Transcript Using Recovery Manager.. Oracle Presentation

®
Using
Recovery Manager
Connie Dialeris, Server Technologies
[email protected]
Last Updated: 8th July 1997
Agenda and Aims
•
•
Agenda
– Overview and Features
– Recovery Manager Components
and Architecture
– Configuring Recovery Manager
– Performing Backups
– Performing Restoration
– Recovery
– Catalog Maintenance
– Running Recovery Manager
Commands
– Reports on the Recovery Catalog
– Miscellaneous
Aims
– To understand what Recovery
Manager is
– How to configure Recovery
Manager
– How to perform backups, restores
and recoveries
•
Note
– This presentation will not show all
Recovery Manager features and
syntax
– For detailed information on
Recovery Manager, please see the
Oracle8 Backup and Recovery
Guide
Recovery Manager - Connie Dialeris
2
Overview and Features
Recovery Manager - Connie Dialeris
3
What is Recovery Manager
• Recovery Manager is a tool that:
– manages the process of creating backups, and also manages the process of
restoring and recovering from them
– directs the Oracle8 server to back itself up, and restore and recover itself
• The features of Recovery Manager are also available through the
Oracle Enterprise Manager - Backup Manager GUI
• Recovery Manager is a Command Line Interface (similar to
SQL*DBA)
– Provides a powerful, Operating System independent scripting language
– Interactive or Batch Mode
Recovery Manager - Connie Dialeris
4
Features
• Incremental Backups (up to 4 levels. Level 0 and levels 1->4)
• Corrupt block detection
– during backup
• v$backup_corruption, v$copy_corruption
• also reported in the database’s alert log and trace files
– restore
• Easily Manages distributing database backups, restores and recoveries
across clustered nodes in an Oracle Parallel Server
• Performance
– Automatic Parallelization of backup, restore and recovery
– Multiplexing prevents flooding any one file with reads and writes while
still keeping a tape drive streaming
– Backups can be restricted to limit reads per file, per second to avoid
interfering with OLTP work
– No generation of extra redo during open database backups
Recovery Manager - Connie Dialeris
5
Features
• Easy backup of Archived redo logs
• Able to limit:
– Number of open files
• Overcome O/S limits on concurrent files opened
• Default 32
– Size of backup piece
• This is important, as a backup piece size may easily be over the maximum file
size allowed by the Operating System, or Media Manager
Recovery Manager - Connie Dialeris
6
Components and
Architecture
Recovery Manager - Connie Dialeris
7
Overview of
Oracle Database Components
•
•
Instance
(memory structures and processes)
Control File
(one or more multiplexed)
server
server
server
server
Redo
Log
116
Current
Online Log
SMON
PMON
DBWR
LGWR
ARCH
Instance
prod
SGA
Redo
Log
117
Redo
Log
118
Online
Log
•
•
•
One or more Data Files
Online Redo Log
Archived Redo Log
(only if in ARCHIVELOG mode)
CF
CF
CF
File
1
File
2
File
3
File
4
File
5
Database
Recovery Manager - Connie Dialeris
Redo
Log
110
Redo
Log
114
Redo
Log
111
Redo
Log
115
Redo
Log
112
Redo
Log
116
Redo
Log
113
Archived
Redo Log
8
Recovery Manager Components
• Components of Recovery Manager
–
–
–
–
Recovery Manager executable (rman)
Target Database
Recovery Catalog (or database’s control file)
Channel
• Recovery Manager
– Translates Recovery Manager commands into a sequence of steps which
operate on physical files
– The Backup, Restore and Recovery Steps are sent to the target database
for execution
– If a Recovery Catalog is used, Recovery Manager generates steps which
update the catalog
– Recovery Manager co-ordinates and monitors the execution of these steps
Recovery Manager - Connie Dialeris
9
Recovery Manager Components
• Target Database
– The database on which the specified backup, restore and recover actions
will be executed
• Recovery Catalog
– Is a repository of information queried and updated by Recovery Manager
– Is an Application, stored in an Oracle database
– Contains information about:
•
•
•
•
•
Physical schema of the target database
Datafile and Archivelog backup sets and pieces
Datafile Copies
Archived Redo Logs
Stored scripts
– Recovery Manager uses information in the Recovery Catalog to decide
how to perform certain operations (if a Recovery Catalog is not used,
Recovery Manager can query the target database’s control file to decide
what actions to perform).
Recovery Manager - Connie Dialeris
10
Recovery Manager Components
• Channel
– Allocating a channel tells Recovery Manager to initiate an Oracle server
process for backup/restore/recovery on the target database
– The number of channels allocated will be the maximum degree of
parallelization
– The type of channel allocated will determine the type of media used
• type disk
• type ‘SBT_TAPE’
– Channels can be used to impose limits:
• read rate (number of buffers read per second, per file)
• backup piece file size
• maximum concurrently open files
Recovery Manager - Connie Dialeris
11
Recovery Manager Architecture
Recovery Manager
server
Recovery Catalog
channel
SBT_TAPE
server
(default)
prod
SGA
Recovery
Catalog
CF
SCF
File
1
server
(polling)
File
2
File
3
server
MML
File
4
Target
Media Manager
Recovery Manager - Connie Dialeris
12
Configuring
Recovery Manager
Recovery Manager - Connie Dialeris
13
Configuring Recovery Manager
• You must have Oracle8 installed. Recovery Manager is distributed
with Oracle8.
