Subject: Backup and Recovery Scenarios Content Type: TEXT/PLAIN Creation Date: 05-JAN-2000 PURPOSE Describe various Backup and Recovery Scenarios SCOPE & APPLICATION All support analysts RELATED DOCUMENTS Backup and Recovery HandBook, Intro to DataServer Course Material Backup and Recovery - an Overview Backup a) Consistent backups A consistent backup means that all data files and control files are consistent to a point in time. I.e. they have the same SCN. This is the only method of backup when the database is in NO Archive log mode. b) Inconsistent backups An Inconsistent backup is possible only when the database is in Archivelog mode. You must apply redo logs to the data files, in order to restore the database to a consistent state. c) Database Archive mode The database can run in either archivelog mode or noarchivelog mode. When you first create the database, you specify if it is to be in archivelog mode. Then in the init.ora file you set the parameter log_archive_start=true so that archiving will start automatically on startup. If the database has not been created with archivelog mode enabled, you can issue the command whilst the database is mounted, not open. SVRMGR> alter database archivelog;. SVRMGR> log archive start SVRMGR> alter database open SVRMGR> archive log list This command will show you the log mode and if automatic archival is set. d) Backup Methods Essentially, there are two backup methods, hot and cold, also known as online and offline, respectively. A cold backup is one taken when the database is shutdown. A hot backup is on taken when the database is running. Commands for a hot backup: 1. Svrmgr>alter database archivelog Svrmgr> log archive start Svrmgr> alter database open 2. Svrmgr> archive log list --This will show what the oldest online log sequence is. As a precaution, always keep the all archived log files starting from the oldest online log SCN. 3. Svrmgr> Alter tablespace tablespace_name BEGIN BACKUP 4. --Using an OS command, backup the database. 5. Svrmgr> Alter tablespace tablespace_name END BACKUP 6. Svrmgr> archive log list ---do this again to obtain the current SCN. You will want to make sure you have a copy of this redo log file, so to force an archived log, issue 7. Svrmgr> ALTER SYSTEM SWITCH LOGFILE 8. Backup all archived log files determined from steps 2 and 6. Do not backup the online redo logs. These will contain the end-of-backup marker and can cause corruption if use doing recovery. 9. Back up the control file: Svrmgr> Alter database backup controlfile to 'filename' e) Incremental backups These are backups that are taken on blocks that have been modified since the last backup. These are useful as they don't take up as much space and time. There are two kinds of incremental backups Cumulative and Non cumulative. Cumulative incremental backups include all blocks that were changed since the last backup at a lower level. This one reduces the work during restoration as only one backup contains all the changed blocks. Noncumulative only includes blocks that were changed since the previous backup at the same or lower level. Using rman, you issue the command "backup incremental level n" f) Support scenarios When the database crashes, you now have a backup. You restore the backup and then recover the database. Also, don't forget to take a backup of the control file whenever there is a schema change. RECOVERY There are several kinds of recovery you can perform, depending on the type of failure and the kind of backup you have. Essentially, if you are not running in archive log mode, then you can only recover the cold backup of the database and you will lose any new data and changes made since that backup was taken. If, however, the database is in archivelog mode you will be able to restore the database up to the time of failure. There are three basic types of recovery: 1. Online Block Recovery. This is performed automatically by Oracle.(pmon) Occurs when a process dies while changing a buffer. Oracle will reconstruct the buffer using the online redo logs and writes it to disk. 2. Thread Recovery. This is also performed automatically by Oracle. Occurs when an instance crashes while having the database open. Oracle applies all the redo changes in the thread that occurred since the last time the thread was checkpointed. 3. Media Recovery. This is required when a data file is restored from backup. The checkpoint count in the data files here are not equal to the check point count in the control file. This is also required when a file was offlined without checkpoint and when using a backup control file. Now let's explain a little about Redo vs Rollback. Redo information is recorded so that all commands that took place can be repeated during recovery. Rollback information is recorded so that you can undo changes made by the current transacation but were not committed. The Redo Logs are used to Roll Forward the changes made, both committed and non- commited changes. Then from the Rollback segments, the undo information is used to rollback the uncommitted changes. Note: using the _corrupted_rollback_segments parameter in the init.ora will mean that startup bypasses the rollback stage of recovery. This may result in an inconsistent database. Media Failure and Recovery in Noarchivelog Mode In this case, your only option is to restore a backup of your Oracle files. The files you need are all datafiles, and control files. You only need to restore the password file or parameter files if they are lost or are corrupted. Media Failure and Recovery in Archivelog Mode In this case, there are several kinds of recovery you can perform, depending on what has been lost. The three basic kinds of recovery are: 1. Recover database - here you use the recover database command and the database must be closed and mounted. Oracle will recover all datafiles that are online. 2. Recover tablespace - use the recover tablespace command. The database can be open but the tablespace must be offline. 3. Recover datafile - use the recover datafile command. The database can be open but the specified datafile must be offline. Note: You must have all archived logs since the backup you restored from, or else you will not have a complete recovery. a) Point in Time recovery: A typical scenario is that you dropped a table at say noon, and want to recover it. You will have to restore the appropriate datafiles and do a point-in-time recovery to a time just before noon. Note, you will lose any transactions that occurred after noon. After you have recovered until noon, you must open the database with resetlogs. This is necessary to reset the log numbers, which will protect the database from having the redo logs that weren't used be applied. The four incomplete recovery scenarios all work the same: Recover database until time '1999-12-01:12:00:00'; Recover database until cancel; (you type in cancel to stop) Recover database until change n; Recover database until cancel using backup controlfile; Note: When performing an incomplete recovery, the datafiles must be online. Do a select name, status from v$datafile to find out if there are any files which are offline. If you were to perform a recovery on a database which has tablespaces offline, and they had not been taken offline in a normal state, you will lose them when you issue the open resetlogs command. This is because the data file needs recovery from a point before the resetlogs option was used. b) Recovery without control file If you have lost the current control file, or the current control file is inconsistent with files that you need to recover, you need to recover either by using a backup control file command or create a new control file. You can also recreate the control file based on the current one using the 'backup control file to trace' command which will create a script for you to run to create a new one. Recover database using backup control file command must be used when using a control file other that the current. The database must then be opened with resetlogs option. c) Recovery of missing datafile with rollback segment The tricky part here is if you are performing online recovery. Otherwise you can just use the recover datafile command. Now, if you are perfoming an online recovery, you must first ensure that in the init.ora file, you remove the parameter rollback_segments. Otherwise, oracle will want to use those rollback segments when opening the database, but can't find them and wont open. Until you recover the datafiles that contain the rollback segments, you need to create some temporary rollback segments in order for new transactions to work. Even if other rollback segments are ok, they will have to be taken offline. So, all the rollback segments that belong to the datafile need to be recovered. If all the datafiles belonging to the tablespace rollback_data were lost, you can now issue a recover tablespace rollback_data. Next bring the tablespace online and check the status of the rollback segments by doing a select segment_name, status from dba_rollback_segs; You will see the list of rollback segments that are in status Need Recovery. Simply issue alter rollback segment online command to complete. Don't forget to reset the rollback_segments parameter in the init.ora. d) Recovery of missing datafile without rollback segment There are three ways to recover in this scenario, as mentioned above. 1. recover database 2. recover datafile 'c:\orant\database\usr1orcl.ora' 3. recover tablespace user_data e) Recovery with missing online redo logs Missing online redo logs means that somehow you have lost your redo logs before they had a chance to archived. This means that crash recovery cannot be performed, so media recovery is required instead. All datafiles will need to be restored and rolled forwarded until the last available archived log file is applied. This is thus an incomplete recovery, and as such, the recover database command is necessary. (i.e. you cannot do a datafile or tablespace recovery). As always, when an incomplete recovery is performed, you must open the database with resetlogs. Note: the best way to avoid this kind of a loss, is to mirror your online log files. f) Recovery with missing archived redo logs If your archives are missing, the only way to recover the database is to restore from your latest backup. You will have lost any uncommitted transactions which were recorded in the archived redo logs. Again, this is why Oracle strongly suggests mirroring your online redo logs and duplicating copies of the archives. g) Recovery with ResetLog option Reset log option should be the last resort, however, as we have seen from above, it may be required due to incomplete recoveries. (recover using a backup control file, or a point in time recovery). It is imperative that you backup up the database immediately after you have opened the database with reset logs. The reason is that oracle updates the control file and resets log numbers, and you will not be able to recover from the old logs. The next concern will be if the database crashes after you have opened the database with resetlogs, but have not had time to backup the database. How to recover? Shut down the database Backup all the datafiles and the control file Startup mount Alter database open resetlogs This will work, because you have a copy of a control file after the resetlog point. Media failure before a backup after resetlogs. If a media failure should occur before a backup was made after you opened the database using resetlogs, you will most likely lose data. The reason is because restoring a lost datafile from a backup prior to the resetlogs will give an error that the file is from a point in time earlier, and you don't have its backup log anymore. h) Recovery with corrupted/missing rollback segments. If a rollback segment is missing or corrupted, you will not be able to open the database. The first step is to find out what object is causing the rollback to appear corrupted. If we can determine that, we can drop that object. If we can't we will need to use the hidden parameters _allow_corruption. So, how do we find out if it's actually a bad object? 