Nielsen c41.tex V4 - 07/23/2009 2:46pm Page 982 Part VI Enterprise Data Management The Contents button can be used to check the disk device for specific backups. If several backup instances are in the backup device, the Contents dialog can be used to select the correct backup. It then fills in the file number in the restore form. Before restoring the msdb database, stop SQL Server Agent. This ensures that the msdb database is not being accessed by the SQL Server Agent and allows the restore to complete. Performing a Complete Recovery If the server has completely failed and all the backups must be restored onto a new server, this is the process to follow: 1. Build the Windows server and restore the domain logins to support Windows authentication. 2. Install SQL Server and any service-pack, cumulative updates, security upgrades, or hotfixes. 3. Start SQL Server in single-user mode and restore the master database. 4. Verify that SQL Server Agent is stopped. Restore the msdb database. 5. If the model database was modified, restore it. 6. Restore the user databases. Best Practice P erforming a flawless recovery is a ‘‘bet your career’’ skill. I recommend taking the time to work through a complete recovery of the production data to a backup server. The confidence it builds will serve you well as a SQL Server DBA. Summary The recovery cycle begins with the backup of the databases. The ability to survive hardware failure or human error is crucial to the ACID properties of a database. Without the transaction’s durability, the database can’t be fully trusted. Because of this, recovery planning and the transaction log provide dura- bility to committed transactions. The recovery cycle transfers data from the past to the present. They key points from this chapter include the following: ■ Invest the time to create a solid backup and recovery plan. ■ Just performing regular backups is not enough. Because the only way to know that the backups are good is by restoring them, regularly restore the backups on a test server. This is a boring task but well worth it when a disaster occurs and you need to recover from the backups. 982 www.getcoolebook.com Nielsen c41.tex V4 - 07/23/2009 2:46pm Page 983 Recovery Planning 41 ■ I recommend that a senior SQL DBA create the backup and recovery plan to ensure that all aspects are taken care of. Once the plan is ready, ascertain that the junior DBA can understand it and perform all the steps in the plan, as it is more than likely that this is the person who will be recovering from the backups in the middle of the night when a disaster occurs. ■ Perform a complete recovery to simulate a server and disk subsystem failure at least every six months and update your backup/recovery plan as required. In the next chapter, you’ll learn how to maintain the database. 983 www.getcoolebook.com Nielsen c41.tex V4 - 07/23/2009 2:46pm Page 984 www.getcoolebook.com Nielsen c42.tex V4 - 07/21/2009 2:25pm Page 985 Maintaining the Database IN THIS CHAPTER Using SQL Server’s Database Console Commands (DBCC) Creating database maintenance plans T he previous chapter covered database recovery planning. This chapter explores various database maintenance tasks that need to be performed regularly, such as database backups, database integrity checks, and index maintenance. It also discusses database maintenance using Transact-SQL Database Console Commands (DBCC) and Maintenance Plans. DBCC Commands Microsoft SQL Server Database Console Commands (DBCC) are used for check- ing database integrity, performing maintenance operations on databases, tables, indexes, and filegroups, and collecting a nd displaying information during trou- bleshooting issues. The first DBCC command to become familiar with is the DBCC HELP command, which returns the syntax and all the options for any DBCC command. The following command returns the syntax for DBCC CHECKDB: DBCC HELP (’CHECKDB’); Result: dbcc CHECKDB ( { ‘database_name’ | database_id | 0 } [ , NOINDEX | { REPAIR_ALLOW_DATA_LOSS | REPAIR_FAST | REPAIR_REBUILD }] ) 985 www.getcoolebook.com Nielsen c42.tex V4 - 07/21/2009 2:25pm Page 986 Part VI Enterprise Data Management [ WITH { [ ALL_ERRORMSGS ] [ , [ NO_INFOMSGS ] ] [ , [ TABLOCK ] ] [ , [ ESTIMATEONLY ] ] [ , [ PHYSICAL_ONLY ] ] [ , [ DATA_PURITY ] ] [ , [ EXTENDED_LOGICAL_CHECKS ] ] } ] DBCC execution completed. If DBCC printed error messages, contact your system administrator. What’s New in SQL Server Database Maintenance? I n earlier versions of SQL Server, DBCC stood for ‘‘Database Consistency Checker’’ — now it is renamed ‘‘Database Console Commands.’’ The following deprecated DBCC commands are still in SQL Server 2008, but they will be removed in the next version. It is recommended to stop using the deprecated DBCC commands for new development work and plan to change existing applications using these commands: ■ DBCC DBREINDEX ■ DBCC INDEXDEFRAG ■ DBCC SHOWCONTIG ■ DBCC PINTABLE ■ DBCC UNPINTABLE While the DBCC PINTABLE and DBCC UNPINTABLE commands are still in SQL Server 2008, they have no effect on SQL Server. Note also that DBCC CONCURRENCYVIOLATION has been discontinued in SQL Server 2008. You can find a complete list of deprecated (outdated or no longer accessible) DBCC commands in Chapter 1, ‘‘The World of SQL Server,’’ in Books Online, and at http://msdn2.microsoft.com/en-US/library/ ms144262.aspx. The following command returns all DBCC commands for which help is available: DBCC HELP (’?’); Result: checkalloc checkcatalog 986 www.getcoolebook.com Nielsen c42.tex V4 - 07/21/2009 2:25pm Page 987 Maintaining the Database 42 checkconstraints checkdb checkfilegroup checkident checktable cleantable dbreindex dropcleanbuffers free freeproccache freesessioncache freesystemcache help indexdefrag inputbuffer opentran outputbuffer pintable proccache show_statistics showcontig shrinkdatabase shrinkfile sqlperf 987 www.getcoolebook.com Nielsen c42.tex V4 - 07/21/2009 2:25pm Page 988 Part VI Enterprise Data Management traceoff traceon tracestatus unpintable updateusage useroptions DBCC execution completed. If DBCC printed error messages, contact your system administrator. All DBCC commands report their activity or errors found, and then conclude with the standard ‘‘DBCC execution completed’’ statement, including any action that might be needed. Database integrity DBCC CHECKDB performs several operations to check the logical and physical integrity of the database. It’s critical for the health of the database that the physical structure is correct. DBCC CHECKDB checks things such as index pointers, data-page offsets, the linking between data pages and index pages, and the structural content of the data and index pages. If a hardware hiccup has left a data page half-written, then DBCC CHECKDB is the best means of detecting the problem. The following command executes DBCC CHECKDB on the AdventureWorks2008 sample database: DBCC CHECKDB (’AdventureWorks2008’); Result (abridged): DBCC results for ‘AdventureWorks2008’. Service Broker Msg 9675, State 1: Message Types analyzed: 14. Service Broker Msg 9676, State 1: Service Contracts analyzed: 6. Service Broker Msg 9667, State 1: Services analyzed: 3. Service Broker Msg 9668, State 1: Service Queues analyzed: 3 . DBCC results for ‘sys.sysrscols’. There are 1406 rows in 14 pages for object "sys.sysrscols". DBCC results for ‘sys.sysrowsets’. There are 263 rows in 3 pages for object "sys.sysrowsets". DBCC results for ‘sys.sysallocunits’. DBCC results for ‘Production.ProductModelProductDescriptionCulture’. There are 762 rows in 4 pages for object "Production.ProductModelProductDescriptionCulture". DBCC results for ‘Sales.Store’. There are 701 rows in 101 pages for object "Sales.Store". DBCC results for ‘Production.ProductPhoto’. There are 101 rows in 50 pages for object "Production.ProductPhoto". 988 www.getcoolebook.com Nielsen c42.tex V4 - 07/21/2009 2:25pm Page 989 Maintaining the Database 42 CHECKDB found 0 allocation errors and 0 consistency errors in database ‘AdventureWorks2008’. DBCC execution completed. If DBCC printed error messages, contact your system administrator. The results you see could vary from those displayed here depending on your SQL Server configuration and any changes you have made to the database. Two options simply determine which messages are reported, without altering the functionality of the integrity check: ALL_ERRORMSGS and NO_INFOMSGS. ALL_ERRORMSGS displays all the error mes- sages; when it is not used, the default displays 200 errors per object. The NO_INFOMSGS option can be used to