Re: ROLLBACK_SEGMENTS definition

From: Andrew Mobbs <andrewm_at_chiark.greenend.org.uk>
Date: 10 Jul 2002 07:12:03 +0100 (BST)
Message-ID: <VUz*mqVsp@news.chiark.greenend.org.uk>

Howard J. Rogers <dba_at_hjrdba.com> wrote:
>
>"Richard Foote" <richard.foote_at_bigpond.com> wrote in message
>news:DwMW8.31665$Hj3.95560_at_newsfeeds.bigpond.com...
>> Hi Howard,
>>
>> I thought the bug only affected Release 2 ?
>
>That's my understanding too.
>
>Guess which version I'm on????? ;-)
>
>>
>> Can you please forward me the details, access to metalink in not that easy
>> for me at the moment :(
>
>Not really, Richard.... I'm ensconced in the study right now, and the
>details of how I get into Metlink are written on a sticky bit of yellow
>paper attached to my monitor at work.
>
>Oops.
>
>Perhaps some kind soul will oblige us both?

OK, I was running a batch process creating large numbers of records, and everything went wonderfully for some time until they started dying with ORA-600 [kcbnew_3]. After that, many DML and DDL commands failed with the same error (not always, e.g. I could truncate some of the tables I'd been filling, but some truncates would fail). Any attempt to restart the batch processes was doomed to failure.

This persisted across a instance shutdown, and as it affected all data, not just the data I was manipulating, I had a hunch it was something undo related. Creating a new undo tablespace and switching to it allowed me to run fine for a while before the problem reoccurred. Giving up on automatic undo and going back to rollback segments worked.

So, TAR off to Oracle. Bunch of trace files and alert logs later they confirm the bug and confirm the workaround.

Bug No. 2436895
Product Version 9.2.0.1
Affects Platforms Generic
Status Code Bug (Response/Resolution)

06/27/02 06:54 am *** TAR:
. PROBLEM:
1. Clear description of the problem encountered: . Customer runs a data generation batch process built using pro*c. He was running 32 concurrent copies and generating data for benchmark testing. There was no other activity on the system. The processes ran as expected for some time, then failed with ORA-600[kcbnew_3]. After the error occurred, many but not all SQL commands failed with similar errors.

.
2. Pertinent configuration information (MTS/OPS/distributed/etc)

N/A
3. Indication of the frequency and predictability of the problem

Everytime when run with undo tablespaces .
4. Sequence of events leading to the problem

Ran multiple copies of a batch process and received the error .
5. Technical impact on the customer. Include persistent after effects.

Large system benchmark scheduled in 2 weeks at the customer site with IBM. This error is preventing them from proceeding using the desired configuration.

.
DIAGNOSTIC ANALYSIS:

--> Changing undo tablespace does not help --> occurs only in 9.2 as the internal error does not exist in prior releases
.
WORKAROUND:

-->Use Rollback segments
--> Possibly use dictionary manage tablespace for undo information .
RELATED BUGS:

2414972 which is still with development
.
REPRODUCIBILITY:

Everytime at customer site
.
TEST CASE:

not available
.
STACK TRACE:

kcbassertb1

kcbnew+0770
ktugnb+08ec     
ktugur+0814 
ktuchg+03d0

ktbchg2+00a4
kddchg+04a0
kduovw+0f9c
.
.
SUPPORTING INFORMATION:

Internal error seems to have been introduced in 9.2 and seems to be related to bitmap information of the file space
.
24 HOUR CONTACT INFORMATION FOR P1 BUGS:

.
DIAL-IN INFORMATION:

.
IMPACT DATE:

*** 06/27/02 07:14 am *** (CHG: Sta-&gt;16)
*** 06/27/02 07:14 am ***
*** 06/27/02 07:17 am ***
*** 06/27/02 07:24 am *** (CHG: Asg-&gt;NEW OWNER)
*** 06/27/02 07:50 am *** (CHG: Sta-&gt;11 Asg-&gt;NEW OWNER)
*** 06/27/02 07:50 am ***

rdbmsrep -> assigning to 2414972's owner for confirmation

*** 06/28/02 03:21 pm ***
*** 06/28/02 05:39 pm *** (CHG: Sta-&gt;30)
*** 06/28/02 05:39 pm ***
*** 07/03/02 11:44 am *** (CHG: FixBy-&gt;9.2.0.2)
*** 07/08/02 11:09 am *** (CHG: Sta-&gt;11)
*** 07/08/02 11:09 am ***

Received on Wed Jul 10 2002 - 01:12:03 CDT