DB2 restart database命令的作用总结
信息中心对于RESTART DATABASE的作用只有一句话的介绍，就是做crash recovery用的。这里总结一下这个命令的作用
信息中心的介绍
Restarts a database that has been abnormally terminated and left in an inconsistent state. At the successful completion of RESTART DATABASE, the application remains connected to the database if the user has CONNECT privilege.
如果数据库当前状态正常(consistent)的话，则只是建立一个新的连接，也不会对其他现有连接产生影响。
indexrec默认设置为RESTART，意思就是发出restart database命令的时候，会自动重建所有的索引（注意，首次连库不会重建索引，除非数据库之前crash了）。
如果系统中有很多的索引失效，那么最好显式地发出restart database命令，以避免等到访问相关表时再创建从而影响性能。
如果数据库Crash之后，某些表空间容器丢失或损坏，则crash recovery会失败。这时候可以使用restart database命令跳过这些表空间的crash recovery，成功连库，连库之后这些表空间会被处于drop pending的状态，只能被删除。
信息中心的介绍
Restarts a database that has been abnormally terminated and left in an inconsistent state. At the successful completion of RESTART DATABASE, the application remains connected to the database if the user has CONNECT privilege.
1. 主要作用是完成正常的crash recovery操作
如果数据库异常终止(inconsistent)，则restart database会做crash recovery，并且做完Crash recovery之后会连接到数据库。如果数据库当前状态正常(consistent)的话，则只是建立一个新的连接，也不会对其他现有连接产生影响。
默认情况下，数据库crash之后，因为autorestart默认设置为ON，连库时会自动做restart db操作，即会自动做crash recovery；如果设置为OFF的话，连库操作由于不会做crash recovery，数据库不一致，会报错SQL1015N，需要显式地发出restart database命令：
$ db2 "update db cfg for sample using autorestart off"
$ db2 "connect to sample"
$ db2 +c "insert into t1 select * from t1" <--这个事务要足够大，以确保日志缓冲区的数据被刷新到磁盘上了，这才样会有crash recovery.
$ db2_kill <--模仿crash
Application ipclean: Removing DB2 engine and client IPC resources for e105q5a
$ db2start
2017-03-14 08:38:58 0 0 SQL1063N DB2START processing was successful.
SQL1063N DB2START processing was successful.
$ db2 "connect to sample" <--没有做Crash recovery，数据库状态不一致
SQL1015N The database is in an inconsistent state. SQLSTATE=55025
2017-03-14-08.43.04.742052+480 E3152A612 LEVEL: Error
PID : 27852862 TID : 3343 PROC : db2sysc 0
INSTANCE: e105q5a NODE : 000 DB : SAMPLE
APPHDL : 0-7 APPID: *LOCAL.e105q5a.170314004304
AUTHID : E105Q5A HOSTNAME: db2b
EDUID : 3343 EDUNAME: db2agent (SAMPLE) 0
FUNCTION: DB2 UDB, base sys utilities, sqledint, probe:2535
MESSAGE : SQL1015N The database is in an inconsistent state.
DATA #1 : String, 91 bytes
Crash Recovery is needed.
Issue RESTART DATABASE on this node before issuing this command.
$ db2 "restart db sample" <--查看诊断日志的话，会发现有做Crash recovery的操作,完成之后，restart db就会成功。
DB20000I The RESTART DATABASE command completed successfully.
2017-03-14-08.44.29.649628+480 E5852A467 LEVEL: Info
PID : 27852862 TID : 3343 PROC : db2sysc 0
INSTANCE: e105q5a NODE : 000 DB : SAMPLE
APPHDL : 0-8 APPID: *LOCAL.e105q5a.170314004429
AUTHID : E105Q5A HOSTNAME: db2b
EDUID : 3343 EDUNAME: db2agent (SAMPLE) 0
FUNCTION: DB2 UDB, base sys utilities, sqledint, probe:5528
MESSAGE : Crash Recovery is needed.
..
2017-03-14-08.44.29.812231+480 E7352A483 LEVEL: Info
PID : 27852862 TID : 3343 PROC : db2sysc 0
INSTANCE: e105q5a NODE : 000 DB : SAMPLE
APPHDL : 0-8 APPID: *LOCAL.e105q5a.170314004429
AUTHID : E105Q5A HOSTNAME: db2b
EDUID : 3343 EDUNAME: db2agent (SAMPLE) 0
FUNCTION: DB2 UDB, recovery manager, sqlpresr, probe:200
MESSAGE : ADM1530I Crash recovery has been initiated.
..
2017-03-14-08.44.30.992167+480 E26963A492 LEVEL: Info
PID : 27852862 TID : 3343 PROC : db2sysc 0
INSTANCE: e105q5a NODE : 000 DB : SAMPLE
APPHDL : 0-8 APPID: *LOCAL.e105q5a.170314004429
AUTHID : E105Q5A HOSTNAME: db2b
EDUID : 3343 EDUNAME: db2agent (SAMPLE) 0
FUNCTION: DB2 UDB, recovery manager, sqlpresr, probe:3100
MESSAGE : ADM1531I Crash recovery has completed successfully.
$ db2 list applications <--restart db会连接到数据库
Auth Id Application Appl. Application Id DB # of
Name Handle Name Agents
-------- -------------- ---------- -------------------------------------------------------------- -------- -----
E105Q5A db2bp 8 *LOCAL.e105q5a.170314004429 SAMPLE 1 2. 重建失效的索引
首先一点，无论何时，只要访问表数据，失效的索引就被会自动重建。还可能有另外一种情形，即发出restart db命令时。indexrec默认设置为RESTART，意思就是发出restart database命令的时候，会自动重建所有的索引（注意，首次连库不会重建索引，除非数据库之前crash了）。
如果系统中有很多的索引失效，那么最好显式地发出restart database命令，以避免等到访问相关表时再创建从而影响性能。
3. 跳过表空间的crash recovery
如果数据库Crash之后，某些表空间容器丢失或损坏，则crash recovery会失败。这时候可以使用restart database命令跳过这些表空间的crash recovery，成功连库，连库之后这些表空间会被处于drop pending的状态，只能被删除。
//首先模式了一个crash操作，之后将表空间的容器删掉，则crash recovery会失败：
$ db2 restart db sample <--restart db会失败，失败的原因是表空间TBS1容器被删，无法做Crash recovery
SQL0290N Table space access is not allowed. SQLSTATE=55039
2017-03-14-11.56.28.645793+480 E138213A633 LEVEL: Info
PID : 37093388 TID : 3343 PROC : db2sysc 0
INSTANCE: e105q5a NODE : 000 DB : SAMPLE
APPHDL : 0-8 APPID: *LOCAL.e105q5a.170314035628
AUTHID : E105Q5A HOSTNAME: db2b
EDUID : 3343 EDUNAME: db2agent (SAMPLE) 0
FUNCTION: DB2 UDB, buffer pool services, sqlbStartPoolsErrorHandling, probe:55
MESSAGE : ADM6023I The table space "TBS1" (ID "6") is in state "0x00000000".
The table space cannot be accessed. Refer to the documentation for
SQLCODE -290.
2017-03-14-11.56.28.646872+480 E138847A1413 LEVEL: Error
PID : 37093388 TID : 3343 PROC : db2sysc 0
INSTANCE: e105q5a NODE : 000 DB : SAMPLE
APPHDL : 0-8 APPID: *LOCAL.e105q5a.170314035628
AUTHID : E105Q5A HOSTNAME: db2b
EDUID : 3343 EDUNAME: db2agent (SAMPLE) 0
FUNCTION: DB2 UDB, buffer pool services, sqlbStartPools, probe:63
MESSAGE : ADM6049E The database cannot be restarted because one or more table
spaces cannot be brought online. To restart the database specify the
"DROP PENDING TABLESPACES" option on the RESTART DATABASE command.
Putting a table space into the drop pending state means that no
further access to the table space will be allowed. The contents of
the table space will be inaccessible throughout the remainder of the
life of the table space; and the only operation that will be allowed
on that table space is "DROP TABLE SPACE". There is no way in which
it can be brought back. It is important that you consider the
consequences of this action as data can be lost as a result. Before
proceeding consult the DB2 documentation and contact IBM support if
necessary. The table spaces to specify in the DROP PENDING
TABLESPACES list are: "TBS1,".
$ db2 "restart db sample drop pending tablespaces (TBS1)" <--跳过表空间TBS1的crash recovery
DB20000I The RESTART DATABASE command completed successfully.
2017-03-14-11.58.42.738412+480 E175431A535 LEVEL: Warning
PID : 37093388 TID : 3343 PROC : db2sysc 0
INSTANCE: e105q5a NODE : 000 DB : SAMPLE
APPHDL : 0-9 APPID: *LOCAL.e105q5a.170314035841
AUTHID : E105Q5A HOSTNAME: db2b
EDUID : 3343 EDUNAME: db2agent (SAMPLE) 0
FUNCTION: DB2 UDB, recovery manager, sqlpresr, probe:2850
MESSAGE : ADM1533W Database has recovered. However, one or more table spaces
are offline.
2017-03-14-11.58.42.739983+480 E175967A492 LEVEL: Info
PID : 37093388 TID : 3343 PROC : db2sysc 0
INSTANCE: e105q5a NODE : 000 DB : SAMPLE
APPHDL : 0-9 APPID: *LOCAL.e105q5a.170314035841
AUTHID : E105Q5A HOSTNAME: db2b
EDUID : 3343 EDUNAME: db2agent (SAMPLE) 0
FUNCTION: DB2 UDB, recovery manager, sqlpresr, probe:3100
MESSAGE : ADM1531I Crash recovery has completed successfully.
$ db2 list tablespaces show detail
Tablespaces for Current Database
..
Tablespace ID = 6
Name = TBS1
Type = Database managed space
Contents = All permanent data. Large table space.
State = 0xc000
Detailed explanation:
Offline
Drop Pending
4. 跳过某个表的crash recovery
同情况3类似，如果某个表上有坏页，crash recovery因为这张表而失败，可以跳过这张表的Crash recovery，成功连库，但该表会被置于drop pending，必须被删除；但这个是隐藏的选项，只有在IBM的支持下才能操作，所以购买IBM的维保服务还是挺重要的。
对于情况3和4，虽然表空间、表处于drop pending的状态，还是可以使用db2dart工具抢救一部分数据的。