1. Make sure that all tablespaces are online and all datafiles are online. This can be checked through v$datafile, under the status column. For tablespaces associated with the datafiles, look in dba_tablespaces. If this doesn't show us anything, ie, all are online, then 2. Put the following in the init.ora: event = "10015 trace name context forever, level 10" This event will generate a trace file that will reveal information about the transaction Oracle is trying to roll back and most importantly, what object Oracle is trying to apply the undo to. Stop and start the database. 3. Check in the directory that is specified by the user_dump_dest parameter (in the init.ora or show parameter command) for a trace file that was generated at startup time. 4. In the trace file, there should be a message similar to: error recovery tx(#,#) object #. TX(#,#) refers to transaction information. The object # is the same as the object_id in sys.dba_objects. 5. Use the following query to find out what object Oracle is trying to perform recovery on. select owner, object_name, object_type, status from dba_objects where object_id = ; 6. Drop the offending object so the undo can be released. An export or relying on a backup may be necessary to restore the object after the corrupted rollback segment goes away. 7. After dropping the object, put the rollback segment back in the init.ora parameter rollback_segments, remove the event, and shutdown and startup the database. In most cases, the above steps will resolve the problematic rollback segment. If this still does not resolve the problem, it may be likely that the corruption is in the actual rollback segment. If in fact the rollback segment itself is corrupted, we should see if we can restore from a backup. However, that isn't always possible, there may not be a recent backup etc. In this case, we have to force the database open with the unsupported, hidden parameters _offline_rollback_segments, and _corrupted_rollback_segments. Please note, that this is potentially dangerous! When these are used, transaction tables are not read on opening of the database. Because of this, the typical safeguards associated with the rollback segment are disabled. Their status is 'offline' in dba_rollback_segs. Consequently, there is no check for active transactions before dropping the rollback segment. If you drop a rollback segment which contains active transactions then you will have logical corruption. Possibly this corruption will be in the data dictionary. If the rollback segment datafile is physically missing, has been offlined dropped, or the rollback segment header itself is corrupt, there is no way to dump the transaction table to check for active transactions. So the only thing to do is get the database open, export and rebuild. - in init.ora comment out the rollback_segments parameter - add _corrupted_rollback_segments parameter including all rollback segments - Startup - export the database and rebuild If you cannot get the database open, there is no other alternative than restoring from a backup. i) Recovery with System Clock change. You can end up with duplicate timestamps in the datafiles when a system clock changes. A solution here is to recover the database until time 'yyyy-mm-dd:00:00:00', and set the time to be later than the when the problem ocurred. That way it will roll forward through the records that were actually performed later, but have an earlier time stamp due to the system clock change. Performing a complete recovery is optimal, as all transactions will be applied. j) Recovery with missing System tablespace. The only option is to restore from a backup. k) Media Recovery of offline tablespace When a tablespace is offline, you cannot recover datafiles belonging to this tablespace using recover database command. The reason is because a recover database command will only recover online datafiles. Since the tablespace is offline, it thinks the datafiles are offline as well, so even if you recover database and roll forward, the datafiles in this tablespace will not be touched. Instead, you need to perform a recover tablespace command. Alternatively, you could restored the datafiles from a cold backup, mount the database and select from the v$datafile view to see if any of the datafiles are offline. If they are, bring them online, and then you can perform a recover database command. l) Recovery of Read-Only tablespaces If you have a current control file, then recovery of read only tablespaces is no different than recovering read-write files. The issues with read-only tablespaces arise if you have to use a backup control file. If the tablespace is in read-only mode, and hasn't changed to read-write since the last backup, then you will be able to media recovery using a backup control file by taking the tablespace offline. The reason here is that when you are using the backup control file, you must open the database with resetlogs. And we know that Oracle wont let you read files from before a resetlogs was done. However, there is an exception with read-only tablespaces. You will be able to take the datafiles online after you have opened the database. When you have tablespaces that switch modes and you don't have a current control file, you should use a backup control file that recognizes the tablespace in read-write mode. If you don't have a backup control file, you can create a new one using the create controlfile command. Basically, the point here is that you should take a backup of the control file every time you switch a tablespaces mode.