Re: Checkpoints - Recoverablity
Date: Wed, 2 Dec 2015 19:16:45 -0600
Message-ID: <CAFH+iffvyB4+moGJLeBGLSuFFVpFw+YpzR_OBcpjt8-+-jF8vQ_at_mail.gmail.com>
Just for grins and giggles, thought I post the results of the research I did on this.
When setting event 10356, as previously noted, updates to the controlfile are not written for unrecoverable SCN, but invalidation redo is still maintained. So it seems recoverablity is maintained, the buffer cache functionality does not change and incremental backups will be valid.
On Thu, Nov 5, 2015 at 11:12 AM, David Barbour <david.barbour1_at_gmail.com> wrote:
> Oracle 10.2.0.5/HPUX 11.23 (no flames please, upgrading next month)
>
> Recently, to improve performance, our Data Warehouse folks implemented
> 'nologging' loads as well as amending their table, index and partition
> creations to include the 'nologging' modifier. This of course caused the
> controlfile updates to go nuts with CF Enqueue waits which slowed the whole
> process to a crawl. To alleviate this, we set Oracle event 10356 (per Doc
> ID 1380939.1) which by-passes the controlfile updates. Loads are working
> quite well now but what are the recoverability ramifications?
>
> My what happens with backups? I understand nologging=norecovery and was
> planning on implementing RMAN using BCT and incremental backups following
> the completion of the nightly load cycle.
>
> Checkpoints occur at log switches, at specified intervals or on request
> and update datafile headers as well as the control files. They write
> information about modifications from the SGA buffer cache.
>
> Is there a danger of either having these age out of the buffer cache or
> consuming so much space that it reduces the efficacy of the cache?
>
> Should we increase checkpoint intervals? Will external BCT capture the
> changes so we can get a good incremental backup?
>
>
>
-- http://www.freelists.org/webpage/oracle-lReceived on Thu Dec 03 2015 - 02:16:45 CET