Re: alarming experience with controlfile autobackup on

The testcase with a file drop
(autobackup is on, the tablespace test is composed of 3 files: test1,test2 and test3)

1. Backup database
2. Alter tablespace test drop datafile 'test2';(first backup of the controlfile is created in fra/autobackup/) for exemple at 12:00 PM
3. Alter tablespace test drop datafile 'test3';(another backup of the controlfile is created in fra/autobackup/) for exemple at 12:05 PM
4. delete obsolete (with redundancy 1 , the first backup of the controlfile is deleted)
5. shutdown
6. startup nomount;
7. restore controlfile from autobackup;
8. restore database to time "to_date(.....12:03PM....)" (to the time before the datafile test3 was deleted)
9. recover database to time "to_date(.....12:03PM....)"
10. alter database open resetlogs;

What is the result ? The database was restored and recovered to the point in time when test3 existed, but this file is not restored since the new controlfile does not see it. So yes, you can say that the database was brought to state that never existed , to the point in time 12:03 PM but without the datafile test3.
This is a normal and expected behaviour, all files were restored and recovered to their point in time,except the files that are unkown to the control file used. One more reason to use the rman catalog. And this behaviour is not specific for autobackup controlfile on feature, every time you are doing a restore/recover with the controlfile that does not see some files, those files will not be restored

The testcase with a file add

1. Backup database
2. Alter tablespace test add datafile 'test2';(first backup of the controlfile is created in fra/autobackup/) for exemple at 12:00 PM
3. Alter tablespace test add datafile 'test3';(another backup of the controlfile is created in fra/autobackup/) for exemple at 12:05 PM
4. delete obsolete (with redundancy 1 , the first backup of the controlfile is deleted)
5. shutdown
6. startup nomount;
7. restore controlfile from autobackup;
8. restore database to time "to_date(.....12:03PM....)" (to the time before the datafile test3 was added)
9. recover database to time "to_date(.....12:03PM....)"
10. alter database open resetlogs;

The datafile test3 was displayed in v$datafile in 'recover' state after the recover command and its checkpoint_time (12:05) was different from all other datafiles (12:03) in v$datafile_header. After <alter database open resetlogs>, it has disapeared from dictionnary views,in spite of the fact that the controlfile saw it, because the database was brought to the point in time before it was created. So, again, this is the expected behaviour

HTH Dragutin

--
http://www.freelists.org/webpage/oracle-l