Where does oracle write rollback information (undo information), if the rollback tablespace is taken to backup mode?
any metalink or link references are wellcome.
svrmgrl>alter tablespace RBS begin backup;
Look for SYSTEM rollback segment(which can not be taken offline by the way). Here are some notes from oracle manuals:
Note: No data can be inserted into any tablespace until the
current instance has acquired at least two rollback segments
(including the SYSTEM rollback segment).
When a database is created, a single rollback segment named SYSTEM is
created in the SYSTEM tablespace. You can create any objects in non-SYSTEM
tablespaces, but you cannot write to them until you have created and brought
online at least one additional rollback segment in a non-SYSTEM tablespace (for
An instance always acquires the SYSTEM rollback segment in addition to any other
rollback segments it needs. However, if there are multiple rollback segments, Oracle
tries to use the SYSTEM rollback segment only for special system transactions and
distributes user transactions among other rollback segments. If there are too many
transactions for the non-SYSTEM rollback segments, Oracle uses the SYSTEM
While RBS tablespace is in backup mode, SYSTEM rollback segment does not play any other role that it plays normaly. It does not have to, because nothing is different with normal rollback segments while their tablespace is in backup mode.
The original question could pe rephrased like: "Where does Oracle write my inserts, updates and deletes on a specific table while its tablespace is in backup mode?" The answer is obvious: In the coresponding datafile(s)! (Almost) nothing is changed while tablespace is in backup mode. Data is still writen to its datafiles like any other time. It is the same with tables, rollbac segments, indexes,... - any kind of segments. The changed blocks are still written to the datafiles.
Jurij Modic ASCII a stupid question, get a stupid ANSI
24 hours in a day .... 24 beer in a case .... coincidence?
see an article "IS RMAN REALLY WORTH THE TROUBLE?
Tim Gorman, SageLogix, Inc." from EVDBT site.
If can not find it, I can send it to you.
In short the idea is that begin backup freeze datafile headers for SCN updates. Then backup is done of a datafile while it is modified by oracle (db blocks written into it)
When need to restored that datafile, changes from redo logs are applied to datafile(restored one) starting from SCN frozen in datafile's header.
To freeze datafile header, a global checkpoint is performed to flush buffers from memory.