SEO

Hi,
While reading backup and recovery manual, i read the following:
---------------------------------------------------------------------------------------------------------------------------------------
transaction recovery
Transaction recovery involves rolling back all uncommitted transactions
of a failed instance. These are "in-progress" transactions that did not
commit and that Oracle needs to roll back. It is possible for
uncommitted transactions to get saved to disk. In this case, Oracle
uses undo data to reverse the effects of any changes that were written
to the datafiles but not yet committed
---------------------------------------------------------------------------------------------------------------------------------------

Can someone please explain how can uncommited transactions gets saved
to disk.
I assumed only commited data gets saved to disk.

Thanks.
Giridhar Kodakalla.

>>Can someone please explain how can uncommited transactions gets saved
to disk. I assumed only commited data gets saved to disk.

On line redo logs.

HTH

Carlos.

Thanks for your reply. So it means, whether the record is commited or
not,data from online redo logs can be written to to the data
files,right?

Can you please explain when commited data can be flushed to data files
and when uncommited data can be flushed to data files?

Thanks again.
Giridhar Kodakalla.

>>So it means, whether the record is commited or not,data from online redo logs can be written to to the data files,right?

No. Entries in the log buffer are written to the log files.

Commits trigger sync writes.
Background (no-commit) writes are triggered based on SGA variables.

HTH

Carlos.

Adding to this: undo is also written to the redolog, so undo is
protected by redo log.
Apart from a lost redo log, one can always rollback and roll forward.

--
Sybrand Bakker
Senior Oracle DBA

"Giridhar" <gkodakalla@gmail.com> wrote in message
news:1147152845.214758.144960@j33g2000cwa.googlegr oups.com...
> Hi,
> While reading backup and recovery manual, i read the following:
> --------------------------------------------------------------------------
-------------------------------------------------------------
> transaction recovery
> Transaction recovery involves rolling back all uncommitted transactions
> of a failed instance. These are "in-progress" transactions that did not
> commit and that Oracle needs to roll back. It is possible for
> uncommitted transactions to get saved to disk. In this case, Oracle
> uses undo data to reverse the effects of any changes that were written
> to the datafiles but not yet committed
> --------------------------------------------------------------------------
-------------------------------------------------------------
>
> Can someone please explain how can uncommited transactions gets saved
> to disk.
> I assumed only commited data gets saved to disk.
>
> Thanks.
> Giridhar Kodakalla.

Since most transactions are committed and not rolledback Oracle writes
uncommitted transactions to the datafiles. So when you issue a commit then
it doesn't have to write the data to the datafiles. It does have to indicate
that the data is committed.
Jim

On 9 May 2006, miotromailcarlos@netscape.net wrote:
>>>So it means, whether the record is commited or not,data from online
>>>redo logs can be written to to the data files,right?
>
> No. Entries in the log buffer are written to the log files.
>
> Commits trigger sync writes.

I thought Oracle also writes to the disk at any point in the
transaction, completely at its discretion. Uncommitted data can be
found in the datafiles during a transaction. Its the redolog that
Oracle completely depends on to either back these records out or leave
them there or flush them if needed, or not write them to disk until a
later time.

The point being, Oracle can write to disk whenever it deems the most
performant time to do so because it is 100% protected by the redologs.
Is this not correct?

--
Galen Boyer

>>The point being, Oracle can write to disk whenever it deems the most
>>performant time to do so because it is 100% protected by the redologs.
>>Is this not correct?

Typically, DBWn writes to disk dirty blocks on log switches, when there
are few database buffers free, and by log_checkpoint_interval &
log_checkpoint_timeout, but I was referring to log writes(LWR), not to
block writes (DBWn / checkpoints).

HTH

Cheers.

Carlos.