Problem when backing up Oracle 10g

Question

I've just started a job and have identified an issue in which the database isn't currently being backed up properly so to speak. We are doing one back up every 6 hours that uses the Oracle native backup utility, but we were also sold a process by a company in which they stated that they could in essence perform "warm" backups of our database by simply taking file system copies of our database files and when we needed to restore we'd simply shut down Oracle and then copy over the files that had been copied, restart Oracle and the world would be whole again. The challenge is the fact that we have not gotten this to work just yet. I need to spend some more time reviewing the message that Oracle is giving, but my primary question is, "Is it possible" to take copies of Oracle files while Oracle is still running and to use those files at a later date to restore the database? I know that it works if the database is shut down, and then copies are made, but this is the first that I've heard that a copy (file system) can be made while the database is running. Any guidance would be greatly appreciated. Here is the error that we are getting.

ORA-00314: log 3 of thread 1, expected sequence# 1939 doesn't match 1944
ORA-00312: online log 3 thread 1: 'E:\ORACLE\ORADATA\ITMS\REDO03.LOG'

Answer 1

Yes, it is possible, but you have to put all the tablespaces into backup mode first and take them out afterwards (e.g. ALTER TABLESPACE x BEGIN BACKUP and ALTER TABLESPACE x END BACKUP; you'll need to check the syntax and make sure it's appropriate for your situation!). Oversimplifying hugely, this tells Oracle not to write to any of the data files, so they're all kept in a consistent state.

The two main problems you get otherwise are that individual files are updated while you're copying them so a single file can be corrupted; and more visibly that different files have different internal timestamps and sequences so Oracle won't allow them to be used.

If you're using a process you've bought in then it should already be taking care of all that though. It sounds the backup is OK and it's the restore that you haven't got working.

I haven't been involved in a restore from a hot backup for some time so someone else will need to give the detail on the actual error. My read of it is that you've tried to open with the restored data files but the later live redo logs. When restoring I think you either have to RECOVER the database using the redo logs generated since the backup was taken; or if you're trying to restore to that point in time then you can open the data with the RESETLOGS directive and lose all the changes from all the redo logs that came later. But really take more informed advice than this...

Answer 2

As far as I know, there are two ways that you can "copy" datafiles from a running Oracle instance.

• The datafiles are copied for a tablespace when the tablespace is in "BEGIN BACKUP" mode.
• You are using a high-end journalling filesystem such as Veritas that can snapshot and track block changes on the filesystem while the copy is taking place.

Answer 3

It is possible. You must must be in ARCHIVELOG mode.

An example script would be for manual:

Alter tablespace USERS begin backup;
host cp -p /u02/oradata/PROD/users01.dbf /u03/backup/PROD/
host cp -p /u02/oradata/PROD/users02.dbf /u03/backup/PROD/
Alter tablespace USERS end backup;

However, I would recommend just using RMAN. RMAN is QUITE ROBUST, included free, and will do the hot backup, as well as cold. It will clone to another instance, clone as a point in time, recover to a certain point in time, etc. Any manual backup procedure should be migrated to using the RMAN.

If you wanted to backup the entire database while it is open (I prefer as Oracle with DBA so you can avoid passwords in scripts, but ymmv):

$ ORAENV_ASK=NO
$ ORACLE_SID=PROD
$ . oraenv
$ rman target=/

Recovery Manager: Release 10.2.0.4.0 - Production on Thu Oct 28 14:23:29 2010

Copyright (c) 1982, 2007, Oracle.  All rights reserved.

connected to target database: PROD (DBID=x)

RMAN> backup as compressed backupset database plus archivelog;

...

Backup Complete.

Answer 4

I've effectively done this with a non-mission-critical database running on Amazon EC2. My backup strategy is to periodically take a snapshot of the EBS volume. To restore a backup, I create a new EBS volume from the snapshot, start up the instance using it, then run RECOVER DATABASE.

This loses any transactions that were in-flight at the time when the snapshot was taken, of course.