RE: UNTO TBS behavior in 9i
- From: "Mark W. Farnham" <mwf@xxxxxxxx>
- To: <veeeraman@xxxxxxxxx>, "Hemant K Chitale" <hkchital@xxxxxxxxxxxxxx>
- Date: Wed, 24 Jan 2007 13:48:15 -0500
Presuming you have enough time to test it, what happens when you do only the
select without attempting the insert?
Possibly you could shoot us the statement in question so it could be
examined for possibly side-effect problems.
-----Original Message-----
From: oracle-l-bounce@xxxxxxxxxxxxx [mailto:oracle-l-bounce@xxxxxxxxxxxxx]On
Behalf Of Ram Raman
Sent: Wednesday, January 24, 2007 12:20 PM
To: Hemant K Chitale
Cc: oracle-l
Subject: Re: UNTO TBS behavior in 9i
Thanks for your replies.
1. We are not doing multiple commits. Just one INSERT INTO SELECT statement
with one commit at the end. There were no other transactions when we started
this one.
2. The space for both data and index is about 5.5Gb.
Last time we ran the process, the 30GB UNDO TBS was close to 100% free when
we started the process and the UNDO TBS became 100% full and then erred out
because of lack of space. Why would the undo TBS grow out to be almost 30GB
causing the process to fail esp. when no other process is running in the
database. The only other active processes are Oracle background processes.
Oracle is 9206.
We do have the option of using Direct path insert and turning the logging
off, but this behavior is perplexing.
On 1/23/07, Hemant K Chitale < hkchital@xxxxxxxxxxxxxx
<mailto:hkchital@xxxxxxxxxxxxxx> > wrote:
1. If you are doing multiple commits during the insert run, each
committed batch remains in UNDO for the duration of 2 hours. You should
reduce UNDO_RETENTION.
2. UNDO is maintained for both Table Rows and Index updates -- eg
when you INSERT,
you are adding entries to indexes on the table as well. So UNDO is
maintained for that
as well.
3. You can monitor the size of your transaction by querying
V$TRANSACTION -- that
would show the number of undo "records" (table+index entries) and
undo blocks used
between every commit.
4. You could avoid undo and redo both -- eg
alter table target_table nologging ;
insert /*+ APPEND */ into target_table ....
<finally> commit ; alter table target_table logging;
At 02:19 PM Tuesday, Ram Raman wrote:
>Hi all,
>
>We are running a batch process to insert data from one table to
>another. The data inserted is over than 10 million rows. We had
>about 20GB of UNDO tablespace. UNDO_RETENTION is 7200. When we start
>the insert process no other process runs and all the time during the
>insert process nothing else runs. We insert about 5.5 Gb of data
>including indexes.
>
Hemant K Chitale
http://web.singnet.com.sg/~hkchital
and
http://hemantoracledba.blogspot.com
Other related posts:
