OLAP error leading to a corrupted AW during cube build

Question

Hi all
My cube load keeps on crashing and the result is a corrupted AW. Sometimes it happens, sometimes it doesn't, thus it is very hard to pin point the cause of the problem
I would like to emphasize that the Alert log and trace files contain general info about the issue but does not help in resolving the problem.
Let me explain what is happening.
I have 6 cubes A, B, C, D , E and F in "MY_AW" and all the cubes are populated via a single dbm_cube.build call as shown below
DBMS_CUBE.BUILD('CUBE_A, CUBE_B, CUBE_C, CUBE_D, CUBE_E , CUBE F', 'C', FALSE, 5, TRUE, TRUE, FALSE);
Here is a brief description of the cube structure
a) CUBE_A, CUBE_B CUBE_C are partitioned along dimension A_A ( i.e. along the same dimension)
b) CUBE_D is partitioned along dimension B_B ( i.e. along a different dimension)
c) CUBE_E and CUBE_F' are NOT partitioned ( These cubes are really tiny, thus no need partitioning them)
All 6 cubes are loaded on a daily basis. Initial reload load is always without any issues. Then the next , say 3 incremental loads would run successfully too.
After that, any further incremental load generates an Oracle error ( as shown below) and the result is a corrupted AW.
It is really frustrating because the error is inconsistent( i.e does not surface always) , although the cube structure is constant across all loads..
Can anybody figure out what is happening here?
Any help would be highly appreciated.
ERROR at line 1:
ORA-37162: OLAP error
XOQ-01707: Oracle job "JOB$_7041" failed while executing slave build "*CUBE_F* USING (LOAD NO SYNCH, SOLVE) AS OF SCN 11027959160835" with error "37162: ORA-37162: OLAP error
XOQ-00703: error executing OLAP DML command "(AW ATTACH MY_AW RO : ORA-01403: no data found )"
ORA-06512: at "SYS.DBMS_CUBE", line 234
ORA-06512: at "SYS.DBMS_CUBE", line 316
ORA-06512: at line 1
ORA-06512: at "SYS.DBMS_CUBE", line 234
ORA-06512: at "SYS.DBMS_CUBE", line 287
ORA-06512: at line 3
Edited by: user9018701 on Jun 11, 2011 4:50 AM

Answer 1

One recent source of AW corruption is
BUG 8664896 - ORA-20000: ORU-10027: BUFFER OVERFLOW - AW VALIDATE
A characteristic error of this bug (which you may see in the alert log) is
+ORA-00600: internal error code, arguments: [xspggepGenPSErase05], [], [], []+
This is fixed in the most recent 11g patches, as described by metalink note 1078454.1, so apply these may clear up the issue.
I also see that you selecting the 'atomic' option in your call to dbms_cube.build. This means you may be a victim of
BUG 11795047 - USING OLAP "ROLLBACK TO FREEZE" CAN DAMAGE AW
The issue is that if you choose to run an atomic build and any error occurs during a slave process, then your AW may end up being corrupted. Choosing FALSE for the atomic argument would not fix the underlying error, but it may stop the subsequent AW corruption. As of writing the fix for this bug is not part of any public OLAP patch, so you would need to get a one-off patch through a service request.
In either case I think a service request would be appropriate here since this is a serious error. You can mention my name if you like and I will help direct the problem to the appropriate developers.