• Decisions required to configure Recovery Manager
– Decide whether you will use Oracle password files
You need to use password files if you:
• Intend to connect to the target database remotely
(for example, if you wish to use the Oracle Enterprise Manager Backup GUI)
• Would like to specify a TNS alias on the connect string
(for example, scott/tiger@tns_alias)
– Decide how to backup supporting O/S files, such as init.ora files and
password files
– Decide whether to use a Recovery Catalog (discussed in more detail)
Recovery Manager - Connie Dialeris
14
Deciding Whether to Use a
Recovery Catalog
• It is possible for Recovery Manager to use just the target databases’
control file.
•
Using RMAN without a catalog
– There is no extra overhead of a Recovery Catalog
– Does not allow easy Point-In-Time Recovery
– If you lose your control file and need to restore and recover the database,
restore and recovery are impossible without the assistance of Oracle
World Wide Customer Support. WWCS can only help if the user
maintains excellent records (I.e. the rman message log files) of which files
were backed up, the date the backup occurred and the names of the output
files
– Optimally, the user should keep current backups of his controlfile in
multiple locations that do not need to be restored using Recovery Manager
– Stored scripts are not available
• Oracle recommends the use of a Recovery Catalog for databases with
more than 20 files.
Recovery Manager - Connie Dialeris
15
Creating a Recovery Catalog
• Choose whether to create a separate database for the Recovery
Catalog, or install it in an existing database.
– If you decide to create a separate database, then create it using the same
procedures you would to create any Oracle database.
– Cautions:
• Do not install the recovery catalog in the same database as the target database
• Do not put this database on the same physical disks as the target database
• Decide whether to create a separate tablespace for the Recovery
Catalog.
Start Server Manager, and:
SVRMGR> connect internal
SVRMGR> create tablespace rcvcat datafile
2>
‘/oracle/rcvcat/prod/rcvcat_1.dbf’ size 20m;
Recovery Manager - Connie Dialeris
16
Creating a Recovery Catalog
• Create the schema owner (this example assumes there is a temp
tablespace), and grant the required privileges.
Start Server Manager, and:
SVRMGR> connect internal
SVRMGR> create user rman identified by rman
2>
temporary tablespace temp
3>
default tablespace rcvcat
4>
quota unlimited on rcvcat;
SVRMGR> grant recovery_catalog_owner to rman;
• Create the Recovery Catalog Application.
Start Server Manager, and:
SVRMGR> connect rman/rman
SVRMGR> @?/rdbms/admin/catrman
Before continuing, check to ensure no errors were reported when
running the script.
Recovery Manager - Connie Dialeris
17
Creating a Recovery Catalog
• If you use a Recovery Catalog, you must decide:
– How, and how frequently, you will backup the Recovery Catalog
– How to automate Catalog resyncs, and how often
Use a utility such as cron (on Unix) to schedule regular resyncs
• Irrespective of whether a Recovery Catalog is used, Recovery Manager
sometimes needs to make a temporary backup of the controlfile to
read.
The location of this file is configurable:
RMAN> set snapshot controlfile name to ‘/dev/vgd01/r300_1’;
The default location of the snapshot controlfile on Unix systems is:
$ORACLE_HOME/dbs/snapcf_<dbname>.f, where <dbname> is the database
name.
Recovery Manager - Connie Dialeris
18
Linking with a Media Manager
• To be able to write to sequential devices such as Tape, you must link
the Oracle server software with software which is written by a Media
Manager.
• This software (usually known as the Media Management Library, or
Media Management Layer) is called by Oracle when it needs to write
to, or read from devices supported by the Media Manager.
• An example of linking Oracle on Solaris, with an MML (the MML is
libobk.so, and is located in /usr/lib):
> cd $ORACLE_HOME/rdbms/lib
> make -f ins_rdbms.mk ioracle LLIBMM=/usr/lib/libobk.so
You should shutdown all databases using that executable set before
linking.
Recovery Manager - Connie Dialeris
19
Starting Recovery Manager
• There are a number of ways to start Recovery Manager, depending on
how you have configured it. Below are a few examples:
• Starting Recovery Manager locally (without a password file)
– Without using a Recovery Catalog
> setenv ORACLE_SID prod
> rman nocatalog
RMAN> connect target
– Without a Recovery Catalog, if SCOTT has SYSDBA role
> setenv ORACLE_SID prod
> rman target scott/tiger nocatalog
– Starting rman locally, with a Catalog
> setenv ORACLE_SID prod
> rman rcvcat rman/rman@rcat
RMAN> connect target
Where:
•
The target database instance is called ‘prod’, and we are connecting internal using a password file
•
The schema containing the Recovery Catalog is rman, with the same password, and it’s in a database called rcat
Recovery Manager - Connie Dialeris
20
Starting Recovery Manager
• Starting rman using TNS alias (e.g. remotely):
– Starting rman, connecting to a SCOTT with SYSDBA priv, without a
catalog
> rman target scott/tiger@prod nocatalog
– Starting rman when you use password files, without a catalog:
> rman target internal/kernel@prod nocatalog
– Starting rman if you use password files and a recovery catalog
> rman target internal/kernel@prod rcvcat rman/rman@rcat
Where:
•
The target database instance is called ‘prod’, and we are connecting internal using a password file
•
The schema containing the Recovery Catalog is rman, with the same password, and it’s in a database called rcat
Recovery Manager - Connie Dialeris
21
Registering a Database in a
Recovery Catalog
• If using a Recovery Catalog, the first time you use Recovery Manager
to perform any backups, you must register the Database with the
Recovery Catalog. The target database you connect to will be the
database registered in the catalog.
RMAN> register database;
• The register database command identifies the database to the Recovery
Catalog.
– All information about the target database’s structure is propagated to the
Recovery Catalog from the target database’s control file.
– The database must be mounted or open to perform this action.
Recovery Manager - Connie Dialeris
22
Performing Backups
Recovery Manager - Connie Dialeris
23
Backup Overview
• The database can be Open or Closed during backup
– If you are performing a closed backup, the database must be mounted, but
not open
SVRMGR> shutdown immediate
SVRMGR> startup mount
– If you are performing a backup while the database is open, the tablespaces
should not be put in ‘hot backup’ mode (I.e. you should not issue an alter
tablespace begin backup command).
– Recovery Manager allows you to backup the whole database, or any
logical part thereof (irrespective of whether the database is open or closed)
•
•
•
•
database
tablespace
datafile
control file
• Archived logs may also be backed up
Recovery Manager - Connie Dialeris
24
New & Changed Terminology
• Whole backup
A backup of all datafiles and the control file.
• Full backup
This replaces the term ‘full database backup’.
A backup of one or more files, which is not an incremental.
• Incremental backup
A backup of datafiles which only includes the blocks which have changed
since the last incremental. The base level 0 incremental will backup all blocks
containing data.
• Operating System Backup
An Oracle7 style file backup, made by Operating System Utilities.
• Open Backup
A backup of any portion of the database while it is open (I.e. online).
• Closed Backup
A backup of any portion of the database while it is closed (I.e. offline).
Recovery Manager - Connie Dialeris
25
Types of Backups
• There are two types of Recovery Manager backups:
– File Copies
– Backup Sets
• Recovery Manager automatically resynchronizes database structure
information in the Recovery Catalog before executing certain
commands (e.g. backup)
• Recovery Manager propagates changes in the databases’ structure to
the recovery catalog, before it executes certain commands (datafiles,
resized files, or dropped tablespaces)
Recovery Manager - Connie Dialeris
26
File Copy
• A File Copy is created by the copy command and
–
–
–
–
–
Is an image copy of the file
Can only be written to disk
Contains only a single input file (datafile, archivelog or controlfile)
Is most like an Operating System backup (contains all blocks)
Can be a part of an incremental strategy; but a file copy always includes
all blocks, and hence can only be a level 0
run {
allocate channel d1 type disk;
copy
level 0
datafile 1 to ‘/oracle/prod/backup/file1.dbf’;
}
– Is usable immediately - does not need to be restored
Datafile Copies can be used in-situ by issuing a switch command, or can
be restored to the original location, if desired
Recovery Manager - Connie Dialeris
27
Parallelization of File Copies
• Parallelization of File Copies is achieved by copying multiple files
within one copy command, as well as allocating multiple channels.
run {
allocate channel d1 type disk;
allocate channel d2 type disk;
allocate channel d3 type disk;
allocate channel d4 type disk;
copy
#first
<- RMAN comment
datafile 1 to ‘/oracle/backup/prod1/d1/prd1.dbf’,
datafile 2 to ‘/oracle/backup/prod1/d2/prd2.dbf’,
datafile 3 to ‘/oracle/backup/prod1/d3/prd2.dbf’;
copy
#second
datafile 4 to ‘/oracle/backup/prod1/d4/prd3.dbf’;
sql ‘alter system archive log current’; }
• Note that recovery manager executes commands serially. I.e. the
previous command must finish before the next command is started.
In this example, the first copy command will perform 3 copies in
parallel. The second copy command is not executed until the
command before it completes.
Recovery Manager - Connie Dialeris
28
Backup Set
• A Backup Set is created by the backup command and
– Usually contains multiple source files (multiplexed in the output)
– It is a logical entity, comprising one or more physical output files called
backup pieces
– Can Be Written to Disk or ‘SBT_TAPE’
– A restore operation is required to extract files from a backup set
– There are two types of Backup Sets
• Datafile backup sets (can include control files)
• Archivelog backup sets (may only contain archivelogs)
• Datafile backup sets
– Can be Incremental or Full (are full, by default)
– Do not include empty blocks
Recovery Manager - Connie Dialeris
29
Backup Set - example
• Whole database backup, of a database with 10 files (including the
control file), writing out to two tape drives
• The number of files per backup set will be restricted to 5, which means
that two sets will be created, each with 5 files multiplexed.
