DBLE和Mycat跨分片查询结果不一致案例分析

背景

　　某一零售业后端使用了分布式中间件+MySQL数据库作为后端存储。但是因为历史问题存在两种分布式中间件，分别是Mycat和DBLE，共用一组后端MySQL实例。分片规则以及后端数据完全一致。最近碰到了一个比较有意思的场景，财务结算单来往明细和业务来往单据的关联查询。一条跨节点join查询在DBLE、Mycat的查询得到的结果不一致。究竟谁对谁错？

环境准备

　　在虚拟机搭建类似架构，模拟场景，比较Mycat-DBLE在跨节点join上的异同点。

测试架构

测试环境架构比较简单，DBLE与Mycat公用数据库。

测试软件版本

表结构

结算单来往明细表

CREATE TABLE `t_bl_detail` (
  `unit_num` int(11) DEFAULT NULL,
  `tenantid` int(11) DEFAULT NULL,
  `detail_num` int(11) DEFAULT NULL,
  `balance_date` datetime DEFAULT NULL
) ENGINE=InnoDB DEFAULT CHARSET=utf8;

业务来往单据表

CREATE TABLE `t_bl_super_detail` (
  `unit_num` int(11) DEFAULT NULL,
  `sup_id` int(11) DEFAULT NULL,
  `tenantid` int(11) DEFAULT NULL,
  `bl_unit_num` int(11) DEFAULT NULL
) ENGINE=InnoDB DEFAULT CHARSET=utf8;

分片规则配置

配置表t_bl_detail、t_bl_super_detail，使用取模算法，数据分布在db1-4四个database中。

schema配置

DBLE：

 <schema name="testdb" sqlMaxLimit="100" dataNode="dn01">
        <table name="t_bl_detail" rule="mod4Series" dataNode="dn01,dn02,dn03,dn04"></table>
        <table name="t_bl_super_detail" rule="mod4Series" dataNode="dn01,dn02,dn03,dn04"></table>
    </schema>
    <dataNode name="dn01" dataHost="group1" database="db1"></dataNode>
    <dataNode name="dn02" dataHost="group2" database="db2"></dataNode>
    <dataNode name="dn03" dataHost="group1" database="db3"></dataNode>
    <dataNode name="dn04" dataHost="group2" database="db4"></dataNode>
    <dataHost name="group1" maxCon="100" minCon="5" balance="0" switchType="-1">
        <heartbeat>select user()</heartbeat>
        <writeHost host="mysql-w9zhkr" url="1.1.1.26:3306" user="ushard" password="VRDkwsUlq++O3HeamsDcuW/2K22si3RIhcTpAdjalbkiinNAeUUDWk11ttls1Z3PXY5TxGJToeK4LsJE3+k0Wg==" id="mysql-w9zhkr" usingDecrypt="1"></writeHost>
    </dataHost>
    <dataHost name="group2" maxCon="100" minCon="5" balance="0" switchType="-1">
        <heartbeat>select user()</heartbeat>
        <writeHost host="mysql-ci2va0" url="1.1.1.27:3306" user="ushard" password="O/iCi0F+9ts5YR78ZcQnKzpb5GqG7xSUzJcO44yZ3BhNkT5E7aiZakmz1tbKIQylnqh20vu5Z9egXBUC3GOKow==" id="mysql-ci2va0" usingDecrypt="1"></writeHost>
    </dataHost>

Mycat：

<schema name="testdb" sqlMaxLimit="100" dataNode="dn01">
        <table name="t1" rule="mod4Series" dataNode="dn01,dn02,dn03,dn04"></table>
        <table name="t2" rule="mod4Series" dataNode="dn01,dn02,dn03,dn04"></table>
    </schema>
    <dataNode name="dn01" dataHost="group1" database="db1"></dataNode>
    <dataNode name="dn02" dataHost="group2" database="db2"></dataNode>
    <dataNode name="dn03" dataHost="group1" database="db3"></dataNode>
    <dataNode name="dn04" dataHost="group2" database="db4"></dataNode>
    <dataHost name="group1" maxCon="100" minCon="5" balance="0" switchType="-1" dbType="mysql" dbDriver="native">
        <heartbeat>select user()</heartbeat>
        <writeHost host="mysql-w9zhkr" url="1.1.1.26:3306" user="ushard" password="ushard"></writeHost>
    </dataHost>
    <dataHost name="group2" maxCon="100" minCon="5" balance="0" switchType="-1" dbType="mysql" dbDriver="native">
        <heartbeat>select user()</heartbeat>
        <writeHost host="mysql-ci2va0" url="1.1.1.27:3306" user="ushard" password="ushard"></writeHost>
    </dataHost>

rule配置

DBLE：

<tableRule name="mod4Series">
        <rule>
            <columns>unit_num</columns>
            <algorithm>mod4DB</algorithm>
        </rule>
    </tableRule>
    <function name="mod4DB" class="Hash">
        <property name="partitionCount">4</property>
        <property name="partitionLength">1</property>
    </function>

Mycat：

<tableRule name="mod4Series">
        <rule>
            <columns>id</columns>
            <algorithm>mod4DB</algorithm>
        </rule>
    </tableRule>
    <function name="mod4DB" class="io.mycat.route.function.PartitionByMod">
        <property name="count">4</property>
    </function>

比对开始

准备测试数据

随便登录哪一台中间件写入测试数据

insert into t_bl_detail values(1,3,123443,'2019-01-01 00:00:00');
insert into t_bl_detail values(2,3,3423524,'2019-01-01 00:00:00');
insert into t_bl_detail values(3,3,245245,'2019-01-01 00:00:00');
insert into t_bl_detail values(4,4,356356,'2019-01-01 00:00:00');


insert into t_bl_super_detail values(1,10342,3,2);
insert into t_bl_super_detail values(2,12355,3,2);
insert into t_bl_super_detail values(3,62542,3,3);
insert into t_bl_super_detail values(4,74235,4,1);

执行跨节点join查询：

select
 m.unit_num,
 n.sup_id,
 n.tenantid,
 m.detail_num,
 n.bl_unit_num,
 m.balance_date
from
 t_bl_detail m
join t_bl_super_detail n on
 m.tenantid = n.tenantid
where
 n.tenantid = 3
 and m.unit_num = n.bl_unit_num;

在通过中间件之前，先在MySQL中执行一遍看下结果，作为预期结果供后续案例使用。

分别通过DBLE、mycat执行跨节点join语句

　　可看到相同的查询语句，DBLE执行结果符合预期，Mycat执行结果缺失。数据差异在于DBLE查询结果相较于Mycat多了跨节点的结果。虽然Mycat执行跨节点join不报错，但是查询结果却和预期不一致。

