TimesTen 数据库复制学习：8. 管理Active Standby Pair（带缓存组）

带缓存组的Active standby pairs(ASP)

在不带缓存组的ASP中，复制发生在TimesTen的表间；而在带缓存组的ASP中，复制发生在cache table之间。带缓冲组的复制仅支持只读和AWT缓存组。对于只读缓存组，复制的意义在于保持状态的连续，而对于AWT，复制可以保证数据不丢失。

设置带只读缓存组的ASP(例)

假设active master为cachedb1，standby master为cachedb2

1. 在active上创建dynamic readonly缓存组

cachedb1>
CREATE DYNAMIC READONLY CACHE GROUP "RO" 
 AUTOREFRESH MODE INCREMENTAL INTERVAL 5 SECONDS
 STATE PAUSED
 FROM
  "TTHR"."A" (
    "ID"   NUMBER(38)        NOT NULL,
    "NAME" VARCHAR2(32 BYTE),
    PRIMARY KEY("ID")
  )

cachedb1> cachegroups;

Cache Group TTHR.RO:

  Cache Group Type: Read Only (Dynamic)
  Autorefresh: Yes
  Autorefresh Mode: Incremental
  Autorefresh State: Paused <- 确保此状态是paused
  Autorefresh Interval: 5 Seconds
  Autorefresh Status: ok
  Aging: LRU on

  Root Table: TTHR.A
  Table Type: Read Only

缓存组的状态必须为paused，否则后续操作会报错，可以使用alter cache group命令修改状态

5166: Autorefresh state ON for TTHR.RO is incompatible with replication store state Idle.  Autorefresh state should be PAUSED. Alter the autorefresh state and reexecute statement.
cachedb1> alter cache group ro set autorefresh state paused;

1. 创建active standby pair，缺省是异步的传输

cachedb1> CREATE ACTIVE STANDBY PAIR cachedb1, cachedb2;

1. 设置active数据库的状态为ACTIVE

cachedb1> call ttRepStateSet('ACTIVE');
cachedb1> call ttRepStateGet;
< ACTIVE, NO GRID >

1. 启动复制代理并动态加载一条数据

cachedb1> call ttrepstart;

cachedb1> select * from a;
cachedb1> select * from a where id = 1;
< 1, beijing >

1. 从active复制出standby, 注意-keepCG选项

cachedb1>create user repadmin identified by timesten;
cachedb1> grant admin to repadmin;

$ ttRepAdmin -duplicate -from cachedb1 -host $(hostname)  -uid repadmin -pwd timesten -keepCG cachedb2
Enter cache administrator UID: cacheadm
Enter cache administrator password: oracle
Waiting for the Duplicate operation to complete ...

1. 登录cachedb2，启动缓存和复制代理，状态自动变为STANDBY

$ ttisql -v1 -e "set prompt 'cachedb2> '" "dsn=cachedb2;uid=tthr;pwd=timesten;oraclepwd=oracle"
cachedb2> call ttCacheStart;
cachedb2> call ttRepStateGet;
< IDLE, NO GRID >
cachedb2> call ttRepStart;
cachedb2> call ttRepStateGet;
< STANDBY, NO GRID >

1. 缓存组的数据已经复制到standby，但standby是只读的，不能运行LOAD操作

cachedb2> select * from a;
< 1, beijing >
cachedb2> select * from a where id = 2; <-只读
 8151: TTHR.RO's replication role disallows the requested operation

cachedb1> select * from a where id = 2;
< 2, shanghai >

cachedb2> select * from a;
< 1, beijing >
< 2, shanghai > <- 可以看到cache group的数据已复制过来
目前一切正常

设置带AWT缓存组的ASP(例)

建立AWT缓存组, 这时active是cachedb2, standby是cachedb1, 整个过程如下：

cachedb2>
CREATE DYNAMIC ASYNCHRONOUS WRITETHROUGH CACHE GROUP "AWT" 
 FROM
  "TTHR"."A" (
    "ID"   NUMBER(38)        NOT NULL,
    "NAME" VARCHAR2(32 BYTE),
    PRIMARY KEY("ID")
  )
  AGING LRU ON
cachedb2> cachegroups;

Cache Group TTHR.AWT:

  Cache Group Type: Asynchronous Writethrough (Dynamic)
  Autorefresh: No
  Aging: LRU on

  Root Table: TTHR.A
  Table Type: Propagate

cachedb2> call ttcachestart;
cachedb2> CREATE ACTIVE STANDBY PAIR cachedb2, cachedb1;
cachedb2> call ttrepstart;
cachedb2> call ttrepstateset('active');
cachedb2> call ttrepstateget;
< ACTIVE, NO GRID >

cachedb2> insert into a values(1, 'beijing');
cachedb2> commit;
cachedb2> select * from a;
< 1, beijing >

$ ttRepAdmin -duplicate -from cachedb2 -host $(hostname)  -uid repadmin -pwd timesten -keepCG cachedb1
Enter cache administrator UID: cacheadm
Enter cache administrator password: 
Waiting for the Duplicate operation to complete ...

$ ttisql -v1 -e "set prompt 'cachedb1> '" "dsn=cachedb1;uid=tthr;pwd=timesten;oraclepwd=oracle"
cachedb1> call ttrepstart;
cachedb1> call ttcachestart;
cachedb1> call ttrepstateget;
< STANDBY, NO GRID >
cachedb1> select * from a;
< 1, beijing >
cachedb1> insert into a values(2, 'shanghai');
16265: This store is currently the STANDBY.  Change to TTHR.A not permitted.
cachedb2> insert into a values(2, 'shanghai');
cachedb1> select * from a;
< 1, beijing >
< 2, shanghai > <- 复制的数据


cachedb1> call ttrepstop;
cachedb1> call ttRepStateSet('ACTIVE');
cachedb1> call ttrepstateget;
< ACTIVE, NO GRID >
cachedb1> call ttRepStateSave('FAILED', 'cachedb2','timesten-hol');
$ ttdestroy cachedb2

$ ttRepAdmin -duplicate -from cachedb1 -host $(hostname)  -uid repadmin -pwd timesten -keepCG -recoveringNode cachedb2
$ ttisql -v1 -e "set prompt 'cachedb2> '" "dsn=cachedb2;uid=tthr;pwd=timesten;oraclepwd=oracle"
cachedb2> call ttrepstart;
cachedb2> call ttcachestart;
calcachedb2> call ttrepstateget;
< STANDBY, NO GRID >
cachedb2> select * from a;
< 1, beijing >
< 2, shanghai >
cachedb2> insert into a values(3, 'guangzhou');
16265: This store is currently the STANDBY.  Change to TTHR.A not permitted.
cachedb1> insert into a values(3, 'guangzhou');
cachedb2> select * from a;
< 1, beijing >
< 2, shanghai >
< 3, guangzhou >

active数据库失效时的恢复

当传输模式为return receipt或异步时

停止复制代理来模拟active失效，以下是缓存组为read-only的情形，如果是AWT，也是类似的

cachedb1> call ttrepstop;

将standby提升为active, 并标记之前的active为失效状态

cachedb2> call ttRepStateSet('ACTIVE');

cachedb2> call ttRepStateSave('FAILED', 'cachedb1','timesten-hol');

这时新的active可以LOAD数据了

cachedb2> select * from a;
< 1, beijing >
< 2, shanghai >
cachedb2> select * from a where id = 3;
< 3, guangzhou >

销毁原来的active，正常的情况下直接用ttdestroy即可

cachedb1> call ttcachestop;
cachedb1> drop active standby pair;
cachedb1> drop cache group ro;
cachedb1> exit
$ ttdestroy cachedb1

从新的active克隆出standby，注意-keepCG -recoveringNode选项

$ ttRepAdmin -duplicate -from cachedb2 -host $(hostname)  -uid repadmin -pwd timesten -keepCG -recoveringNode cachedb1
Enter cache administrator UID: cacheadm
Enter cache administrator password: 
Waiting for the Duplicate operation to complete ...

启动standby的复制和缓存代理

$ ttisql -v1 -e "set prompt 'cachedb1> '" "dsn=cachedb1;uid=tthr;pwd=timesten;oraclepwd=oracle"
cachedb1> call ttrepstart;
cachedb1> call ttcachestart;
cachedb1> call ttrepstateget;
< STANDBY, NO GRID >

数据已复制到standby，但standby的缓存组不能进行LOAD操作

cachedb1> select * from a;
< 1, beijing >
< 2, shanghai >
< 3, guangzhou >
cachedb1> unload cache group ro;
16265: This store is currently the STANDBY.  Change to TTHR.A not permitted.

