某一零售業後端使用了分佈式中間件+MySQL數據庫做爲後端存儲。可是由於歷史問題存在兩種分佈式中間件,分別是Mycat和DBLE,共用一組後端MySQL實例。分片規則以及後端數據徹底一致。最近碰到了一個比較有意思的場景,財務結算單來往明細和業務來往單據的關聯查詢。一條跨節點join查詢在DBLE、Mycat的查詢獲得的結果不一致。究竟誰對誰錯?mysql
DBLE官方網站: https://opensource.actionsky.com 能夠詳細瞭解DBLE的背景和應用場景,本文不涉及到的細節均可在官方文檔得到更細節都信息;對於剛瞭解到同窗,能夠以本文爲快速入門基礎git
DBLE 官方項目: https://github.com/actiontech/dble 如對源碼有興趣或者須要定製的功能的能夠經過源碼編譯安裝github
DBLE 下載地址: https://github.com/actiontech/dble/releases算法
DBLE 官方社區交流:669663113sql
在虛擬機搭建相似架構,模擬場景,比較Mycat-DBLE在跨節點join上的異同點。數據庫
測試環境架構比較簡單,DBLE與Mycat共用數據庫。後端
軟件名稱 | 軟件版本 | 端口 | 管理端口 |
---|---|---|---|
DBLE | DBLE-2.18.10.1-cb392c3-20181106093917 | 3309 | 3310 |
Mycat | Mycat-1.6-RELEASE-20161028204710 | 8066 | 9066 |
MySQL | 5.7.21-log | 3306 | 無 |
結算單來往明細表架構
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;
業務來往單據表分佈式
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;
配置表t_bl_detail、t_bl_super_detail,使用取模算法,數據分佈在db1-db4四個database中。測試
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>
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>
隨便登陸哪一臺中間件寫入測試數據
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);
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;
在經過中間件以前,如今MySQL中執行一遍看下結果,做爲預期結果供後續案例使用。
分別經過DBLE、mycat執行跨節點join語句
可看到相同的查詢語句,DBLE執行結果符合預期,Mycat執行結果缺失。數據差別在於DBLE查詢結果相較於Mycat多了跨節點的結果。雖然Mycat執行跨節點join不報錯,可是查詢結果卻和預期不一致。
只從結果上判斷並無辦法知道是什麼緣由致使了Mycat結果缺失,查看查詢計劃,比較二者差別。
經過DBLE的執行計劃可看出,DBLE內部分別對結算明細表、業務單據表作了各自的數據查詢,將查詢結果在中間層作了merge。最後得到跨節點join的結果
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對於跨節點join的處理則相對暴力,直接將查詢語句下發到各個節點,最後將結果進行彙總,若是錶鏈接涉及到跨節點。則跨節點的數據沒法進行join。
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)
Mycat是一款很是優秀的分佈式中間件,可是在某些細節方面處理的不盡人意。在跨節點關聯查詢場景下,Mycat採起的策略是直接將語句透傳到各個節點上,將獲取到的結果整合後返回,獲得的結果集和預期結果有出入,缺失了跨節點關聯的數據。
DBLE處理跨節點的關聯查詢是先獲取到關聯須要的數據,提取到中間件進行融合,獲得關聯查詢的結果並返回。獲得的結果集符合預期,與MySQL執行結果一致。可見DBLE在跨節點關聯查詢方面作了優化,可以提供準確的查詢結果。