Oracle rownum usage for splitting a Table into two equal parts.

Question

Hi All,
I have a table which has like 1.2 billion records and i would have to split the table in two parts for my Archiving needs.Unfortunately that table does not have any unique key or primary key or data stamp which i can rely for.
I would have to use the rownum concept to divide the table.
I am using the below
SELECT * FROM (SELECT ENAME, SAL FROM EMP ORDER BY SAL DESC) WHERE ROWNUM < 5000000;
But the problem is that the table is taking forever to retrieve as it has to do a order by and then retrieve the data as per the where clause.
The question i have is that instead of using a orderby clause to get the same rownum for the row every time, can i directly rely on the fact that the database is read only and the Rownum would remain same even without oder by clause....
Thanks....

Answer 1

WARNING! There is a bug in the code, see EDIT: at bottom of post for details
Justin,
It makes sense that Oracle could order over rowid without sorting, but I see a sort in the explain plan:
SQL> create table t as select 1 as data
  2  from all_objects
  3  where rownum <= 100000;
Table created.
SQL> explain plan for select *
  2  from (select t.*, row_number() over (order by rowid) rn from t)
  3  where rn < 50000;
Explained.
SQL> select * from table(DBMS_XPLAN.DISPLAY);
PLAN_TABLE_OUTPUT
Plan hash value: 327232321
| Id  | Operation                | Name | Rows  | Bytes |TempSpc| Cost (%CPU)| Time     |
|   0 | SELECT STATEMENT         |      | 99651 |  2530K|       |   489   (3)| 00:00:07 |
|*  1 |  VIEW                    |      | 99651 |  2530K|       |   489   (3)| 00:00:07 |
|*  2 |   WINDOW SORT PUSHED RANK|      | 99651 |  2432K|  7056K|   489   (3)| 00:00:07 |
|   3 |    TABLE ACCESS FULL     | T    | 99651 |  2432K|       |    31   (7)| 00:00:01 |
Predicate Information (identified by operation id):
   1 - filter("RN"<50000)
   2 - filter(ROW_NUMBER() OVER ( ORDER BY ROWID)<50000)875820,
What are you doing with the results of the select to archive the table in two pieces? If the archive is in the DB in two seperate tables, multi table insert would be an option:
SQL> create table archive_1 (data number);
Table created.
SQL> create table archive_2 (data number);
Table created.
SQL> explain plan for insert when mod(rn, 2) = 0 then into archive_2 (data) values (data)
  2  else into archive_1 (data) values(data)
  3  select rownum as rn, data
  4  from t;
Explained.
SQL> select * from table(DBMS_XPLAN.DISPLAY);
PLAN_TABLE_OUTPUT
Plan hash value: 828723766
| Id  | Operation             | Name      | Rows  | Bytes | Cost (%CPU)| Time     |
|   0 | INSERT STATEMENT      |           | 99651 |  2530K|    31   (7)| 00:00:01 |
|   1 |  MULTI-TABLE INSERT   |           |       |       |            |          |
|   2 |   INTO                | ARCHIVE_2 |       |       |            |          |
|   3 |   INTO                | ARCHIVE_1 |       |       |            |          |
|   4 |    VIEW               |           | 99651 |  2530K|    31   (7)| 00:00:01 |
|   5 |     COUNT             |           |       |       |            |          |
|   6 |      TABLE ACCESS FULL| T         | 99651 |  1265K|    31   (7)| 00:00:01 |
SQL> insert when mod(rn, 2) = 0 then into archive_2 (data) values (data)
  2  else into archive_1 (data) values(data)
