DR Design Using Standby Database

[Former Member]

Hi ,

We have a large SAP landscape based on Oracle . Recently we are planning to move to Oracle Standby Database solution . One of our Vendor provides the following suggestion which I personally do not agree to .

1.For each primary database there is a standby database at alternate site

2.The standby database is created by normal procedure as suggested by Oracle . i.e backup /restore from the primary using standby control file .

3.The primary online redo logfiles are stored in a nfs based storage provided by one of the Top storage vendors

4.The primary DB control files are also stored in in this storage

5.The Offline Archive Files are also on this storage

6.3,4,5 and snapmirroed to the secondary site every 15 minutes

7.In a planned outage the vendor has given a demo as follows

a.Shutdown the Primary , Take the snap of primary and made it available at the standby site

b.Use recover database using standby controlfile until canel which takes oracle offline archives from the snap mirrot

c.After all the archives are applied ,the replace the standby control file , redo logfiles with that of the primary database

d.Then they use recover database until cancel

e This demo works .

f. They again go back to primary by following the above steps

From my Point of view , this does not follow Oracle Recovery Mechanism Rather the vendor is applying some unusual way to sell it's product .

I am asking the experts their opinion on the above scenario . This will really help

Best Regards and Thanks in Advance.

[Former Member]

Hello Stepan,

Thanks for your response . You have fully understood the scenario . I fully want to go with dataguard , however due to some circumstances , this is not getting approved .

We have RPO of 30 minutues , keeping this in mind vendor is suggesting the solution for data loss <30 minutes . My main concern is how someone is replacing the standby control file and log file and make the standby database recognised as primary database .The Oracle has not kept any information of the standby database except in the control file which may sometimes geopardise the data recovery in various scenarios . My biggest fear is it should not corrupt the standby database and make our life hell.

We are not accepting the solution , but still under pressure to adopt to the solution as the demo works .

best Regards,

Frank

[Former Member]

Frank,

This solution might as long as all Oracle datafiles are present in single Netapp FlexVol. The reason is that SnapMirror(replication) works by creating crash-consistent snap of the FlexVol and mirroring it. You are even safer if everything (all data files, control files) is single place[1]

If you have multiple SnapMirrors in place, Snaps are not going to be consistent with each other and for this reason you need to have think about putting Oracle into hot-backup mode before replicating FlexVols... plus you definetley want to be sure you have flushed host kernel memory from in flight I/O (synced the FS/VG metadata and so on). Why? you don't want to replicate Oracle in two parts... (e.g. half of Oracle data files from 11:39 AM and second part from 11:40 AM)...

If i may recommend something then do something like SMSAP/SMO + Snapdrive + Protection Manager and always verify the DR by doing rman validate during testing. My biggest concern is that in this solution is that they are playing hard with control files. IMHO it should be always replicated in storage replication solution... you just connect to SnapMirror destination FlexVols (or FlexClones), mount storage, startup Oracle in mount mode, do the recovery from 2nd FlexVol containing archived redo log files and that's it...

-Jakub.

[1] - but then you may hit capacity limits on Netapp DataOnTAP 7.3.x series...

[stefan_koehler]

Hello Frank,

this sounds like some NetApp appliances

At first i have to say that there is no "bad" solution in general - in a DR design everything depends on your requests and needs.

If i understand your scenario correct - the suggested way is the following:

1) Built up Oracle Standby database with RMAN or manually

2) Mirror the primary control files, primary online redlogs and archivelogs every 15 minutes via storage solution every 15 minutes

What you have to think about in this scenario (without knowing any details):

1) Manual recovery is required after mirroring the archivelogs - otherwise your recovery time will be huge and the storage for archive logs will be exhausting.

2) You will have data loss of round about 15 minutes in case of disaster

By the way you don't need the primary controlfiles on standby site to do a recovery - you just need to do it "the Oracle way".

We already have implented DR scenarios with a SYNC mirror (no data loss) in huge SAP environments (> 4 TB database). The design itself depends on your requirements, Oracle release, the purchased Oracle EE options, amount of redo stream and the network speed / bandwith of course.

In my experience - most of the implementations can be realised with Oracle Data Guard itself and no storage features are needed. The advantage of a "single product" environment (Oracle / Data Guard) is that nearly no manual steps are required.

Regards

Stefan

P.S.: If you need any further (commercial) assistance - check my SDN profile / homepage.