当传输模式为return twosafe时

这种情况比较简单，因为复制时，数据是先到standby然后在active上提交的。因此很多操作都是自动的。
大不了，从standby升级为active后，再克隆出新的standby即可

standby数据库失效时的恢复

如果复制是return twosafe, 以下是缓存组为read-only的情形

CREATE ACTIVE STANDBY PAIR cachedb1, cachedb2 return twosafe
cachedb2> unload cache group ro;
 8099: TWOSAFE operation not permitted with AutoCommit = 1.
cachedb2> set autocommit 0;
cachedb2> unload cache group ro;
cachedb2> commit;
cachedb2> select * from a;
cachedb2> select * from a where id = 1;
< 1, beijing >
cachedb1> call ttrepstateget;
< STANDBY, NO GRID >
cachedb1> select * from a;
< 1, beijing >
cachedb1> call ttrepstop;
cachedb2> select * from a where id = 2; <- HANG了一会，结果还是出来了，
< 2, shanghai >
cachedb1> select * from a;
< 1, beijing > <- 不过数据自然是不会复制过去的了

如果缓存组为AWT，并且传输模式为return twosafe，这时会阻止active端的事务提交，这时需要设置主端：

call ttRepSyncSet( null, null, 2);
commit;
ttRepStateSave('FAILED','standby_database','host_name')

然后再克隆standby

交换active和standby的角色

假设active为cachedb1，standby为cachedb2
首先停止应用更新数据库。
cachedb1> call ttRepSubscriberWait(NULL, NULL, 'cachedb2', 'timesten-hol', 10);
< 00 > <- 返回值必须为00
call ttRepDeactivate();
cachedb1> call ttrepstop;
cachedb1> call ttRepDeactivate();
cachedb1> call ttrepstateget;
< IDLE, NO GRID >

cachedb2> call ttRepStateSet('ACTIVE');
cachedb2> call ttrepstateget;
< ACTIVE, NO GRID >

cachedb1> call ttrepstart;
cachedb1> call ttrepstateget;
< IDLE, NO GRID >
过一小会
cachedb1> call ttrepstateget;
< STANDBY, NO GRID >
好了，角色换过了了

修改复制的用户名和口令

修改schema user的口令

这个比较简单，如果DDLReplicationLevel 是2，则口令的修改会自动复制到standby。否则就必须手工在每一个库中修改。例如：

cachedb2> alter user tthr identified by oracle;

User altered.

[oracle@timesten-hol ~]$ ttisql -v1 -e "set prompt 'cachedb1> '" "dsn=cachedb1;uid=tthr;pwd=timesten;oraclepwd=oracle"
 7001: User authentication failed
[oracle@timesten-hol ~]$ ttisql -v1 -e "set prompt 'cachedb1> '" "dsn=cachedb1;uid=tthr;pwd=oracle;oraclepwd=oracle"

修改cache admin 用户的口令

这个其实是不常见的。
cache admin用户名在Oracle和TimesTen中必须是一样的，但口令可以不一样
所以改口令没什么特别的，alter user即可。例如

Command> ALTER USER cacheadmin IDENTIFIED BY newpwd;
Command> passthrough 3; 
Command> ALTER USER cacheadm IDENTIFIED BY newpwd;

如果改用户名，就比较麻烦了，必须先删除掉所有的缓存组，因此不建议。

创建灾备用的subscriber

目前为止，谈的都是数据中心内部的HA场景，我们也可以建立一个位于灾备端的subscriber ，在故障时使其成可以继续与Oracle同步数据，第二个Oracle数据库也位于灾备端。

大致过程如下，由于需要建立第二个Oracle数据库，这里就不演示了。

1. 在主点建立AWT缓存组的ASP
2. 建立灾备端的Oracle数据库
3. 在灾备端建立一个subscriber, 这里关键是有一个ttRepAdmin -duplicate的-initCacheDR选项

参考

http://www.oracledistilled.com/linux/configuring-an-oracle-timesten-logical-server-name-on-unix-based-systems/