Re: Fssnap and Oracle 9i

"Howard J. Rogers" <hjr_at_dizwell.com> wrote in message news:<4167074d$0$23893$afc38c87_at_news.optusnet.com.au>...
> Roel Toledo wrote:
>
> > Id' like to solicit opinions from other DBA's in this forum about
> > Solaris 9's online backup utility fssnap and using it to backup an
> > Oracle database server. Do the database instances need to be shut
> > down first before issuing fssnap? In the event that fssnap backup is
> > made while the database is up and running, can the backups be safely
> > restored without a problem?
>
> I don't know fssnap, so I could be wrong. But I do know Oracle's internal
> shenanigans that get stuffed by O/S backups done incorrectly, so I'll bet
> I'm not.

http://sunsolve.sun.com/search/document.do?assetkey=1-9-49622-1&searchclause=fssnap

>
> The instances would indeed need to be shutdown before taking your backup.
> The problem is simply a generic one: an operating system utility cannot
> know the significance of the first block in a data file's header. Nor can
> it understand that 16 512 byte blocks need to be copied in synchronism
> before you can say you've backed up one Oracle block. Only an Oracle-aware
> utility can do that (such as RMAN). Therefore, if the instances are still
> running, you risk fractured blocks and inappropriate data file headers.

Isn't that what hot backups are for?

>
> Ana's suggestion to test it is not, in this case, sufficient, I would
> suggest. Because whether or not you actually suffer from fractured blocks
> or inappropriate data file headers is actually a matter of chance. You
> might test it, and it all works perfectly... but that was just good luck.
> When it comes to backup and recovery, you don't take chances. You adopt the
> strategy that you know will work 100% of the time.

Mostly true and correct, but why wouldn't hot backups deal with the data file headers?

>
> In your specific case, go ask Sun (or whoever) whether fssnap is
> Oracle-aware, and if it isn't, then don't use it with running instances.
>

I don't think that is really the right answer. Like most of these snapshot backups, nothing is really copied unless a change is made. In the case of Oracle, it updates the headers of files in read/write tablespaces periodically (on the order of seconds), so the first time that Oracle accesses each of the files, it would be copied. With many G's of files, I would imagine that there would be some waiting, and don't believe the Sun claim of otherwise, since in short order _almost everything_ would have to be physically copied. Which kind of moots the reason for snapping, wouldn't you say? You might as well just get an old backup and recovery book with hot backup scripts in it that use cp. Or get some super-duper mirrors and put everything into backup mode and break the mirrors.

> For Oracle, you should be using RMAN.

Agreed. For the benefit of the OP, RMAN is smarter than anything that uses alter tablespace begin backup, for the simple reason that it doesn't have to write extra redo to account for OS copies that don't know what fractured blocks are. And it is a lot simpler and more powerful than any hot backup scripts, and has internal optimizations to speed things up.

jg

--
@home.com is bogus.
I can think of plenty worse jobs: 
http://www.itmanagersjournal.com/article.pl?sid=04/10/11/1835200