cancel
Showing results for 
Search instead for 
Did you mean: 

Crash consistent backups

jamal_asi
Participant
0 Kudos

We create snapshots(known as replays on the Compellent disk array) on our local SAN and then replicate these snapshots to our DR SAN. These snapshots contain all SAP and Oracle file systems. The snapshots are created while SAP and Oracle are up and running.

We have tested these snapshots and we were able mount these snapshots on a different server and the instance came up.

We also tested these snapshots where we mounted the snapshots of all the sapdata directories from time N and then mounted the snapshots of the directories of the online and offline redolog files  and the control file from time N+t where t>0 and then recovered using the command: RECOVER AUTOMATIC DATABASE. We were able to recover to time N+t.

I was wondering if somebody is using this "Crash Consistent" recovery and if there are any comments or suggestion that some may have.

Thanks.

Accepted Solutions (0)

Answers (3)

Answers (3)

volker_borowski2
Active Contributor
0 Kudos

Hi,

in my opinion you lack a couple of things

(which you might in fact perhaps do, but did not write them here).

Are you doing anything around the snapshots?

Like "begin backup" / "end backup" or any kind of suspending writes

in a consistent manner accross all filesystems involved?

(i.e. snapmounting oraarch while a redolog is half way written is a bad idea)

In addidtion, check this

http://www.oracle.com/technetwork/database/features/availability/rman-fra-snapshot-322251.html

Volker

jamal_asi
Participant
0 Kudos

Hi Volker,

I'm not doing anything around the time of the snapshots. The database is not in hot backup mode while the snapshot is performed(begin backup / end backup are not invoked). The snapshots are performed while everything is running and production is in progress.

You have a good point about an offline redo log file which is snapshoted before it is closed, but isn't that included in the online redo log file which is still not deleted because it is not finished writing to the offline redo in the oraarch directory? The online redo log files are snapshoted along the offline redo log files and of course the control file is snapshoted along too.

Let me know what you think.

Thanks,

thiago_cavalheiro
Active Participant
0 Kudos

Hi Jamal,

I wonder how you were able to restore this backups if they were taken with the database not being in backup mode. I supposed you had done that.

Yes, I am with them. The database needs to be in backup mode for that to work. No backup can be consistent by simply copying the datafiles with the database up.

Regards,

Thiago

jamal_asi
Participant
0 Kudos

Hi Thiago,

Take a look at Oracle document 604683.1 and let me know what you think. This document addresses the possiblity of snapshotting a database while it is NOT in backup mode.

Let me know what you think.

Thank.

thiago_cavalheiro
Active Participant
0 Kudos

Hi Jamal,

Thanks for the information.

Yes, this is something that you need validation from your SAN vendor. This is a backup method that's not supported by SAP, even though it is permitted.

By as I said earlier, for maintaining a backup strategy all you need to do is test it frequently to make sure it is consistent. Also avoid any issues in case you have a crash and need to use it. It is good to have a document to guide on the restore process as well.

Best regards,

Thiago

Former Member
0 Kudos

Hi Janal,

>> The snapshots are created while SAP and Oracle are up and running.

The tablespaces are in the backup mode while creating the snapshots, isn't it?

Because all the configurations and files have been backed up in the snapshot, it will work properly. The problem may arise just only when you restore a single file from the snapshot, such as an offline redolog file. Many devices support this feature, but to be safe side check this option on your device.

Best regards,

Orkun Gedik

jamal_asi
Participant
0 Kudos

Hi Orkun,

These are snapshots taken when the tablespaces are NOT in hot backup mode. They are similar to a database that was powered down at once and then when restarted, it performs an instance recovery. This is the case if all the data files and the online and off line redo logs and the control files were snapshoted at time N.

If the time on the data files snapshot is N and the time on the online and latest offline redo logs and the control file is N+t where t >0, then we recover using the statement:

RECOVER AUTOMATIC DATABASE;

Applying the above two procedures, we were able to bring up the instance. The question here is:

Was there any data loss?

That is why I would like to see more comments regarding these procedure and if anybody uses them in production environments and what are the draw backs or issues that may ensue.

Thanks,

Former Member
0 Kudos

Hi,

Under normal conditions, the database should be in the backup mode during the online backup, as I noted in my previous message. I heard that Oracle supports some of the storage vendors consistent crash recovery solution without hot backup mode, such as EMC or NetApp.

Additionally, RMAN support this feature, as noted in the SAP note, below on item 53;

Note 105047 - Support for Oracle functions in the SAP environment

In this case, I suggest you contact to your storage vendor, to learn support on online backup without hot backup mode.

Best regards,

Orkun Gedik

thiago_cavalheiro
Active Participant
0 Kudos

Hello Jamal,

In my optinion you have it all already.

The most important when defining a backup strategy is to chose the option that fits better your case and perform restore tests with a frequency. I'd say at least once a quarter.

I see you have already done these steps.

Best regards,

Thiago