• If two channels are allocated, these sets will be created in parallel
run {
allocate channel t1 type ‘SBT_TAPE’;
allocate channel t2 type ‘SBT_TAPE’;
backup
filesperset 5
format ‘df_%t_%s_%p’
(database);
}
• The format string determines the output file name (passed to sbtopen)
• Recovery Manager will automatically perform the partitioning of files
to channels, multiplex the files, and skip any unused blocks
Recovery Manager - Connie Dialeris
30
Parallelization of Backup Sets
• Parallelization of Backup Sets is achieved by allocating multiple
channels, and specifying filesperset.
– Recovery Manager establishes one database connection (server process) to
the target database per channel
– In this example, if the keyword filesperset is not included only one
channel would be used, and that channel would create one piece which
contains all five files. The other two channels would be idle.
– Recovery Manager uses all 3 tape drives, each one to write a backup set,
each set with a maximum of 1 file.
run {
allocate channel t1 type ‘SBT_TAPE’;
allocate channel t2 type ‘SBT_TAPE’;
allocate channel t3 type ‘SBT_TAPE’;
backup
format ‘df_%t_%s_%p’
filesperset 1
(datafile 10,11,12);
sql ‘alter system archive log current’;
}
Recovery Manager - Connie Dialeris
31
Backup Pieces
• A Backup Set is a logical entity, which is composed of one or more
physical output files, called backup pieces
• Typically, a backup set only has one output piece
• When using ‘SBT_TAPE’ the only time a backup set is made of more
than one piece, is when the piece size is limited by the user
e.g. Tablespace backup to one tape drive, max. file size is 4Gb
run {
allocate channel t1 type ‘SBT_TAPE’;
set limit channel t1 kbytes 4194304;
backup
format ‘df_%t_%s_%p’ filesperset 5
(tablespace user_data);
}
If the output file is < 4Gb, only one backup piece will be written. If the output
size is > 4Gb, more than one backup piece will be written. Each backup piece
will have blocks from five files interspersed together
Recovery Manager - Connie Dialeris
32
Incremental Datafile Backup
• Is a backup of a datafile which only copies out Oracle blocks which
have been modified since a previous incremental
– An incremental backup at level N (where N > 0) backs up all blocks
modified since the previous incremental at a level <= N
•
•
•
•
Are initially based on a Level 0 backup set, or a Level 0 File Copy
Are faster than performing full (or level 0) backups
Write out fewer blocks than when performing full (or level 0) backups
Are, by default non-cumulative
changed
blocks
all nonempty
blocks
Level
Day
0
Sun
2
Mon
2
Tue
1
Wed
2
Thu
Recovery Manager - Connie Dialeris
2
Fri
2
Sat
0
Sun
33
Cumulative Incremental Backups
• Contain all blocks changed since the previous backup at a level less
than N
• Are provided for recovery speed
• Cumulative Incrementals backup blocks previously backed up at the
same level
– May take longer and will write out more blocks than non-cumulative
– However fewer backups at each level must be applied when recovering
changed
blocks
all nonempty
blocks
Level
Day
0
Sun
2
Mon
2C
Tue
1
Wed
2
Thu
Recovery Manager - Connie Dialeris
2C
Fri
2C
Sat
0
Sun
34
Archivelog Backup Sets
• Only include archivelogs
• Are always full backups
• As Recovery Manager has access to control file information, it knows
which logs have been archived
This prevents a common problem experienced by users, of not
knowing whether an archive log has been completely copied out to the
archive log destination before attempting to back it up
Example
run {
allocate channel t1 type ‘SBT_TAPE’;
backup
format ‘al_%t_%s_%p’
filesperset 10
(archivelog all
delete input);
}
Recovery Manager - Connie Dialeris
35
Backup Sets, File Copies
and O/S backups
• Backup Sets and File Copies can be written to disk, then later backedup to ‘SBT_TAPE’ (one or many times)
This allows the most recent backups to be made to disk, then migrated
to sequential media before reusing the disk space for a more current
backup
• Oracle8 continues to support making O/S Backups
– Valid O/S backups on disk can be cataloged in the Recovery Catalog for
immediate use
e.g. alter tablespace begin backup, break the mirror, then catalog the
broken mirror half, alter tablespace end backup.
Note in the case of breaking mirrors to make O/S backups, you must immediately notify recovery manager that
the backup is no longer available when you resilver the mirror. This can be done using a recovery manager
change...unavailable command.
– Once the O/S backups have been cataloged, they can then be backed up by
Recovery Manager to disk or to the MML
Recovery Manager - Connie Dialeris
36
Restore
Recovery Manager - Connie Dialeris
37
Restore Concepts
• To reconstruct, or bring back an original file from backup
• Recovery Manager allows you to restore any logical unit of a database:
–
–
–
–
database
tablespace
datafile
control file
restore (database);
restore (tablespace <name>,..,<name>);
restore (datafile <name>,..,<name>);
restore controlfile to ‘<location>’;
• Archived logs may also be manually restored, however typically
Recovery Manager automatically restores them if required during
Recovery.
restore (archivelog all|like <filename>|<archivelog range>|
etc…);
Recovery Manager - Connie Dialeris
38
Restore Concepts
• What state the database should be in for various recoveries:
– If you need to restore the control file, the instance must be started, but not
mounted:
SVRMGR> startup nomount
– If you need to restore the whole database, or the system tablespace, the
database can be mounted, but not open:
SVRMGR> startup mount
– If you need to restore a tablespace or datafile, the database can be open
but the tablespace or datafile must be offline:
SVRMGR> alter tablespace user_data offline temporary;
or
SVRMGR> alter database datafile
2> ‘/oracle/prod/file/user_data_1.dbf’ offline;
Recovery Manager - Connie Dialeris
39
Restore
• Recovery Manager:
– Queries the Recovery Catalog to determine the configuration of the
database, and to find out what backups are available for use.
– Decides which backups are the optimal ones to use to restore the specified
files
• When compiling lists of files to restore, Recovery Manager takes into
consideration:
– Time Specifications (e.g. for a Point-In-Time Restore)
– Any tags specified by users
– The channel type (disk/’SBT_TAPE’) allocated. To ensure Recovery
Manager will find the most recent backups, allocate the appropriate
channel type.
– Note that Datafile Copies do not need to be restored, they can be used insitu by executing a ‘switch’ command, which directs the database to use
this file as the actual file in the database.
Recovery Manager - Connie Dialeris
40
Restore Example
• Restore the whole database, including control file (to each file’s
original location).
As the whole database and control file are being restored, the database
should be started, but not mounted.
Start Server Manager, then:
SVRMGR> shutdown immediate
SVRMGR> startup nomount
Recovery Manager - Connie Dialeris
41
Restore Example
Then start Recovery Manager:
rman target internal/kernel@prod rcvcat rman/rman@rcat
RMAN> run {
2>
allocate channel t1 type ‘SBT_TAPE’;
3>
allocate channel t2 type ‘SBT_TAPE’;
4>
allocate channel d1 type disk;
5>
restore controlfile to ‘/oracle/prod/files/ctrl_prod_1.ctl’;
6>
replicate controlfile from
7>
‘/oracle/prod/files/ctrl_prod_1.ctl’;
8>
sql ‘alter database mount’;
9>
restore (database);
10> }
The database has been restored (but not recovered).
If the backup is consistent, the database could be opened without the
application of redo (I.e. the database can be opened without recovery,
however, a fake recovery is needed to open the database resetlogs)
Recovery Manager - Connie Dialeris
42
Parallelization of Restore
• Parallelization of Restore is controlled by the number of channels
allocated and the number of backup sets which are required for restore.
– This applies equally for the restore of level 0 or full backups, as well as
the application of incremental backups of levels > 0
– In the previous example:
– If there were 12 backup sets which needed to be read to restore the whole
database, it would have been faster to allocate 12 channels, thereby
increasing the degree of parallelization.
– However if there had been only one backup set which needed to be read to
restore the whole database, the second channel would not have been used
at all.
Recovery Manager - Connie Dialeris
43
Preparing for Point-In-Time
Recovery
• If a user error has necessitated the database be restored then recovered
to a point-in-time immediately before the error, you will need to
specify an until clause.
The until clause can be specified for each restore and recovery command in the script, but it can also
be defined first - in this way it will apply to all commands executed in that session.
• With an until clause, Recovery Manager will restore a backup taken
before that time. If this clause is omitted, the most recent backups of
the database will be restored. This means a subsequent ‘recover.. until’
command may fail, if the copy of the database restored is already
ahead of the desired recovery point-in-time.
• A ‘time’ to restore until may specified by:
– a time
– a log sequence number
or
– an SCN (System Commit Number)
Recovery Manager - Connie Dialeris
44
Preparing for Point-In-Time
Recovery: Example
• Restore the whole database, including control file (to the original
locations) until time ‘1997/06/15 14:20:00’.
> setenv NLS_DATE_FORMAT ‘YYYY/MM/DD HH24:MI:SS’
> setenv NLS_LANG american
SVRMGR> shutdown abort
SVRMGR> startup nomount
SVRMGR> exit
rman target internal/kernel@prod rcvcat rman/rman@rcat
RMAN> run {
2>
set until time ‘1997/06/15 14:20:00’;
3>
allocate channel t1 type ‘SBT_TAPE’;
4>
allocate channel t2 type ‘SBT_TAPE’;
5>
allocate channel d1 type disk;
6>
restore controlfile to ‘/oracle/prod/files/ctrl_prod_1.ctl’;
7>
replicate controlfile from
8>
‘/oracle/prod/files/ctrl_prod_1.ctl’;
9>
sql ‘alter database mount’;
10>
restore database;
11>
#recover database; #recover should be in same job, if possible
12> }
Recovery Manager - Connie Dialeris
45
Recovery
Recovery Manager - Connie Dialeris
46
Recovery Concepts
• Recovery is the process whereby
– A restored file is made current
or
– A restored file is made current to a specific point in time
• If the database is in ARCHIVELOG mode, it is possible to either
completely recover the database, or recover the database to some
arbitrary point-in-time (incomplete recovery)
• For NOARCHIVELOG mode databases, the only option is to restore
from a consistent whole database backup
• If a database is closed during recovery, all files in that database must
be consistent with each other before the database can be opened.
• If a database is open during a tablespace or datafile recovery, the
tablespace or datafile must be consistent with the remaining datafiles
before it can be brought online.
Recovery Manager - Connie Dialeris
47
Recovery Concepts
• There are two types of recovery:
– Complete recovery
A recovery where all changes made to the database since the restored
backup have been applied (including changes recorded in the online redo
logs).
– Incomplete recovery
A recovery where the recovery is stopped before all changes made to the
database since this backup have been applied. Incomplete recoveries are
performed:
• To undo a user error (recover to the point-in-time before the error occurred)
• When all redo information to complete the recovery is not available (this can
happen for archivelogs as well as online logs)
• When the online logs are lost (and are not mirrored)
A ‘time’ to recover to may be identified by a time, a log sequence number
or an SCN (System Commit Number)
All files must be recovered to the same point-in-time.
Recovery Manager - Connie Dialeris
48
Phases of Recovery
• Recovery Manager performs recovery in 4 phases:
– If a controlfile was restored, repair it
– Apply any incremental backups which will make the files newer
(Incremental backups are applied to a level 0 or full)
– Apply redo which is currently on disk
– If all redo required for recovery is not on disk, restore archivelogs, then
apply them and any redo in the online redo logs needed for recovery
Recovery Manager - Connie Dialeris
49
Tablespace Recovery: Example
• This example shows the restoration and recovery of all files in the
user_data tablespace.
Assume the database is open.
rman target internal/kernel@prod rcvcat rman/rman@rcat
RMAN> run {
2>
allocate channel t1 type ‘SBT_TAPE’;
3>
sql ‘alter tablespace user_data offline immediate’;
4>
restore
5>
tablespace user_data;
6>
recover
7>
tablespace user_data;
8>
sql ‘alter tablespace user_data online’;
9> }
• The tablespace is now available for use.
Recovery Manager - Connie Dialeris
50
Parallelization of Recovery
• The parallelization of applying incremental backups is controlled by
the number of channels allocated and the number of incremental
backup sets which are needed for recovery (this is exactly the same as
parallelizing restore)
• Redo application
– Can be parallelised by setting the init.ora parameter
RECOVERY_PARALLELISM
– Archived Redo Logs can be manually restored to disk in preparation for
Recovery, or Recovery Manager will restore them automatically as they
are required
The restore of archived logs can be parallelized by allocating multiple
channels
Recovery Manager - Connie Dialeris
51
Example of Recovery With
Incrementals
• If a file had been backed up with the schedule shown below, and it
required recovery on Sunday, the following actions would be initiated
by Recovery Manager:
• Restore the previous Sunday’s level 0 backup
– Apply the level 1 backup from Wednesday
– Apply the level 2’s from Thursday, Friday and Saturday
– Apply any archived redo generated after the Saturday backup, and then
apply the online redo logs
Time of Failure
Restored
file
Increm.
applied
Not
required
Redo
Level
Day
0
Sun
2
Mon
2
Tue
1
Wed
2
Thu
Recovery Manager - Connie Dialeris
2
Fri
2
Sat
0
Sun
52
Point-In-Time Recovery
• To complete the PIT Restore example:
> setenv NLS_DATE_FORMAT ‘YYYY/MM/DD HH24:MI:SS’
> setenv NLS_LANG american
SVRMGR> shutdown abort
SVRMGR> startup nomount
SVRMGR> exit
rman target internal/kernel@prod rcvcat rman/rman@rcat
RMAN> run {
2>
set until time ‘1997/06/15 14:20:00’;
3>
allocate channel t1 type ‘SBT_TAPE’;
4>
allocate channel t2 type ‘SBT_TAPE’;
5>
allocate channel d1 type disk;
6>
restore controlfile to ‘/oracle/prod/files/ctrl_prod_1.ctl’;
7>
replicate controlfile from
8>
‘/oracle/prod/files/ctrl_prod_1.ctl’;
9>
sql ‘alter database mount’;
10> restore database;
11> recover database;
12> }
Recovery Manager - Connie Dialeris
53
Catalog Maintenance
Recovery Manager - Connie Dialeris
54
Catalog Maintenance Commands
• There are a number of Recovery Manager commands provided which
will allow you to maintain (modify) information in the Recovery
Catalog.
• Resyncing the Recovery Catalog
RMAN> resync catalog;
Causes Recovery Manager to compare the recovery catalog to the
target database’ controlfile. The Catalog is then updated with new or
changed information.
Should be performed:
– Each time the physical database structure changes e.g.
• Addition (or resize) of datafile
• Tablespace drop
• Rollback segment create or drop
– At least once a day (depending on how many archived logs are produced,
and whether automatic file resize is turned on)
Recovery Manager - Connie Dialeris
55
Catalog Maintenance Commands
• The change command is used to update the availability information in
the Recovery Catalog e.g.
– To mark a backup piece, file copy or archivelog as unavailable or
available for restore or recovery
RMAN> change archivelog
2> ‘/disk1/archivelog/al_prd1_123.rdo’ unavailable;
– To delete a backup piece, file copy or archivelog which is no longer
required from the controlfile, recovery catalog and from the media it
resides on
– RMAN> allocate channel t1 for delete type ‘SBT_TAPE’;
RMAN> change backuppiece <key> delete;
– To remove references to a backup piece, file copy or archivelog from the
controlfile and recovery catalog (e.g. if it was deleted by an O/S utility)
– RMAN> change archivelog
2> ‘/disk1/archivelog/al_prd1_123.rdo’ uncatalog;
Recovery Manager - Connie Dialeris
56
Catalog Maintenance Commands
• If the DBA has deleted a large number of files on disk (e.g. archived
logs), a more efficient method of reconciling the Recovery Catalog
(and controlfile) with what is on disk, is to use the validate keyword
RMAN> change archivelog all validate;
This verifies that the files the Recovery Catalog believes are on disk,
are there. If they are not there, these files are identified as deleted in
the catalog.
Recovery Manager - Connie Dialeris
57
Catalog Maintenance Commands
• The catalog command:
– If there are archivelogs, datafiles copies etc. on disk which were created
before Recovery Manager was used, it is possible to identify the existence
of these to the Recovery Catalog:
RMAN> catalog archivelog ‘/disk1/archivelog/al_prd1_1234.rdo’;
– This command is also used to identify O/S backups made to disk in the
Recovery Catalog. The creation of O/S backups by breaking mirrors (with
alter tablespace begin/end backup)
Once the mirror has been resilvered, the backup cataloged must be
uncataloged using a change...uncatalog command.
Recovery Manager - Connie Dialeris
58
Running Recovery Manager
Commands
Recovery Manager - Connie Dialeris
59
Running Recovery Manager
commands
• Can be run by:
– Typing them into the command line interface (as we have seen in all
examples)
RMAN> run {
2> allocate channel c1 type disk;
3> copy
4>
datafile 10 to ‘/oracle/prod/backup/prod_10.dbf’;
5> }
– Typed into an Operating System file, which is then invoked on the
Recovery Manager command line
The script above is typed in to an O/S file called copy_10, then is called
when invoking rman:
rman target internal/kernel@prod rcvcat rman/rman@rcat cmdfile
copy_10
This is a good method if you plan to schedule backups using O/S utilities
(e.g. cron)
Recovery Manager - Connie Dialeris
60
Running Recovery Manager
commands
•
The commands can also be stored in the Recovery Catalog, with a name. It is
then called a Stored Script.
A stored script is only associated with one target database.
RMAN> create script copy_10 {
2> allocate channel c1 type disk;
3> copy
4>
datafile 10 to ‘/oracle/prod/backup/prod_10.dbf’;
5> }
To execute the stored script:
RMAN> run {
2> execute script copy_10;
3> }
Recovery Manager - Connie Dialeris
61
Lists and Reports on the
Recovery Catalog
Recovery Manager - Connie Dialeris
62
List Command
• Queries the Recovery Catalog and produces a listing of it’s contents.
You can list:
– All databases known to the recovery catalog or incarnations of a database
RMAN> list incarnation of database;
– Backup Sets or File Copies containing a backup of a specified list of files
RMAN> list backupset of datafile 2;
RMAN> list backupset of datafile ‘/d3/dbf/user_prd1.dbf’;
– Backup Sets or File Copies containing a backup of any datafile which is a
member of a list of tablespaces
RMAN> list copy of tablespace system;
– Backup Sets or File Copies of the whole database, optionally restricted by
time, datafile copy name, device name and tag
– Backup Sets containing Archivelogs restricted by name, or by range. For
example, list which backup sets contain archivelogs between the sequence
123 and 145
RMAN> list backupset of archivelog
2> from logseq 123 until logseq 145 thread 1;
Recovery Manager - Connie Dialeris
63
Report Command
• Analyzes the content of the recovery catalog in more detail than list
• Produces reports which can answer questions such as:
– What files need a backup?
report need backup
– What files are not recoverable due to unrecoverable operations?
report unrecoverable
– What backups can be deleted (.e. are obsolete)?
report obsolete
– What was the physical schema of the database at a previous point-in-time?
report schema
Recovery Manager - Connie Dialeris
64
Report Command
•
and report
regular basis to ensure:
report need backup
unrecoverable
should be used on a
– Recovery is possible
RMAN> report unrecoverable;
– Recovery is possible within an acceptable time
For Example, the following command reports all datafiles which would
require the application of 3 or more incrementals to be recovered to their
current state
RMAN> report need backup incremental 3 database;
The following command reports all datafiles belonging to the SYSTEM
tablespace that haven’t had a backup (full or incremental) in 5 days or
more.
RMAN> report need backup days 5 tablespace system;
Recovery Manager - Connie Dialeris
65
Miscellaneous
Recovery Manager - Connie Dialeris
66
Troubleshooting
• Loss of the Recovery Catalog
– First method should be: to always have a valid backup of your Recovery
Catalog, and use the normal Oracle recovery mechanisms
– Second method:
• create a new recovery catalog
• catalog all objects (archivelogs, backup sets, file copies) which exist on disk
• resync the catalog
– From current controlfile (if available)
– From backup controlfile (if not available), then change...uncatalog all obsolete
records inserted into the recovery catalog from the resync from backup
controlfile
Recovery Manager - Connie Dialeris
67
Troubleshooting
•
“Failed to start RPC call”
RMAN-08010:
RMAN-10030:
RMAN-10036:
RMAN-08010:
channel c8: including datafile number 47 in backupset
RPC call appears to have failed to start on channel c9
RPC call ok on channel c9
channel c3: including datafile number 18 in backupset
This is not a problem. It indicates one of two things has occurred:
– The target database instance is slow
– Temporary timing problem
• Killing a Recovery Manager job
– interactively: ^c
– batch: kill the rman process
Recovery Manager - Connie Dialeris
Troubleshooting
• Additional Information Messages
RMAN-03007: exception occurred during execution, error is retryable
RMAN-07004: unhandled exception during command execution on channel c4
RMAN-10032: unhandled exception during execution of job step 4: ORA-06512: at line
158
RMAN-10035: exception raised in RPC: ORA-19624: operation failed, retry possible
ORA-19506: failed to create sequential file, name="df_98_1", parms=""
ORA-27007: failed to open file
HP-UX Error: 1003: Unknown system error
Additional information: 7004
Additional information: 1
ORA-06512: at "SYS.DBMS_BACKUP_RESTORE", line 410
RMAN-10031: ORA-19624 occurred during call to DBMS_BACKUP_RESTORE.BACKUPPIECEC
REATE
The 7004 above translates to: no tape device found
Decode them using skgfqsbt.h, or look in the README
Recovery Manager - Connie Dialeris
Message Stack:
read bottom up
Troubleshooting
• If a backup/restore/recover does not seem to be progressing, check the
progress of the job from within oracle by querying v$session_longops
select sid, serial#, context,
round(sofar/totalwork*100,2) "% Complete",
substr(to_char(sysdate,'yymmdd hh24:mi:ss'),1,15) "Time Now"
from v$session_longops
where compnam = 'dbms_backup_restore';
This will produce an output such as:
SID
SERIAL#
CONTEXT
% Complete Time Now
---------- ---------- ---------- ---------- --------------12
56
970408 14:21:07
If the % Complete is not increasing between intervals of 2 minutes,
there is a problem; it should actually increase faster. Check
v$session_wait to see what events within Oracle are being waited for:
select *
from v$session_wait
where wait_time = 0;
• Also check sbtio.log
Recovery Manager - Connie Dialeris
70
Troubleshooting
• Default SBT Interface
– It is possible to write to disk via the SBT interface which comes default
with Oracle.
– This is useful for simulating writing to tape when trying to identify
whether a problem encountered is a Media Manager problem, or an Oracle
problem.
– To use the interface:
• Ensure you do not have the Media Management library linked in
• Allocate channels of type ‘SBT_TAPE’
• Fully qualify the location to write the piece out to (must be a file system)
Recovery Manager - Connie Dialeris
Media Manager Issues
•
•
•
•
Maximum supported file sizes (size of backup piece)
Uniqueness of backup piece name
Restoring to another node
Reconciling the Recovery Catalog and the Media Manager's catalog
Ensure the Recovery Catalog is made aware of the unavailability of
any backups invalidated by Tape (re)cycling. Basically, the Recovery
Catalog must be kept in sync with the Media Manager’s catalog.
You can use rman change…uncatalog commands to do this.
Recovery Manager - Connie Dialeris
OPS Issues
• Restoring in parallel across nodes
• Snapshot control file
set snapshot controlfile to '/oracle/db_files/snapshot/snap_prod.cf';
• Archivelog Backup
All files must be readable by the ‘default’ connection during command
compilation time
• Archivelog Restore
Archivelogs may be restored to any node performing a Restore
Operation
• Archivelog Recovery
The archivelogs must be readable by the ‘default’ channel, as it is this
one that performs the recovery
Recovery Manager - Connie Dialeris
Performance
• Using I/O Slaves (An option if Async I/O is not available)
– ‘Slave’ server processes perform the reads and writes on behalf of the
channel, and allow the server process to continue processing while reads
are performed.
– Configured by setting the following init.ora parameters:
backup_tape_io_slaves = (true/false)
backup_disk_io_slaves = n
Where n is the number of slaves per channel
• Avoiding saturating one file with too many read-requests is already
achieved by Recovery Manager multiplexing files together, and
switching files frequently
• This has the benefit of keeping tape drive streaming, while also not
flooding a particular file with too many read requests. If backup is still
adversely affecting OLTP:
–
–
set limit readrate X (limits buffers read per file, per second)
each buffer is of size db_block_size*db_file_direct_io_count
Recovery Manager - Connie Dialeris
74
After Resetlogs
• Once a resetlogs has been performed on a database, you must notify
Recovery Manager of this before you attempt any further backup
operations
RMAN> reset database;
• Returning to an old Incarnation
– If the user has decided that they need to restore and recover the database
to a point-in-time before the resetlogs was performed, they need to
indicate this intent to Recovery Manager, by performing a:
RMAN> reset database to <incarnation>;
– The <incarnation> to revert to is determined by:
RMAN> list incarnation of database <identifier>;
Recovery Manager - Connie Dialeris
75
Questions
Recovery Manager - Connie Dialeris
76