Summary
Beginning with Oracle Database 10g, there are certain configurations where Data Guard Redo Apply (physical standby) can support different Oracle binaries between primary and standby databases as long as all systems in the Data Guard configuration are running
the same OS family. For example, you may have a primary database with 32bit Oracle binaries on Red Hat 32bit, and a physical standby database with 64bit Oracle binaries on Red Hat 64bit. Such configurations must follow additional procedures during Data Guard
role transitions (switchover and failover) as described in this MetaLink note, below. If these procedures are not followed then errors will occur when trying to open the new primary database. One example error sequence would be:
從Oracle10g開始，支持在不同的位數（32位和64位）之間做物理standby，但是前提必須是同樣的系統平臺，如都是REDHAT的。例如在REDHAT 32位平臺上配置一個32bit的主庫，64bit的物理備庫在REDHAT 64bit操作系統上。這樣的主備庫搭建，在進行角色切換（包括switchover and failover），切換後必須額外執行下面的操作步驟來保證角色切換完成，否則會出現如下錯誤：
ORA-00604: error occurred at recursive SQL level 1
ORA-06553: PLS-801: internal error [56327]
For additional information on how Data Guard supports configurations in which the primary database/system may not be identical to its standby database(s)/system(s), please see MetaLink Note 413484.1
oracle Data Guard 異構支持列表可以參照MetaLink Note 413484.1，也可以參照http://blog.csdn.net/java3344520/article/details/8075876
Restrictions
SQL Apply (logical standby) always requires that the bit-ness of the Oracle Database is identical on all systems in a Data Guard configuration. Continuing with the example provided in the summary above of a Red Hat 32bit/64bit primary/standby, if you are using
SQL Apply you must use 32 bit Oracle binaries on both primary and standby.
The bit-ness of the Oracle database must be identical if you are using any Oracle release before Oracle Database 10g for both SQL Apply and Redo Apply. Continuing with the example provided in the summary above of a Red Hat 32bit/64bit primary/standby, if you are using Oracle9i you must use 32 bit Oracle binaries on both primary and standby (either physical or logical standby).
Oracle Data Guard 10g以前的物理和邏輯standby的搭建，主備庫都必須在相同的位數（32位和64位）操作系統平臺上。
The Data Guard Broker does not support different binaries in the same Data Guard configuration in Oracle Database 10g. All such configurations must be managed from the SQL*Plus command line. Beginning with Data Guard 11g onward, this restriction is lifted and
the Data Guard Broker does support Oracle 11g databases of mixed word-size in the same Data Guard configuration.
Data Guard Fast-start Failover will failover in a mixed Oracle binary configuration but it will be unable to open the new primary database until the procedure below is followed.
Data Guard Broker 不支持配置不同位數的standby的搭建，只能通過SQL*PlusS命令行操作。11g之後版本，這個限制不存在，Data Guard Broker支持不同位數配置操作。Data Guard 不同位數主備庫的Failover切換必須執行下列步驟，否則無法打開（切換後）新的主庫。
Additional Role Transition Procedures when using mixed Oracle binaries:
If you are running a physical standby configuration where the primary and standby have different Oracle binaries you will need to follow the instructions below on the new primary as part of any Data Guard role transition (switchover or failover) after a standby becomes a primary database (supported configurations are defined in MetaLink Note 413484.1). This procedure is necessary when transitioning roles in either direction.
（不同位數主備庫角色切換必須執行的步驟）
Note: If you are using Oracle OLAP then you must also follow MetaLink Note 352306.1 . Additionally, as a precaution for a failover, you may want to backup/export the user-created AW’s ahead of time and periodically.
1. Shutdown the new primary database
If using RAC then set database to unclustered: SQL> alter system set cluster_database=false scope=spfile;
SQL> shutdown immediate
2. Startup the new primary in UPGRADE mode
SQL> startup upgrade
3. Invalidate existing PL/SQL modules
Invalidate existing PL/SQL modules in preparation to convert the format required by the new primary. Note that there should be no other DDL on the database while running the script. Not following thisrecommendation may lead to deadlocks.
SQL> spool utlirp.log
SQL> @?/rdbms/admin/utlirp
SQL> spool off
4. Restart the database
If using RAC then set database back to clustered:
SQL> alter system set cluster_database=true scope=spfile;
SQL> shutdown immediate
SQL> startup
5. Recompile all invalid PL/SQL modules
Recompile all invalid PL/SQL modules to be in the format required by the new primary database.
SQL> spool utlrp.log
SQL> @?/rdbms/admin/utlrp
SQL> spool off
6. Verify that all expected packages and classes are valid.
SQL> SELECT count(*) FROM dba_objects WHERE status=’INVALID’;
SQL> SELECT distinct object_name FROM dba_objects WHERE status=’INVALID’;
