Re: Usage of COMMIT NOWAIT, BATCH
- From: Hemant K Chitale <hkchital@xxxxxxxxxxxxxx>
- To: exriscer@xxxxxxxxx, jonathan@xxxxxxxxxxxxxxxxxx
- Date: Sat, 02 Feb 2008 21:27:37 +0800
From what I understand of these options, the real concern would be
a. Can you identify _which_ updates [insert/update/delete etc] did not
go into the redologs in case of instance failure / media failure
b Can the application re-apply only those specific missing transactions
after restart / recovery.
Hemant K Chitale
At 06:52 PM Saturday, LS Cheng wrote:
Hi Jonathan
I forgot to mention that this is a insert intensive application so 3
seconds would not apply I guess. The insert rate is around 1000 to
2000 per seconds.
I guess the upper bound would be
1/3 log buffer
3 seconds
when the log buffer is full
no?
LSC
On 2/2/08, Jonathan Lewis
<<mailto:jonathan@xxxxxxxxxxxxxxxxxx>jonathan@xxxxxxxxxxxxxxxxxx> wrote:
As much data as had not been protected by the log writer since
it last write - i.e. as much as three seconds of log.
Of course, if you had a long running transaction that took
three hours to complete, and the COMMIT for that transaction
was in that three seconds, then the three hours would be rolled
back on recovery.
Regards
Jonathan Lewis
Hemant K Chitale
http://hemantoracledba.blogspot.com
--
//www.freelists.org/webpage/oracle-l
Other related posts:
