我们在学习Oracle基础知识的时候会了解到实例恢复(Instance Recovery)或者说崩溃恢复(Crash recovery)的概念，有时候甚至于这2个名词在我们日常的语言中表达同样的意思。实际上Instance Recovery与Crash Recovery是存在区别的：针对单实例(single instance)或者RAC中所有节点全部崩溃后的恢复，我们称之为Crash Recovery。而对于RAC中的某一个节点失败，存活节点(surviving instance)试图对失败节点线程上redo做应用的情况，我们称之为Instance Recovery。
不管是Instance Recovery还是Crash Recovery，都由2个部分组成：cache recovery继以transaction recovery。
根据官方文档的介绍，Cache Recovery也叫Rolling Forward，就是我们常说的前滚；而Transaction Recovery也叫Rolling Back，就是我们常说的回滚。前滚和回滚贯穿Oracle恢复的基本概念，是我们入门必要学习的知识，在次不多介绍。
有文事者，必济之以武略。理论学得再好，不实践也无用。所幸Crash Recovery是很容易做成的实验，我们不妨来看一看:
SQL> shutdown abort;
ORACLE instance shut down.
SQL> startup mount;
ORACLE instance started.
Total System Global Area 1065353216 bytes
Fixed Size 2089336 bytes
Variable Size 486542984 bytes
Area
iis7站长之家 Buffers 570425344 bytes
Redo Buffers 6295552 bytes
Database mounted.
SQL> alter database open;
Crash Recovery将从alter database open开始，我们来观察其日志
==================alert.log====================
alter database open
Tue Jun 14 18:19:53 2011
Beginning crash recovery of 1 threads
parallel recovery started with 2 processes
Tue Jun 14 18:19:53 2011
Started redo scan
Tue Jun 14 18:19:53 2011
Completed redo scan
0 redo blocks read, 0 data blocks need recovery
Tue Jun 14 18:19:53 2011
Started redo application at
Thread 1: logseq 1004, block 1124, scn 17136185
Tue Jun 14 18:19:53 2011
Recovery of Online Redo Log: Thread 1 Group 2 Seq 1004 Reading mem 0
mem# 0: /flashcard/oradata/g10r2/onlinelog/o1_mf_2_6v34jokt_.log
Mem# 1: /s01/flash_recovery_area/G10R2/onlinelog/o1_mf_2_6v34jotq_.log
Tue Jun 14 18:19:53 2011
Completed redo application
Tue Jun 14 18:19:53 2011
Completed crash recovery at
Thread 1: logseq 1004, block 1124, scn 17156186
0 data blocks read, 0 data blocks written, 0 redo blocks read
Tue Jun 14 18:19:53 2011
LGWR: STARTING ARCH PROCESSES
ARC0: Archival started
LGWR: STARTING ARCH PROCESSES COMPLETE
ARC0 started with pid=16, OS id=7829
Tue Jun 14 18:19:53 2011
Thread 1 advanced to log sequence 1005 (thread open)
Thread 1 opened at log sequence 1005
Current log# 3 seq# 1005 mem# 0: /flashcard/oradata/G10R2/onlinelog/o1_mf_3_6v34jpmp_.log
Current log# 3 seq# 1005 mem# 1: /s01/flash_recovery_area/G10R2/onlinelog/o1_mf_3_6v34jpyn_.log
Successful open of redo thread 1
Tue Jun 14 18:19:53 2011
ARC0: Becoming the 'no FAL' ARCH
ARC0: Becoming the 'no SRL' ARCH
ARC0: Becoming the heartbeat ARCH
Tue Jun 14 18:19:53 2011
SMON: enabling cache recovery
Tue Jun 14 18:19:53 2011
db_recovery_file_dest_size of 204800 MB is 6.81% used. This is a
user-specified limit on the amount of space that will be used by this
database for recovery-related files, and does not reflect the amount of
space available in the underlying filesystem or ASM diskgroup.
Tue Jun 14 18:19:54 2011
Successfully onlined Undo Tablespace 1.
Tue Jun 14 18:19:54 2011
SMON: enabling tx recovery
Database Characterset is UTF8
Opening with internal Resource Manager plan
where NUMA PG = 1, CPUs = 2
replication_dependency_tracking turned off (no async multimaster replication found)
Starting background process QMNC
QMNC started with pid=17, OS id=7831
Tue Jun 14 18:19:55 2011
Completed: alter database open
注意上述单实例Crash Recovery到数据库打开的整个过程:
alter database open
Beginning crash recovery of 1 threads
Started redo scan
Completed redo scan
Started redo application at
Completed redo application
Completed crash recovery at
SMON: enabling cache recovery
Successfully onlined Undo Tablespace 1
SMON: enabling tx recovery
Completed: alter database open
从上述步骤中我们可以看到三种恢复名词，即：
crash recovery
cache recovery
tx recovery
这和官方文档所描述的Crash Recovery概念是不一致的，我们现在来理清这几种recovery。
crash recovery包含对redo的scan和application，显然其完成的是Rolling Forward前滚的工作，告警日志中出现的crash recovery等同于官方文档中介绍的”cache recovery”，我们可以将” Completed crash recovery”看做前滚完成的标志。而tx recovery从字面就可以看出实际上是Transaction Recovery，tx recovery发生在Undo Tablespace online之后(回滚事务的前提是Undo可用)，数据完成打开操作之前(“Completed: alter database open”)。注意tx recovery并不要求数据库打开前完成，仅仅是在数据库打开之前由smon启动(“SMON: enabling tx recovery”)。
剩下的唯一的问题是，这里的cache recovery是什么？显然它不是官方文档中所描述的”cache recovery”，几乎没有任何文档介绍存在这样一个recovery操作，这也是本文重点要介绍的。
我们来看另一个演示，这个演示用以说明cache recovery还存在于最普通的不包含Crash Recovery的数据库打开过程中:
SQL> shutdown immediate;
Database closed.
Database dismounted.
ORACLE instance shut down.
SQL> startup mount;
ORACLE instance started.
Total System Global Area 1065353216 bytes
Fixed Size 2089336 bytes
Variable Size 486542984 bytes
Database Buffers 570425344 bytes
Redo Buffers 6295552 bytes
Database mounted.
SQL> alter database open;
Database altered.
BANNER
----------------------------------------------------------------
Oracle Database 10g Enterprise Edition Release 10.2.0.4.0 - 64bi
PL/SQL Release 10.2.0.4.0 - Production
CORE 10.2.0.4.0 Production
TNS for Linux: Version 10.2.0.4.0 - Production
NLSRTL Version 10.2.0.4.0 - Production
SQL> select * from global_name;
GLOBAL_NAME
--------------------------------------------------------------------------------
==================alert.log====================
alter database open
Tue Jun 14 18:43:52 2011
LGWR: STARTING ARCH PROCESSES
ARC0: Archival started
LGWR: STARTING ARCH PROCESSES COMPLETE
ARC0 started with pid=14, OS id=8133
Tue Jun 14 18:43:52 2011
Thread 1 opened at log sequence 1005
Current log# 3 seq# 1005 mem# 0: /flashcard/oradata/G10R2/onlinelog/o1_mf_3_6v34jpmp_.log
Current log# 3 seq# 1005 mem# 1: /s01/flash_recovery_area/G10R2/onlinelog/o1_mf_3_6v34jpyn_.log
Successful open of redo thread 1
Tue Jun 14 18:43:52 2011
ARC0: Becoming the 'no FAL' ARCH
ARC0: Becoming the 'no SRL' ARCH
ARC0: Becoming the heartbeat ARCH
Tue Jun 14 18:43:52 2011
SMON: enabling cache recovery
Tue Jun 14 18:43:53 2011
Successfully onlined Undo Tablespace 1.
Tue Jun 14 18:43:53 2011
SMON: enabling tx recovery
Database Characterset is UTF8
Opening with internal Resource Manager plan
where NUMA PG = 1, CPUs = 2
replication_dependency_tracking turned off (no async multimaster replication found)
Tue Jun 14 18:43:53 2011
incremental checkpoint up to rba [0x3ed.624.0], current log tail at rba [0x3ed.944.0]
Tue Jun 14 18:43:53 2011
Starting background process QMNC
QMNC started with pid=15, OS id=8135
Tue Jun 14 18:43:53 2011
Completed: alter database open
因为是clean shutdown，所以这里不存在crash recovery。但这里同样出现了”SMON: enabling cache recovery”，可见cache recovery是每次实例启动instance startup必要执行的一种恢复操作。但问题是，这个恢复操作到底针对何种对象？