"Howard J. Rogers" <hjr_at_dizwell.com> wrote in message news:<40586171$0$3958$afc38c87_at_news.optusnet.com.au>...
> "robson" <dbanewbie_at_yahoo.com.br> wrote in message
> news:514f866a.0403170543.391d00a2_at_posting.google.com...
> > hi, i am brasilian and i know a little inglish.
> > i am working with oracle9i RAC (Oracle9i Enterprise Edition Release
> > 9.2.0.3.0 - Production) and a don't understand about the recover in
> > RAC. what is the best configuration to archived log, i used OCFS. each
> > node should have the directory to write archived log or can use a OCFS
> > to write all archived log of all nodes in the same local.
> > thanks.
>
>
> A lot of RACs use raw devices as the shared disk. But archived redo logs
> cannot ever be written to raw devices. Therefore, such RACs cannot share
> their archives by simply writing them to the shared disk. And that's a
> problem, since during a media recovery, the recovering instance will need to
> see all archives produced by all instances. Therefore, when you RAC with raw
> you have to cheat a little, by getting each instance to write its archives
> to an NFS mount point, which is then visible to the other instances.
>
> But you are in the happy position of using a Cluster File System, and
> archived redo logs *can* be written to a CFS. Therefore, you can get each
> instance to archive its thread of redo onto the one, shared, hard disk
> array. No NFS stuff is involved, since that disk array will already be
> physically shared between nodes.
>
> Can all instances write their archives to the same *directory* on the shared
> disk array? Absolutely: not a problem. *PROVIDED* that one instance's
> archives don't threaten to over-write those already written there by another
> instance. For example, with a log_archive_format=arch%s.rdo, you'll have
> problems. Instance 1's log sequence 7 will merrily over-write instance 2's
> log sequence 7.
>
> Which is why the log_archive_format parameter should always include the "%t"
> or "%T" variable as well, with an underscore. In other words, set
> log_archive_format=arch%s_%t.rdo This way, you'll have log "arch7_1.rdo" and
> log "arch7_2.rdo" from each of two instances, and there is no danger of
> over-writing.
>
> I'm not sure if that was exactly what you were asking, but I hope it helps a
> bit anyway.
>
> Regards
> HJR
thanks about your explanation.
i have more questions.
I use OCFS, and am thinking about creating two partition to record
archived log, one for each instance.
in this way I I would have that to define archive_log_dest_2 equal to
archive_log_dest_1 of the other instance.
it is ok?
Received on Thu Mar 18 2004 - 09:20:27 CST
