
compute:
Will calculate actual stat of the table but it take much time when compare to estimate method
now I have used compute method dbms_stat such as
BEGIN
  SYS.DBMS_STATS.GATHER_TABLE_STATS (
      OwnName        => user
     ,TabName        => 'XXXXXXXX'
    ,Estimate_Percent  => 0
    ,Method_Opt        => 'FOR ALL COLUMNS SIZE 1'
    ,Degree            => 4
    ,Cascade           => FALSE
    ,No_Invalidate     => FALSE);
END;
SELECT table_name,sum(num_rows) FROM user_tab_partitions where partition_name like 'GL_ACCT_HIST_PART%'
group by table_name
ORDER BY table_name;
TABLE_NAME                     SUM(NUM_ROWS)
GL_ACCOUNT_HISTORY                  26886632
1 row selected.
select count(*) from gl_account_history; 
  COUNT(*)
  26886632
1 row selected.
Thanks & Regards
Sami.
Edited by: Sami on Sep 30, 2012 2:17 PM