suppress all informational messages. The ESTIMATE_ONLY option returns the estimated size of the tempdb required by DBCC CHECKDB without actually running DBCC CHECKDB against the database. If the database is large, then the NOINDEX option can be used to skip checking the integrity of all user- table non-clustered indexes. For additional time savings, the PHYSICAL_ONLY option performs only the most critical checks on the physical structure of the pages. Use these options only when time prevents a complete DBCC CHECKDB or when the indexes are about to be rebuilt. If you want to perform logical consistency checks on indexed views, XML indexes, and spatial indexes, then use the new option EXTENDED_LOGICAL_CHECKS. This option increases the performance impact on the SQL Server, and currently its progress cannot be tracked. It is recommended that you first run this on a similar database on a similar test SQL Server to get an estimate of the time it takes to run DBCC CHECKDB with this option and the performance impact on the SQL Server. For databases upgraded from SQL Server 2000 or earlier, it is recommended to run DBCC CHECKDB with DATA_PURITY as a post-upgrade step to enable the column-value integrity checks, as they are dis- abled by default on SQL Server 2000 and earlier. After successful completion, column-value integrity checks are enabled for the database, and you do not need to use the DATA_PURITY option for future DBCC CHECKDB executions. Column-value integrity checks are enabled by default beginning with SQL Server 2005. To reduce blocking and concurrency problems when integrity checks are done, DBCC CHECKDB uses an internal database snapshot to perform the checks. If a snapshot cannot be created or you use the TABLOCK option, then DBCC CHECKDB uses locks, which include a short-term exclusive lock on the database. The TABLOCK option can reduce the time it takes for DBCC CHECKDB to run during peak usage, but it reduces the concurrency on the database. If the TABLOCK option is used, then DBCC CHECKCATALOG is skipped, and Service Broker data is not checked. Review the percent_complete and command columns of the sys.dm_exec_requests cata- log view to display the progress and current phase of DBCC CHECKDB. Repairing the database When an error is found and DBCC CHECKDB can fix it, DBCC CHECKDB indicates the repair level that is needed to repair the error(s). When an error is reported by DBCC CHECKDB, it is recommended to restore the database from a known good backup. Use the repair option with DBCC CHECKDB only when there is no known good backup. Repairing the database is a separate operation from the normal DBCC 989 www.getcoolebook.com Nielsen c42.tex V4 - 07/21/2009 2:25pm Page 990 Part VI Enterprise Data Management CHECKDB command because the database needs to be placed in single-user mode with the ALTER DATABASE command before a DBCC CHECKDB can be executed with the REPAIR option. The following command places the AdventureWorks2008 sample database in single-user mode: ALTER DATABASE AdventureWorks2008 SET SINGLE_USER; More on Single-User Mode T he preceding command will wait indefinitely if there is a lock on the database or users are connected to the database. You can use a termination clause, WITH ROLLBACK AFTER integer [SECONDS] or WITH ROLLBACK IMMEDIATE, to indicate that SQL Server should either roll back incomplete transactions after the specified number of seconds or roll back immediately and close any active connections to the database. The following command places the AdventureWorks2008 sample database in single-user mode and specifies that SQL Server should roll back all incomplete transactions immediately and close any active connections to the database: ALTER DATABASE AdventureWorks2008 SET SINGLE_USER WITH ROLLBACK IMMEDIATE; If the AUTO_UPDATE_STATISTICS_AYSNC option for the database is set to ON, you will be unable to place the database in single-user mode, as the background thread that is used to update the statistics takes a connection against the database. You can query the is_auto_update_stats_async_on column in the sys.databases catalog view to check if the AUTO_UPDATE_STATISTICS_ASYNC option is set to ON.If so,thenfirstsetittoOFF using the ALTER DATABASE command. DBCC offers two repair modes: ■ REPAIR_REBUILD: This mode performs a repair that does not lead to any data loss. ■ REPAIR_ALLOW_DATA_LOSS: This mode performs the repairs and fixes corrupted database structures. As the name suggests, this mode can result in data loss. Earlier versions of SQL Server included a third repair mode called REPAIR_FAST. This mode still exists in SQL Server 2008 but it does not perform any activity and is kept only for backward compatibility. The following example places the AdventureWorks2008 sample database in single-user mode and runs DBCC CHECKDB with the REPAIR_ALLOW_DATA_LOSS option. It then sets the database back to multi-user mode: ALTER DATABASE AdventureWorks2008 SET SINGLE_USER WITH ROLLBACK IMMEDIATE; BEGIN TRANSACTION; DBCC CheckDB (’AdventureWorks2008’, REPAIR_ALLOW_DATA_LOSS); Check for any data loss ROLLBACK TRANSACTION if data loss is not acceptable else COMMIT TRANSACTION; ALTER DATABASE AdventureWorks2008 SET MULTI_USER; 990 www.getcoolebook.com Nielsen c42.tex V4 - 07/21/2009 2:25pm Page 991 Maintaining the Database 42 Result (abridged): DBCC results for ‘AdventureWorks2008’. Service Broker Msg 9675, State 1: Message Types analyzed: 14. Service Broker Msg 9676, State 1: Service Contracts analyzed: 6. Service Broker Msg 9667, State 1: Services analyzed: 3. DBCC results for ‘sys.sysrowsets’. There are 291 rows in 3 pages for object "sys.sysrowsets". DBCC results for ‘Production.TransactionHistoryArchive’. There are 89253 rows in 620 pages for object "Production.TransactionHistoryArchive". CHECKDB found 0 allocation errors and 0 consistency errors in database ‘AdventureWorks2008’. DBCC execution completed. If DBCC printed error messages, contact your system administrator. Best Practice S ince SQL Server 7.0, the Storage Engine’s quality has significantly improved, reducing the need to run DBCC CHECKDB frequently. However, a foolproof recovery plan will include a full restore and a DBCC CHECKDB to ensure that all portions of the recovery plan are working. The frequency with which you run DBCC CHECKDB depends on your comfort level, your environment, and the importance of your data. I would never say that DBCC CHECKDB is not needed; it is recommended to run it now and then, as well as to restore from backups. It is also recommended to run DBCC CHECKDB after any hardware malfunction. If an error is detected, then restore from a known good database backup. As mentioned, use the repair option only as the last option. If DBCC CHECKDB asks you to use REPAIR_ALLOW_DATA_LOSS, then take a full database backup first and then run DBCC CHECKDB with the repair option in a user transaction. That way, you can verify the data loss after the command is executed and roll back the transaction if the data loss is not acceptable. After successfully repairing the database, take a full database backup. Because DBCC CHECKDB is a resource-intensive operation, run it during low peak hours. Multi-user concerns DBCC CHECKDB without any repair option can be executed while users are in the database. However, DBCC CHECKDB is very processor- and disk-intensive, so run it when the database has the fewest users. By default, DBCC CHECKDB can check objects in parallel across all CPUs in the box, which can boost the performance of DBCC CHECKDB and reduce the run times. The degree of parallelism is determined by the SQL Server query processor, and the algorithm used is similar to running parallel queries. Because running DBCC CHECKDB in parallel is processor intensive, you may have environments or situa- tions when you want to disable parallel checking of objects by DBCC CHECKDB. SQL Server enables you to do that using the trace flag 2528. 991 www.getcoolebook.com . UNPINTABLE commands are still in SQL Server 2008, they have no effect on SQL Server. Note also that DBCC CONCURRENCYVIOLATION has been discontinued in SQL Server 2008. You can find a complete list. database, stop SQL Server Agent. This ensures that the msdb database is not being accessed by the SQL Server Agent and allows the restore to complete. Performing a Complete Recovery If the server has. mode can result in data loss. Earlier versions of SQL Server included a third repair mode called REPAIR_FAST. This mode still exists in SQL Server 2008 but it does not perform any activity and is