RE: Converting from physical to logical standby
- From: Carel-Jan Engel <cjpengel.dbalert@xxxxxxxxx>
- To: "Orr, Steve" <sorr@xxxxxxxxxxxx>
- Date: Thu, 30 Sep 2004 21:49:24 +0200
Content-Type: text/plain
Content-Transfer-Encoding: 7bit
Hi Steve,
See inline
On Wed, 2004-09-29 at 23:55, Orr, Steve wrote:
> I know about the datatype "limitations" with LSBs but that's not a
> concern here. (I'm thinking the usage of NCLOB, LONG, LONG RAW, BFILE,
> and ROWID datatypes should be relatively rare.) Other than that what are
> some of the "many restrictions" to which you refer? If datatype failures
> aren't an issue what else could make LSB less dependable than PSB? I've
> had SPOFs with straight RMAN backups but I monitor them daily and
> resolve them so I figure I would do the same if there was a problem with
> an RMAN backup due to a standby DB failure.
I've attached a small script that I use to check whether a database can
be duplicated with LSB.
>
> I know that LSB uses LogMiner SQL Apply under the covers so it's not as
> efficient as straight PSB redo application but the overhead for a
> standby database seems to be so light that I can't imagine it would be
> much of an issue. Has anyone experienced standby database server
> performance problems with LSBs?
I'm preparing a Data Guard performance paper. It will be presented at
Miracle DBF and UKOUG in October/November this year. I'll submit an
abstract for Hotsos 2005 as well, so, maybe I'll present it in the US.
>
> I'm getting ready to test LSBs and switchovers in my test environment. I
> have a Python script which automagically performs the switchover
> complete with error checking and it works great. Any suggestions for my
> testing?
I'm getting curious. I'd ask the same to you, for my performance paper.
For LSB I'd concentrate on the restricted DML/DDL. Violating the
restrictions can stop the SQL Apply process. Monitoring that, and
recovering from that seems the hardest part of LSB management to me.
Furthermore, because all SQL will get applied with all the redo and undo
consequences, performane can be an issue. I haven't digged into it, but
I'd expect more parsing than the primary, and maybe serialization due to
the limted number of SQL Apply processes. This will not hurt the primary
(although supplemental logging won't come for free), but might catch you
when you have to wait for completion at failover time. Making backups
from the LSB with RMAN doesn't cover your primary! It's another
database, not a block-for-block copy. So, partial recovery of the
primary cannot be done based on RMAN backups from LSB.
What requirements do you have?
Normally, when someone needs an open database for reporting and needs a
redundant database, I recommend running two standby's, one LSB and one
PSB.
Best regards, Carel-Jan
>
>
> Steve Orr
> Bozeman, Montana
>
>
> -----Original Message-----
> From: oracle-l-bounce@xxxxxxxxxxxxx
> [mailto:oracle-l-bounce@xxxxxxxxxxxxx] On Behalf Of Carel-Jan Engel
> Sent: Wednesday, September 29, 2004 1:51 PM
> To: Orr, Steve
> Cc: DWILLIAMS@xxxxxxxxxxxxx; oracle-l@xxxxxxxxxxxxx
> Subject: RE: Converting from physical to logical standby
>
>
> Hi Steve,
> Creating LSB from scratch DOES need a coalesce of the primary. So, it is
> inaccessible for a couple of minutes.
>
> My reply on your follow-up question: what version do you have in mind?
> I've worked with both PSB and LSB. One LSB in production, and that has
> been abandoned (it was a mixed setup: two LSB's for publishing web
> content and one PSB for the content management system. It was a
> newspaper). PSB on many sites, and quite succesfull. I notice a slight
> change in naming LSB by Oracle: It's called SQL Apply many times now. It
> might hint you to the fact that it isn't actually a Standby Solution. In
> my honest opinion it is not, at least not in 9iR2. Ita has many
> restrictions, and many failures/wrong datatypes/wrong SQL can stop it,
> leaving you with the job to resync the thing.
>
> Doing RMAN backups on the standby is officially supported, I haven't
> done it though. One reason to create your backups on the primary might
> be that relying on the standby for backups is chaining the
> backup-process: when the standby fails your backup will fail likewise. I
> don't like SPOFs.
>
> Best regards,
>
> Carel-Jan Engel
>
> ===
> If you think education is expensive, try ignorance. (Derek Bok) === On
> Wed, 2004-09-29 at 17:36, Orr, Steve wrote:
>
> > Thanks.
> >
> > RE
> > > converting PSB to LSB is not possible.
> > What about creating LSB from scratch... Can it be done without making
> > the primary inaccessible?
> >
> > Follow up question... Anyone have much experience with LSB vs PSB? I
> > know PSB performs better but is LSB just as dependable? If performance
>
> > is okay with LSB is it otherwise just as good as PSB?
> >
> > Performing a full export on the primary could impact performance but
> > could be quite suitable on a standby. One possible advantage of LSB is
>
> > that you don't have to toggle the recovery/database open status.
> >
> > Other thoughts on PSB vs. LSB? Any reason whatsoever not to perform
> > RMAN backups on the standby?
> >
> >
> > Standing by...
> > Steve Orr
> >
> >
> >
> > -----Original Message-----
> > From: oracle-l-bounce@xxxxxxxxxxxxx
> > [mailto:oracle-l-bounce@xxxxxxxxxxxxx] On Behalf Of Carel-Jan Engel
> > Sent: Tuesday, September 28, 2004 11:58 PM
> > To: DWILLIAMS@xxxxxxxxxxxxx
> > Cc: 'oracle-l@xxxxxxxxxxxxx '
> > Subject: RE: Converting from physical to logical standby
> >
> >
> > Steve,
> > For 9iR2 I attempted to extend my instatiate scripts for Physical
> > Standby (PSB) with Logical Standby (LSB) capabilities. It actually
> > worked, using vanilla hot backups.....4 out of 5 times. The quiesce
> > database is still needed, although it didn't take too long. So, no
> > major outage/standstill was needed. But, a real instantiation 'on the
> > fly' is not possible. AFAIK converting PSB to LSB is not possible.=20
> >
> > Whenever the script failed, for no obvious reason the next run went OK
>
> > most of the times. I haven't put too much effort into solving this,
> > because of the instability and restrictions of LSB, and lack of demand
>
> > for LSB. I'm a poor self-employed contractor ;-)
> >
> > Best regards,
> >
> > Carel-Jan Engel
> >
> > =3D=3D=3D
> > If you think education is expensive, try ignorance. (Derek Bok) =
> > =3D=3D=3D=20
> >
> >
> >
> > On Tue, 2004-09-28 at 19:02, DENNIS WILLIAMS wrote:=20
> >
> >
> > Steve - In answer to question #2, Robert Freeman on page 417 of
> > Oracle9i RMAN Backup & Recovery says "RMAN backups cannot be used to
> > create a logical Standby database, because RMAN cannot quiesce a
> > database". =20 Dennis Williams
> > DBA
> > Lifetouch, Inc.
> >
> >
> > --
> > //www.freelists.org/webpage/oracle-l
> >
> > --
> > //www.freelists.org/webpage/oracle-l
>
>
>
>
> --
> //www.freelists.org/webpage/oracle-l
>
===
If you think education is expensive, try ignorance. (Derek Bok)
===
DBA!ert, Independent Oracle consultancy
Kastanjelaan 61C
2743 BX Waddinxveen
The Netherlands
tel. +31 (0) 182 64 04 28
fax +31 (0) 182 64 04 29
e-mail info.dbalert@xxxxxxxx
--
//www.freelists.org/webpage/oracle-l
Other related posts:
