TimesTen 应用层数据库缓存学习：18. 利用TimesTen实现Sharding或数据分区

考虑到恢复和磁盘加载到内存的速度，单个TimesTen数据库可以缓存的数据一般不超过100G，如果需要缓存的数据较大，一般可以利用多个TimesTen来实现数据分区或分片(Sharding)。

虽然TimesTen有Cache Grid即Global Cache Group的概念，但由于其还不太成熟，一般采用手工分区的方式，即通过应用服务器定向到指定的TimesTen的方式。

以下给出了只读和读写数据分区的例子。

另外强调一点，TimesTen没有Oracle Partition的概念。

TimesTen can cache Oracle Database partitioned tables at the table level, but individual partitions cannot be cached. The following describes how operations on partitioned tables affect cache groups:

DDL operations on a table that has partitions do not affect the cache group unless there is data loss. For example, if a partition with data is truncated, an AUTOREFRESH operation does not delete the data from the corresponding cached table.

WHERE clauses in any cache group operations cannot reference individual partitions or subpartitions. Any attempt to define a single partition of a table returns an error.

只读分区

利用只读分区可以指定where条件的便利，可以在不同的TimesTen中建立带where条件的cache group，来缓存Oracle数据库不同部分的数据。

以下的例子虽然Oracle中的表具有分区，但只是示例而已，即使没有分区结果也是一样的。

在Oracle中建表，分为东西南北四个区

create table datapart
( region varchar2(1),
id int,
city varchar2(20),
primary key(id)
)
partition by list(region)
(
partition part_1 values ( 'N' ),
partition part_2 values ( 'S' ),
partition part_3 values ( 'W' ),
partition part_4 values ( 'E' )
);

insert into datapart values ('N', 1, 'Beijing');
insert into datapart values ('S', 2, 'Guangzhou');
insert into datapart values ('E', 3, 'Shanghai');
insert into datapart values ('W', 4, 'Chengdu');

SQL> select * from datapart partition(part_2);

R         ID CITY
- ---------- --------------------
S          2 Guangzhou

grant select, delete, update, insert on datapart to cacheadm;

分别在两个TimesTen数据库cachedb1和cachedb2中建立Cache Group，缓存北区和南区的数据

cachedb1>
CREATE READONLY CACHE GROUP "DATAPART1"
AUTOREFRESH MODE INCREMENTAL INTERVAL 5 SECONDS
STATE ON
FROM
"TTHR"."DATAPART" (
"REGION" VARCHAR2(1 BYTE) ,
"ID"     NUMBER(38)        NOT NULL,
"CITY"   VARCHAR2(20 BYTE),
PRIMARY KEY("ID")
)
WHERE (REGION='N')

cachedb1> select * from datapart;
< N, Beijing >

cachedb2>
CREATE READONLY CACHE GROUP "DATAPART2"
AUTOREFRESH MODE INCREMENTAL INTERVAL 5 SECONDS
STATE ON
FROM
"TTHR"."DATAPART" (
"REGION" VARCHAR2(1 BYTE) ,
"ID"     NUMBER(38)        NOT NULL,
"CITY"   VARCHAR2(20 BYTE),
PRIMARY KEY("ID")
)
WHERE (REGION='S')

cachedb2> select * from datapart;
< S, Guangzhou >

在Oracle中插入数据

insert into datapart values ('N', 5, 'Qingdao');
insert into datapart values ('S', 6, 'Haikou');

然后在两个TimesTen中可自动得到新数据

cachedb1> select * from datapart;
< N, 1, Beijing >
< N, 5, Qingdao >

cachedb2> select * from datapart;
< S, 2, Guangzhou >
< S, 6, Haikou >

可写分区

可写分区以AWT为例，和只读分区不一样，AWT在建立时不能指定where条件，但是可以在LOAD操作时指定where条件来进行数据分区

在cachedb1中建立Cache Group, LOAD时指定缓存北区的数据:

CREATE ASYNCHRONOUS WRITETHROUGH CACHE GROUP "AWT1"
FROM
"TTHR"."DATAPART" (
"REGION" VARCHAR2(1 BYTE) ,
"ID"     NUMBER(38)        NOT NULL,
"CITY"   VARCHAR2(20 BYTE),
PRIMARY KEY("ID")
)

cachedb1> call ttrepstart;

cachedb1> load cache group awt1 where region = 'N' commit every 256 rows;
cachedb1> select * from datapart;
< N, 1, Beijing >
< N, 5, Qingdao >

cachedb1> truncate table datapart;
8238: Cannot truncate cache group table DATAPART; please use DROP/REFRESH/UNLOAD CACHE GROUP instead

insert into datapart values ('N', 7, 'Tianjing');

在cachedb2中建立Cache Group, LOAD时指定缓存南区的数据:

CREATE ASYNCHRONOUS WRITETHROUGH CACHE GROUP "AWT2"
FROM
"TTHR"."DATAPART" (
"REGION" VARCHAR2(1 BYTE) ,
"ID"     NUMBER(38)        NOT NULL,
"CITY"   VARCHAR2(20 BYTE),
PRIMARY KEY("ID")
)

cachedb2> call ttrepstart;

cachedb2> load cache group awt2 where region = 'S' commit every 256 rows;
cachedb2> select * from datapart;
< S, 2, Guangzhou >
< S, 6, Haikou >
cachedb2>

insert into datapart values ('S', 8, 'Changsha');

在Oracle中可以看到由各个TimesTen新插入的数据

SQL> select * from datapart order by id;

R         ID CITY
- ---------- --------------------
N          1 Beijing
S          2 Guangzhou
E          3 Shanghai
W          4 Chengdu
N          5 Qingdao
S          6 Haikou
N          7 Tianjing
S          8 Changsha

8 rows selected.

在Oracle中插入数据，然后各个缓存组通过LOAD/REFRESH操作得到新数据。

注意，对于Explicitly load的缓存组，LOAD和REFRESH的作用是一样的。

SQL>
insert into datapart values ('N', 9, 'Baoding');
insert into datapart values ('S', 10, 'Wuhan');
commit;

cachedb1> select * from datapart;
< N, 1, Beijing >
< N, 5, Qingdao >
< N, 7, Tianjing >

cachedb1> refresh cache group awt1 where region = 'N' commit every 256 rows;
cachedb1> select * from datapart;
< N, 1, Beijing >
< N, 5, Qingdao >
< N, 7, Tianjing >
< N, 9, Baoding >

cachedb2> select * from datapart;
< S, 2, Guangzhou >
< S, 6, Haikou >
< S, 8, Changsha >

cachedb2> load cache group awt2 where region = 'S' commit every 256 rows;

cachedb2> select * from datapart;
< S, 2, Guangzhou >
< S, 6, Haikou >
< S, 8, Changsha >
< S, 10, Wuhan >

参考

https://community.oracle.com/thread/3594994?start=0&tstart=0