  3  select rownum as rn, data
  4  from t;
100000 rows created.Another option would be to use the last digit of rowid to split the table into two groups, but they will not be equal sized.
SQL> explain plan for select *
  2  from t
  3  where substr(rowid, length(rowid), 1) = upper(substr(rowid, length(rowid), 1))
  4  or substr(rowid, length(rowid), 1) in ('0', '1', '2', '3', '4');
Explained.
SQL> select * from table(DBMS_XPLAN.DISPLAY);
PLAN_TABLE_OUTPUT
Plan hash value: 2153619298
| Id  | Operation         | Name | Rows  | Bytes | Cost (%CPU)| Time     |
|   0 | SELECT STATEMENT  |      | 59025 |  1441K|    98  (71)| 00:00:02 |
|*  1 |  TABLE ACCESS FULL| T    | 59025 |  1441K|    98  (71)| 00:00:02 |
Predicate Information (identified by operation id):
   1 - filter(SUBSTR(ROWIDTOCHAR(ROWID),LENGTH(ROWIDTOCHAR(ROWID)),1)='0
              ' OR SUBSTR(ROWIDTOCHAR(ROWID),LENGTH(ROWIDTOCHAR(ROWID)),1)='1' OR
              SUBSTR(ROWIDTOCHAR(ROWID),LENGTH(ROWIDTOCHAR(ROWID)),1)='2' OR
              SUBSTR(ROWIDTOCHAR(ROWID),LENGTH(ROWIDTOCHAR(ROWID)),1)='3' OR
              SUBSTR(ROWIDTOCHAR(ROWID),LENGTH(ROWIDTOCHAR(ROWID)),1)='4' OR
              SUBSTR(ROWIDTOCHAR(ROWID),LENGTH(ROWIDTOCHAR(ROWID)),1)=UPPER(SUBSTR(ROW
              IDTOCHAR(ROWID),LENGTH(ROWIDTOCHAR(ROWID)),1)))
Note
   - dynamic sampling used for this statement
23 rows selected.
SQL> explain plan for select *
  2  from t
  3  where substr(rowid, length(rowid), 1) <> upper(substr(rowid, length(rowid), 1))
  4  and substr(rowid, length(rowid), 1) not in ('0', '1', '2', '3', '4');
Explained.
SQL> select * from table(DBMS_XPLAN.DISPLAY);
PLAN_TABLE_OUTPUT
Plan hash value: 2153619298
| Id  | Operation         | Name | Rows  | Bytes | Cost (%CPU)| Time     |
|   0 | SELECT STATEMENT  |      | 40627 |   991K|    41  (30)| 00:00:01 |
|*  1 |  TABLE ACCESS FULL| T    | 40627 |   991K|    41  (30)| 00:00:01 |
Predicate Information (identified by operation id):
   1 - filter(SUBSTR(ROWIDTOCHAR(ROWID),LENGTH(ROWIDTOCHAR(ROWID)),1)<>'
              0' AND SUBSTR(ROWIDTOCHAR(ROWID),LENGTH(ROWIDTOCHAR(ROWID)),1)<>'1' AND
              SUBSTR(ROWIDTOCHAR(ROWID),LENGTH(ROWIDTOCHAR(ROWID)),1)<>'2' AND
              SUBSTR(ROWIDTOCHAR(ROWID),LENGTH(ROWIDTOCHAR(ROWID)),1)<>'3' AND
              SUBSTR(ROWIDTOCHAR(ROWID),LENGTH(ROWIDTOCHAR(ROWID)),1)<>'4' AND
              SUBSTR(ROWIDTOCHAR(ROWID),LENGTH(ROWIDTOCHAR(ROWID)),1)<>UPPER(SUBSTR(RO
              WIDTOCHAR(ROWID),LENGTH(ROWIDTOCHAR(ROWID)),1)))
Note
   - dynamic sampling used for this statement
23 rows selected.
SQL> select count(*)
  2  from t
  3  where substr(rowid, length(rowid), 1) = upper(substr(rowid, length(rowid), 1))
  4  or substr(rowid, length(rowid), 1) in ('0', '1', '2', '3', '4');
  COUNT(*)
     59242
SQL> select count(*)
  2  from t
  3  where substr(rowid, length(rowid), 1) <> upper(substr(rowid, length(rowid), 1))
  4  and substr(rowid, length(rowid), 1) not in ('0', '1', '2', '3', '4');
  COUNT(*)
     40758
EDIT:
I realized that I screwed up above. In hind sight I don't know what I was thinking. I was attempting to use X = upper(X) to find the upper case characters A-Z. So the two queries to split rows based on the last character of rowid are not right. I don't have time to fix now, but wanted to leave a note of warning.
Edited by: Shannon Severance on Jul 29, 2011 1:34 AM