Re: RMAN rant
- From: John Thomas <jt2354@xxxxxxxxxxxxxx>
- Date: Thu, 28 Oct 2010 22:32:33 +0100
If you want to back up an online Oracle database you can:
- write a script to generate begin backup...
- copy all the datafiles and logfiles to the backup device (split the
mirrors, whatever)...
- end backup for all the tablespaces;
- switch logfiles
- hope your scripts are correct and did not miss any files
OR
- logon to RMAN TARGET / and type BACKUP DATABASE PLUS ARCHIVELOG;
Oh and with the "old-fashioned" method be aware that I/O waits on your redo
logs might go throught the roof for a while. RMAN just rewrites "fractured"
blocks.
Things get a bit more complicated if you want to select incremental or
incrementally refreshed image backups, but what external tool could do that,
regardless of requiring four or five more keywords RMAN needs.
But the FRA and retention policy/windows can also manage your archive logs
to suit your business's requirement. Or you could write complex, potentially
unreliable scripts in Korn-shell/PERL/Windows (do pull the other one) to do
the same.
The gap Oracle *is* missing is in promoting the universal use of the
database as a storage platform.
RMAN/Oracle has been able to do Flashback to a point in time, or efficient,
change-only backups for years.
But for nearly every site recovering the database to DD-MON-YY HH:MI:SS is
no use unless you can also recover the external files to the same point in
time.
So the SAN providers provide proprietary techniques to roll the entire
filesystem back to any point in time. Maybe it works. Maybe there are no
bugs in the de-duplication technology. Or bugs in the time-consistent backup
of datafiles from the logical volume manager, .
It's a subtle difference, Oracle has the odd bug or two from time-to-time.
But serious bugs in recoverability very rarely get past beta testers.
Serious bugs in SAN-based logging systems primarily built to recover
unstructured data may go unnoticed. Until far too late.
Cheers,
JT
On 28 October 2010 10:59, Nuno Souto <dbvision@xxxxxxxxxxxx> wrote:
> Stephane Faroult wrote,on my timestamp of 28/10/2010 7:08 AM:
>
> I have just tried to compile all the RMAN keywords (by "keyword" I mean
>> any
>> terminal term without space that you find in all those wonderful
>> syntactical
>> diagrams that brighten up our reading of Oracle reference docs).
>>
>
> You mean you FOUND *any* syntactical diagrams for RMAN?
>
>
>
> I like to quote to developers the "Art Poétique" of Boileau, in particular
>> things such as:
>> “What is conceived well is expressed clearly.”
>>
>
> Bingo!
>
>
> RMAN is not too bad in 10gr2. That is indeed the first release in which
> it's made any sense to me to use it.
> Previously I simply could not subscribe to the notion that I'd have to
> learn an entire new language at every point release of Oracle in order to
> use a backup product...
>
> And that feeling still goes for a lot of other features in the latest
> releases of Oracle. ASM and the cacophony of IO methods and IO interfaces
> comes to mind.
>
> No: the "easy and familiar SQL" used to command ASM is NOT easy nor is it
> FAMILIAR! And don't get me started on the hidden logging of the thing!
> Nor the multitude of "cluster-specific" file systems and cryptic stuff put
> out by Oracle in the last 5 years.
>
> Rather than re-inventing the wheel at every new release in order to
> increase revenue from training and "certifications", it'd do Oracle better
> to stabilize its current range, commands and features.
>
> M$ and IttyBittyMachines did it, why can't they?
> Ah well...
> --
> Cheers
> Nuno Souto
> dbvision@xxxxxxxxxxxx
>
> --
> //www.freelists.org/webpage/oracle-l
>
>
>
--
Cheers,
John
Other related posts:
