Re: Questions: transaction log and commit

Stefan Pommerenk <spommere_at_de.oracle.com> wrote:

>Hi,
>you're absolutely right with this reasoning.
>This can happen when having placed the redo logs on
>RAID5 logical drives and having write cache enabled. So a commit won't
>guarantee
>to get the committed tx recorded on disk. In case of a system crash,
>transaction
>recovery could fail due to a incomplete thread recovery (rollforward with the
>redo logs) und could end up with an inconsistent database.
>
>Hope this helps
>
>*******************************************************************************
>* Stefan Pommerenk, Central Technical Support Germany, RDBMS/Languages *
>*-----------------------------------------------------------------------------*
>* Phone: +49(0)89/14350-100 MailId: spommere.DE *
>* Fax: +49(0)89/14977-144 Internet: spommere_at_de.oracle.com *
>* Compuserve: 100427.1516_at_compuserve.com *
>* *
>* --- And,... the ball is gone,... it's a HOMERUN ! --- *
>*******************************************************************************
>

This is true as well. If you are buying disk controllers that provide write caching, they should have a battery backup to hold the cache in case of power failure. Better yet, don't use write caching on the controller and you won't have to worry about any type of failure. The original question was about the OS, not the drive controllers. Oracle does force a write through on every OS I've ever seen.

Steve Received on Sat Jun 01 1996 - 00:00:00 CEST