ORA-00600 internal error code

Error Description
In our production server while doing data pump import operation at the end of import it fails with following errors, then the instance crashed and database went down.
Errors in file /var/opt/dumpfile/bdump/shiak_mman_23373.trc:
ORA-00600: internal error code, arguments: [kmgs_pre_process_request_6], [4], [2
], [0], [3], [0xC992C04D0], [], []
Sun Oct 19 18:16:20 2008
MMAN: terminating instance due to error 822
Instance terminated by MMAN, pid = 23373

In the trace file we got the call stack as,
ksedst, ksedmp, ksfdmp, kgerinv, kgeasnmierr, kmgs_pre_process_request, kmgsdrv, ksbabs, ksbrdp, opirip, opidrv, sou2o, opimai_real, main.

Cause of the problem
As we know in Oracle ORA-600 error series are oracle bug. This is oracle Bug 4433838. From alert log we see our MMAN (Memory Manager) process has been terminated. So something wrong happen with oracle MMAN.

This bug actually fires when the initialization parameter SGA_TARGET is set to an exact multiple of 4Gb. So, while setting big SGA be conscious.

A little bit more about this bug is if the SGA_TARGET value in HEXADECIMAL has 00000000 as the last bytes then you can hit this bug.

Detail Explanation of this bug
In out production server we set our SGA_TARGET=48G and this bug fires. We see 48G is multiple of 12 times of 4G. So this bug can error. A little bit elaborate of this issue,
Of 48G we get decimal value as,
SQL> col dec for 9999999999999
SQL> select 48*1024*1024*1024 dec from dual;
DEC
--------------
51539607552

Now let's get hexadecimal value of this number,
SQL> select to_char(51539607552,'XXXXXXXXXXXXX') hex from dual;
HEX
---------------
C00000000

As we see in the hexadecimal representation we get last 8 letters are all 0s; 00000000. The bug happens whenever in hexadecimal representation the 4 bytes have 0. Here in setting of 48G we get last 4 bytes all 0. As for last 0 in binary it represent 0000. So 2 letters of 0 in hexadecimal just form 1 byte(8 bits). So last 8 letters in hexadecimal means we got last 4 bytes 0. And forming this type causes to fire hug.

Solution of the problem

Solution 1)
As bug fires whenever in HEXADECIMAL representation we have 00000000 as the last 4 bytes(8 letters in hex value). So our target is to set SGA_TARGET so that it does not form like this way(4 bytes 0 at last). We can do this if we set our SGA_TARGET not exact multiple of 4G.
Doing set our SGA_TARGET to 47G would happen not the last 4 bytes as all 0s of binary representation.

ALTER SYSTEM SET SGA_TARGET=47G SCOPE=BOTH;
SQL> select to_char(47*1024*1024*1024,'XXXXXXXXXXXXX') hex from dual;
HEX
---------------
BC0000000
We see last 7 letters as 0, and hence the bug will not fire.

Solution 2)
Out database version was 10.2.0.1; base version. And this bug fires. This bug remains in oracle patchset 10.2.0.2 and 10.2.0.3. And is fixed in 10.2.0.4 patchset. So applying patchset 10.2.0.4 would fix the problem.

Solution 3)
Upgrade your database to 11.1.0.6 also will fix the problem.

Solution 4)
Apply the one-off Patch 4433838.

Comments

Post a Comment

Popular posts from this blog

ORA-00923: FROM keyword not found where expected

Problem Symptom While performing a simple select statement it fails with error ORA-00923: FROM keyword not found where expected like below. SQL> select 48*1024*1024*1024 decimal from dual; select 48*1024*1024*1024 decimal from dual * ERROR at line 1: ORA-00923: FROM keyword not found where expected Cause of the problem The ORA-00923 can be caused by two different reasons. 1)Oracle itself expect FROM keyword in the position but it could not find any. Like below, SQL> select 48*1024 om dual; select 48*1024 om dual * ERROR at line 1: ORA-00923: FROM keyword not found where expected As no from keyword found so error arises. Oracle expected from in place of keyword dual. 2)The secong cause can sometimes mislead you. Like in case of SQL> select 48*1024*1024*1024 decimal from dual; select 48*1024*1024*1024 decimal from dual * ERROR at line 1: ORA-00923: FROM keyword not found where exp
Read more

How to make partitioning in Oracle more Quickly

As the table size grows and full table scans happens more frequently then there is no alternative than partitioning in oracle. Partition greatly enhance performance of a query. I will start my partitioning system simply with an example. In my database the following query took 30~35 seconds to complete. After partitioning the performance increased amazingly and it then took only 1 second. My query was, SELECT DISTINCT fs.type, fs.id , fs.pid, fs.cid , fs.cr_id, fs.created_date FROM summary fs where fs.id in (select fa.id from forms fa where fa.sar in (select la.sar from login la where la.login_id=1 and fa.created_date BETWEEN to_date('1-MAY-2008') AND to_date('21-SEP-2008') and fs.created_date BETWEEN to_date('1-MAY-2008') AND to_date('21-SEP-2008') ) )and mode=0; After seeing above query I decide to make range partition on column created_date both in summary table and forms table. Below is the list of procedur
Read more

Copy files between Unix and Windows with rcp

There are many third party tools by which you can copy files between windows and unix machine. Some common tools are winscp, SSHSecureShellClient etc. Instead of using any third party tool, you can use original unix and windows built-in tool to do the task. The RCP (Remote CoPy) is a standard UNIX tool and a Microsoft Windows 2000 tool that copies files/folder to and from computer running the RCP service. Before you can use the RCP tool from a Windows-based computer you must do something in your UNIX computer. you must turn on the RCP service and update the security files to allow the Windows-based computer to connect to your unix machine over this service. The following steps will help you. Step 01: Turn on RCP service on UNIX Host Turn on the RCP service on unix machine by using inetd: 1. Log on as root. 2. Edit the file /etc/Inetd.conf. 3. Uncomment the lines that start with shell and that start with exec. 4. Save the file. 5. Use ps -ef |grep inetd and determine inetd service p
Read more