1.E-R关系策略的由来
join是关系数据库最常用的一个特性，然而在分布式环境中，跨分片的join最复杂，最难解决。
这是官方文档的描述。
具体点，比如：
mycat逻辑库hello，两张表格t1,t2。做了分库策略，t1放到了datanode1，t2放到了datanode2。如果我t1 join t2检索数据，
怎么办？
这就是E-R关系策略要解决的问题。
mycat借鉴了table group的概念，将子表的存储位置依赖于子表，并且在物理上紧邻存放，解决了join的效率和性能问题。E-R关系的数据分片策略，根据这一思路，将子表的记录和所关联的父表记录存放在同一个数据分片上。
2.测试官方教程文档上的E-R关系表
customer采用sharding-by-intfile（分片枚举）策略，分片在dn1,dn2上，orders依赖父表进行分片，两个表的关联关系为orders.customer_id=customer.id。示意图如下：
<table name="customer" primaryKey="ID" dataNode="dn1,dn2"
rule="sharding-by-intfile">
<childTable name="orders" joinKey="customer_id" parentKey="id"/>
</table>
解释：
<table name="customer" primaryKey="ID" dataNode="dn1,dn2"
rule="sharding-by-intfile">
这一行是定义customer表，主键是id,分片部署在dn1,dn2,分片规则是sharding-by-intfile
<childTable name="orders" joinKey="customer_id" parentKey="id"/>
这一行是定义orders是childtable。
childtable是依赖父表的结构，就是前面时候的E-R关系的表。
childtable的joinkey会按照父表的parentkey一起切分。
</table>
这是对应 <table name= 的结束格式，参考xml格式。
3.
表格设计：
customer表
id(primarykey) name city （用city做分片）
orders表
customer_id(primary key) orders
两表格关系：
customer表的主键id为orders表主键customer_id的外键
4.mycat上实际测试：
停止mycat服务，修改配置文件，如下：
[[email protected] conf]# cat schema.xml
<?xml version="1.0"?>
<!DOCTYPE mycat:schema SYSTEM "schema.dtd">
<mycat:schema xmlns:mycat="http://io.mycat/">
<schema name="hello" checkSQLschema="false" sqlMaxLimit="100">
<!-- auto sharding by id (long) -->
<table name="t1" dataNode="dn1,dn2" rule="sharding-by-intfile" />
<!-- global table is auto cloned to all defined data nodes ,so can join
with any table whose sharding node is in the same data node -->
<table name="t2" primaryKey="ID" type="global" dataNode="dn1,dn2" />
<table name="t3" dataNode="dn1" />
<table name="t4" dataNode="dn2" />
<table name="customer" primaryKey="id" dataNode="dn1,dn2"
rule="sharding-by-intfile">
<childTable name="orders" joinKey="customer_id" parentKey="id"/>
</table>
</schema>
<!-- <dataNode name="dn1$0-743" dataHost="localhost1" database="db$0-743"
/> -->
<dataNode name="dn1" dataHost="mysql1" database="db1" />
<dataNode name="dn2" dataHost="mysql3" database="db2" />
<dataHost name="mysql1" maxCon="1000" minCon="10" balance="3"
writeType="0" dbType="mysql" dbDriver="native" switchType="1" slaveThreshold="100">
<heartbeat>select user()</heartbeat>
<!-- can have multi write hosts -->
<writeHost host="hostM1" url="192.168.211.138:3306" user="root"
password="[email protected]">
</writeHost>
<writeHost host="hostS1" url="192.168.211.139:3306" user="root"
password="[email protected]">
</writeHost>
</dataHost>
<dataHost name="mysql3" maxCon="1000" minCon="10" balance="0"
writeType="0" dbType="mysql" dbDriver="native" switchType="1" slaveThreshold="100">
<heartbeat>select user()</heartbeat>
<!-- can have multi write hosts -->
<writeHost host="hostM1" url="192.168.211.142:3306" user="root"
password="[email protected]">
<!-- can have multi read hosts -->
<readHost host="hostS2" url="192.168.211.142:3306" user="root" password="[email protected]"/>
</writeHost>
</dataHost>
</mycat:schema>
定义分片规则：
<tableRule name="sharding-by-intfile">
<rule>
<columns>city</columns>
<algorithm>hash-int</algorithm>
</rule>
</tableRule>
<function name="hash-int"
class="io.mycat.route.function.PartitionByFileMap">
<property name="mapFile">partition-hash-int.txt</property>
<property name="type">1</property>
<property name="defaultNode">0</property>
</function>
[[email protected] conf]# cat partition-hash-int.txt
gz=0
sz=1
启动mycat，创建表格：
mysql> create table customer(id int not null primary key,name varchar(10),city varchar(20));
Query OK, 0 rows affected (0.11 sec)
mysql> create table orders (customer_id int not null primary key,orders int not null,foreign key(customer_id) references customer(id)
on delete cascade on update cascade);
Query OK, 0 rows affected (0.25 sec)
customer插入数据测试：
mysql> insert into customer(id,name,city) values(1,'am1','gz'),(2,'am2','gz'),(3,'am3','sz');
mysql> select * from customer where city='gz';
+----+------+------+
| id | name | city |
+----+------+------+
| 1 | am1 | gz |
| 2 | am2 | gz |
+----+------+------+
2 rows in set (0.08 sec)
mysql> explain select * from customer where city='gz';
+-----------+----------------------------------------------------+
| DATA_NODE | SQL |
+-----------+----------------------------------------------------+
| dn1 | SELECT * FROM customer WHERE city = 'gz' LIMIT 100 |
+-----------+----------------------------------------------------+
1 row in set (0.01 sec)
gz的数据都在dn1实现了分片。
orders插入数据测试：
mysql> insert into orders(customer_id,orders) values(1,10001);
Query OK, 1 row affected (0.33 sec)
mysql> insert into orders(customer_id,orders) values(2,10002);
Query OK, 1 row affected (0.29 sec)
mysql> insert into orders(customer_id,orders) values(3,10003);
Query OK, 1 row affected (0.48 sec)
根据E-R分片规则，orders表格根据外键的值也就是customer的主键值切分，
也就是orders.customer_id=customer.id的数据分在一个区。
分别在db1,db2检索数据，看看是否达到E-R分片的设计要求。
原文地址：http://blog.51cto.com/goome/2058958