D ata Protection with Oracle Data Guard

1. Data Protection with Oracle Data Guard Jacek Wojcieszuk, CERN/IT-DM Distributed Database Operations Workshop November 12th, 2008

2. Outline • Main challenges • Oracle Maximum Availability Architecture • Backup and recovery evolution @ CERN • Data Guard Physical Standby Database • Disaster recovery for production DBS • Standby technology for large-scale testing • Possible RMAN backup improvements Distributed Databse Operations Workshop - 2

3. Databases for LHC experiments - challenges Data volume Very quick, linear growth of databases (especially during LHC run) Some databases may reach 10 TB already next year Database availability On-line databases highly critical for data taking Few hours of downtime can already lead to data loss Off-line databases critical for data distribution and analysis Distributed Databse Operations Workshop - 3

4. ... More challenges With data volume increase: Increases probability of physical data corruption Increases frequency of human errors Traditional RMAN and tape-based approach doesn’t fit well into this picture: Leads to backup and recovery times proportional to or dependent on data volume Makes certain recovery scenarios quite complex e.g: recovery of a single database object Distributed Databse Operations Workshop - 4

5. Maximum Availability Architecture (MAA) Oracle's best practices blueprint Goal: to achieve the optimal high availability architecture at the lowest cost and complexity Helps to minimize impact of different types of unplanned and planned downtimes Is based on such Oracle products/features like: RAC ASM RMAN Flashback Data Guard http://www.oracle.com/technology/deploy/availability/htdocs/maa.htm Distributed Databse Operations Workshop - 5

6. Evolution towards MMA at CERN – DAS era Clients WAN/Intranet DB1 DB2 DB3 RMAN RMAN RMAN Distributed Databse Operations Workshop - 6

7. Evolution towards MMA at CERN – RAC+ASM era Clients WAN/Intranet RAC database with ASM SAN RMAN Distributed Databse Operations Workshop - 7

8. Evolution towards MMA at CERN – RAC + ASM + on-disk copies Clients WAN/Intranet RAC database with ASM and on-disk copy SAN RMAN Distributed Databse Operations Workshop - 8

9. Strong and weak points Server failure: RAC keeps the database available Failure at the storage level: ASM keeps data healthy Small physical corruption: On-disk or on-tape backup can be efficiently used to resolve the problem Logical corruption (human error): On-disk backup can be used restore damaged data quite quickly However the procedure is labour-intensive and error-prone Wide-range physical data corruption: On-disk backup can be used if not corrupted On-tape backup can be used but time-to-restore proportional to the database size Disasters (flood, fire, overvoltage, etc): On-tape backup can be usedbut time-to-restore proportional to the database size Distributed Databse Operations Workshop - 9

10. Solution – Data Guard Physical Standby Database Mature and stable technology: Introduced in Oracle 8i Relying on old and proven functionality: redo generation media recovery Flexible configuration Synchronous or asynchronous propagation of data changes Immediate or delayed standby database’s updates Different protection levels: Maximum performance Maximum availability Maximum protection Small performance overhead on the primary system Distributed Databse Operations Workshop - 10

11. Data Guard Physical Standby Database Physical Standby RAC database with ASM WAN/Intranet Data changes Primary RAC database with ASM RMAN Distributed Databse Operations Workshop - 11

12. Data Guard configuration details Standby databases configured as RAC Servers sized to handle moderated applications’ load Enough disk space to fit all the data and few days of archived redo logs No Flash Recovery Area Identical OS and Oracle RDBMS version on primary and standby The same patch level Asynchronous data transmission with the LGWR process Standby redo logs necessary Shipped redo data applied with 24 hours delay No fast-start-failover Distributed Databse Operations Workshop - 12

13. Handling human errors If a logical data corruption is discovered within assumed lag period (24 hours) standby database can be used to recover corrupted data Procedure: Stop all standby RAC instances but one Disable managed recovery mode: STDBY> ALTER DATABASE RECOVER MANAGED STANDBY DATABASE CANCEL; • If needed roll standby database forward to a desired point in time • Create a guaranteed restore point on a standby database: STDBY> CREATE RESTORE POINT bef_stop GUARANTEE FLASHBACK DATABASE; • Switch logs and disable log transmission: PRIMARY> ALTER SYSTEM ARCHIVE LOG CURRENT; PRIMARY>ALTER SYSTEM SET LOG_ARCHIVE_DEST_STATE_x=DEFER SID='*'; STDBY> ALTER SYSTEM SET LOG_ARCHIVE_DEST_STATE_x=DEFER SID='*'; Distributed Databse Operations Workshop - 13

14. Handling human errors (2) Open standby database for writing: STDBY> ALTER DATABASE ACTIVATE STANDBY DATABASE; STDBY> ALTER DATABASE OPEN; • Copy over corrupted data to the primary database • Flashback database and resume standby: STDBY> STARTUP MOUNT FORCE; STDBY> FLASHBACK DATABASE TO RESTORE POINT bef_stop; STDBY> ALTER DATABASE CONVERT TO PHYSICAL STANDBY; STDBY> STARTUP MOUNT FORCE; STDBY>ALTER DATABASE RECOVER MANAGED STANDBY DATABASE DISCONNECT; • Cleanup: STDBY> DROP RESTORE POINT bef_stop; STDBY>ALTER SYSTEM SET LOG_ARCHIVE_DEST_STATE_x=ENABLE SID='*'; PRIMARY>ALTER SYSTEM SET LOG_ARCHIVE_DEST_STATE_x=ENABLE SID='*'; Distributed Databse Operations Workshop - 14

15. Running tests Standby Database can be open read-write to perform applications’ tests Performance tests Change/migration tests Etc... Especially useful in case of applications with a lot of data Procedure is basically identical to the one used for handling human errors While standby database is open read-write it it can’t receive data changes from the primary system Distributed Databse Operations Workshop - 15

16. Handling wide-range corruptions and disasters In case the primary system becomes unavailable clients can be failed over to the standby database Data Guard environment can be configured for either automatic (fast-start-failover) or manual failover The most tricky part is re-directing clients to the standby database in an efficient way Two approaches: DNS alias-based Oracle service-based Distributed Databse Operations Workshop - 16

17. DNS alias based failover A DNS alias assigned to each VIP of the primary RAC Clients specify those DNS aliases in their connection descriptors During failover aliases have to be moved to VIPs of the standby cluster Pros: Simplicity Cons: Moving aliases requires noticeable time (~30 min at CERN) Problematic when standby RAC has more nodes than the primary one alias1 alias2 alias3 Oracle Service_A Oracle Service_A Primary RAC database Standby RAC database App_A = DESCRIPTION= (ADDRESS= (PROTOCOL=TCP) (HOST=alias1) (PORT=1521) ) (ADDRESS= (PROTOCOL=TCP) (HOST=alias2) (PORT=1521) ) (ADDRESS= (PROTOCOL=TCP) (HOST=alias3) (PORT=1521) ) ... (SERVICE_NAME=Service_A) ... Distributed Databse Operations Workshop - 17

18. Oracle service based failover Client specify both primary and standby db VIPs in the connection descriptor Oracle service used by the client started only on the primary After failover a trigger starts the service on the standby database Clients use SQLNET.OUTBOUND_CONNECTION_TIMEOUT Pros: Relatively short downtime associated with the failover Cons: Quite complex setup stdby1 prim1 stdby2 prim2 stdby3 prim3 Oracle Service_A Primary RAC database Standby RAC database App_A = DESCRIPTION= (ADDRESS= (PROTOCOL=TCP) (HOST=prim1) (PORT=1521) ) (ADDRESS= (PROTOCOL=TCP) (HOST=prim2) (PORT=1521) ) (ADDRESS= (PROTOCOL=TCP) (HOST=prim3) (PORT=1521) ) (ADDRESS= (PROTOCOL=TCP) (HOST=stdby1) (PORT=1521) ) (ADDRESS= (PROTOCOL=TCP) (HOST=stdby2) (PORT=1521) ) (ADDRESS= (PROTOCOL=TCP) (HOST=stdby3) (PORT=1521) ) ... (SERVICE_NAME=Service_A) ... Oracle Service_A Distributed Databse Operations Workshop - 18

19. Possible improvements – Active Data Guard New feature of Oracle 11gR1: Extra-cost option for Oracle RDBMS 11g Enterprise Edition Allows clients to connect to and query a physical standby database while ‘managed recovery’ mode enabled: Better resource utilization Excellent idea for read-only workload Many potential use-cases at CERN e.g. CMS on-line database Extensive tests started Distributed Databse Operations Workshop - 19

20. Possible improvements – LAN-free tape backups Traditionally tape backups are sent over a general purpose network to a media management server: This may limit backup speed to ~80 MB/s At the same time tape drives can archive data with the speed of 100-200 MB/s compressed Tivoli Storage Manager supports so-called LAN-free backup: Backup data flows to tape drives directly over SAN Media Management Server used only to register backups Preliminary tests show that with 2 tape drives a database can be backed up with the speed of ~400 MB/s Media Management Server Metadata Database RMAN backups RMAN backups Tape drives Distributed Databse Operations Workshop - 20

21. Possible improvements - Using a disk pool instead of tapes Tape infrastructure is expensive and difficult to maintain: costly hardware and software noticeable maintenance effort tape media is quite unreliable and needs to be validated At the same time disk space is getting cheaper and cheaper: 1.5 TB SATA disks already available Pool of disks can be easily configured as destination for RMAN backups: Can simplify backup infrastructure Can improve backup performance Can increase backup reliability Several configurations possible: NFS mounted disks SAN-attached storage with CFS SAN-attached storage with ASM Tests is progress Distributed Databse Operations Workshop - 21

22. Conclusions Distributed Databse Operations Workshop - 22 • Production databases are growing very large • Recovery time in case of failure becomes critical • Procedures successfully tested to make recovery time scale: • On-disk backup • Physical Standby for disaster recovery • Physical Standby for large-scale testing • RMAN backup optimisations

23. Q&A Thank you Distributed Databse Operations Workshop - 23