执行计划

只从结果上判断并没有办法知道是什么原因导致了Mycat结果缺失，查看查询计划，比较两者差异。

DBLE

通过DBLE的执行计划可看出，DBLE内部分别对结算明细表、业务单据表做了各自的数据查询，将查询结果在中间层做了merge。最后获得跨节点join的结果

• DBLE的执行计划

mysql> explain select m.unit_num,n.sup_id,n.tenantid,m.detail_num,n.bl_unit_num,m.balance_date from t_bl_detail m join t_bl_super_detail n on m.tenantid=n.tenantid where n.tenantid = 3 and m.unit_num=n.bl_unit_num;
+-----------------+---------------+----------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| DATA_NODE       | TYPE          | SQL/REF                                                                                                                                                              |
+-----------------+---------------+----------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| dn01_0          | BASE SQL      | select `m`.`unit_num`,`m`.`detail_num`,`m`.`balance_date`,`m`.`tenantid` from  `t_bl_detail` `m` where m.tenantid = 3 ORDER BY `m`.`tenantid` ASC,`m`.`unit_num` ASC |
| dn02_0          | BASE SQL      | select `m`.`unit_num`,`m`.`detail_num`,`m`.`balance_date`,`m`.`tenantid` from  `t_bl_detail` `m` where m.tenantid = 3 ORDER BY `m`.`tenantid` ASC,`m`.`unit_num` ASC |
| dn03_0          | BASE SQL      | select `m`.`unit_num`,`m`.`detail_num`,`m`.`balance_date`,`m`.`tenantid` from  `t_bl_detail` `m` where m.tenantid = 3 ORDER BY `m`.`tenantid` ASC,`m`.`unit_num` ASC |
| dn04_0          | BASE SQL      | select `m`.`unit_num`,`m`.`detail_num`,`m`.`balance_date`,`m`.`tenantid` from  `t_bl_detail` `m` where m.tenantid = 3 ORDER BY `m`.`tenantid` ASC,`m`.`unit_num` ASC |
| merge_1         | MERGE         | dn01_0; dn02_0; dn03_0; dn04_0                                                                                                                                       |
| shuffle_field_1 | SHUFFLE_FIELD | merge_1                                                                                                                                                              |
| dn01_1          | BASE SQL      | select `n`.`sup_id`,`n`.`tenantid`,`n`.`bl_unit_num` from  `t_bl_super_detail` `n` where n.tenantid = 3 ORDER BY `n`.`tenantid` ASC,`n`.`bl_unit_num` ASC            |
| dn02_1          | BASE SQL      | select `n`.`sup_id`,`n`.`tenantid`,`n`.`bl_unit_num` from  `t_bl_super_detail` `n` where n.tenantid = 3 ORDER BY `n`.`tenantid` ASC,`n`.`bl_unit_num` ASC            |
| dn03_1          | BASE SQL      | select `n`.`sup_id`,`n`.`tenantid`,`n`.`bl_unit_num` from  `t_bl_super_detail` `n` where n.tenantid = 3 ORDER BY `n`.`tenantid` ASC,`n`.`bl_unit_num` ASC            |
| dn04_1          | BASE SQL      | select `n`.`sup_id`,`n`.`tenantid`,`n`.`bl_unit_num` from  `t_bl_super_detail` `n` where n.tenantid = 3 ORDER BY `n`.`tenantid` ASC,`n`.`bl_unit_num` ASC            |
| merge_2         | MERGE         | dn01_1; dn02_1; dn03_1; dn04_1                                                                                                                                       |
| shuffle_field_3 | SHUFFLE_FIELD | merge_2                                                                                                                                                              |
| join_1          | JOIN          | shuffle_field_1; shuffle_field_3                                                                                                                                     |
| shuffle_field_2 | SHUFFLE_FIELD | join_1                                                                                                                                                               |
+-----------------+---------------+----------------------------------------------------------------------------------------------------------------------------------------------------------------------+
14 rows in set (0.00 sec)

Mycat

Mycat对于跨结点join的处理则相对暴力，直接将查询语句下发到各个节点，最后将结果进行汇总，如果表连接涉及到跨节点。则跨节点的数据无法进行join。

• Mycat的执行计划

mysql> explain select m.unit_num,n.sup_id,n.tenantid,m.detail_num,n.bl_unit_num,m.balance_date from t_bl_detail m join t_bl_super_detail n on m.tenantid=n.tenantid where n.tenantid = 3 and m.unit_num=n.bl_unit_num;
+-----------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| DATA_NODE | SQL                                                                                                                                                                                                    |
+-----------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| dn01      | select m.unit_num,n.sup_id,n.tenantid,m.detail_num,n.bl_unit_num,m.balance_date from t_bl_detail m join t_bl_super_detail n on m.tenantid=n.tenantid where n.tenantid = 3 and m.unit_num=n.bl_unit_num |
| dn02      | select m.unit_num,n.sup_id,n.tenantid,m.detail_num,n.bl_unit_num,m.balance_date from t_bl_detail m join t_bl_super_detail n on m.tenantid=n.tenantid where n.tenantid = 3 and m.unit_num=n.bl_unit_num |
| dn03      | select m.unit_num,n.sup_id,n.tenantid,m.detail_num,n.bl_unit_num,m.balance_date from t_bl_detail m join t_bl_super_detail n on m.tenantid=n.tenantid where n.tenantid = 3 and m.unit_num=n.bl_unit_num |
| dn04      | select m.unit_num,n.sup_id,n.tenantid,m.detail_num,n.bl_unit_num,m.balance_date from t_bl_detail m join t_bl_super_detail n on m.tenantid=n.tenantid where n.tenantid = 3 and m.unit_num=n.bl_unit_num |
+-----------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
4 rows in set (0.00 sec)

总结

Mycat是一款非常优秀的分布式中间件，但是在某些细节方面处理的不尽人意。在跨节点关联查询场景下，Mycat采取的策略是直接将语句透传到各个节点上，将获取到的结果整合后返回，得到的结果集合预期结果有出入，缺失了跨节点关联的数据。

DBLE处理跨节点的关联查询是先获取到关联需要的数据，提取到中间件进行融合，得到关联的结果并返回。得到的结果集符合预期，与MySQL执行结果一致。可见DBLE在跨节点关联查询方面做了优化，能够提供准确的查询结果。