MapReduce 能够计算非常复杂的聚合逻辑,非常灵活,但是,MapReduce非常慢,不应该用于实时的数据分析中。MapReduce能够在多台Server上并行执行,每台Server只负责完成一部分wordload,最后将wordload发送到Master Server上合并,计算出最终的结果集,返回客户端。
MapReduce的基本思想,如下图所示:
在这个例子中,我们以一个求和为例。首先执行Map阶段,把一个大任务拆分成若干个小任务,每个小任务运行在不同的节点上,从而支持分布式计算,这个阶段叫做Map(如蓝框所示);每个小任务输出的结果再进行二次计算,最后得到结果55,这个阶段叫做Reduce(如红框所示)。
使用MapReduce方式计算聚合,主要分为三步:Map,Shuffle(拼凑)和Reduce,Map和Reduce需要显式定义,shuffle由MongoDB来实现。
- Map:将操作映射到每个doc,产生Key和Value
- Shuffle:按照Key进行分组,并将key相同的Value组合成数组
- Reduce:把Value数组化简为单值
我们以下面的测试数据(员工数据)为例,来为大家演示。
db.emp.insert( [ {_id:7369,ename:'SMITH' ,job:'CLERK' ,mgr:7902,hiredate:'17-12-80',sal:800,comm:0,deptno:20}, {_id:7499,ename:'ALLEN' ,job:'SALESMAN' ,mgr:7698,hiredate:'20-02-81',sal:1600,comm:300 ,deptno:30}, {_id:7521,ename:'WARD' ,job:'SALESMAN' ,mgr:7698,hiredate:'22-02-81',sal:1250,comm:500 ,deptno:30}, {_id:7566,ename:'JONES' ,job:'MANAGER' ,mgr:7839,hiredate:'02-04-81',sal:2975,comm:0,deptno:20}, {_id:7654,ename:'MARTIN',job:'SALESMAN' ,mgr:7698,hiredate:'28-09-81',sal:1250,comm:1400,deptno:30}, {_id:7698,ename:'BLAKE' ,job:'MANAGER' ,mgr:7839,hiredate:'01-05-81',sal:2850,comm:0,deptno:30}, {_id:7782,ename:'CLARK' ,job:'MANAGER' ,mgr:7839,hiredate:'09-06-81',sal:2450,comm:0,deptno:10}, {_id:7788,ename:'SCOTT' ,job:'ANALYST' ,mgr:7566,hiredate:'19-04-87',sal:3000,comm:0,deptno:20}, {_id:7839,ename:'KING' ,job:'PRESIDENT',mgr:0,hiredate:'17-11-81',sal:5000,comm:0,deptno:10}, {_id:7844,ename:'TURNER',job:'SALESMAN' ,mgr:7698,hiredate:'08-09-81',sal:1500,comm:0,deptno:30}, {_id:7876,ename:'ADAMS' ,job:'CLERK' ,mgr:7788,hiredate:'23-05-87',sal:1100,comm:0,deptno:20}, {_id:7900,ename:'JAMES' ,job:'CLERK' ,mgr:7698,hiredate:'03-12-81',sal:950,comm:0,deptno:30}, {_id:7902,ename:'FORD' ,job:'ANALYST' ,mgr:7566,hiredate:'03-12-81',sal:3000,comm:0,deptno:20}, {_id:7934,ename:'MILLER',job:'CLERK' ,mgr:7782,hiredate:'23-01-82',sal:1300,comm:0,deptno:10} ] );
(案例一)求员工表中,每种职位的人数
var map1=function(){emit(this.job,1)} var reduce1=function(job,count){return Array.sum(count)} db.emp.mapReduce(map1,reduce1,{out:"mrdemo1"})
(案例二)求员工表中,每个部门的工资总和
var map2=function(){emit(this.deptno,this.sal)} var reduce2=function(deptno,sal){return Array.sum(sal)} db.emp.mapReduce(map2,reduce2,{out:"mrdemo2"})
(案例三)Troubleshoot the Map Function
定义自己的emit函数: var emit = function(key, value) { print("emit"); print("key: " + key + " value: " + tojson(value)); } 测试一条数据: emp7839=db.emp.findOne({_id:7839}) map2.apply(emp7839) 输出以下结果: emit key: 10 value: 5000 测试多条数据: var myCursor=db.emp.find() while (myCursor.hasNext()) { var doc = myCursor.next(); print ("document _id= " + tojson(doc._id)); map2.apply(doc); print(); }
(案例四)Troubleshoot the Reduce Function
一个简单的测试案例 var myTestValues = [ 5, 5, 10 ]; var reduce1=function(key,values){return Array.sum(values)} reduce1("mykey",myTestValues) 测试:Reduce的value包含多个值 测试数据:薪水、奖金: var myTestObjects = [ { sal: 1000, comm: 5 }, { sal: 2000, comm: 10 }, { sal: 3000, comm: 15 } ]; 开发reduce方法: var reduce2=function(key,values) { reducedValue = { sal: 0, comm: 0 }; for(var i=0;i<values.length;i++) { reducedValue.sal += values[i].sal; reducedValue.comm += values[i].comm; } return reducedValue; } 测试: reduce2("aa",myTestObjects)