MapReduce实现手机上网日志分析(排序)
一、背景
1.1 流程
实现排序,分组拍上一篇通过Partitioner实现了。
实现接口,自动产生接口方法,写属性,产生getter和setter,序列化和反序列化属性,写比较方法,重写toString,为了方便复制写够着方法,不过重写够着方法map里需要不停地new,发现LongWritable有set方法,text也有,可以用,产生默认够着方法。
public void set(String account,double income,double expense,double surplus) { this.account = account; this.income = income; this.expense = expense; this.surplus = income-expense; }
1.2 数据集
为了和上一篇保在知识上持递进,数据及换了,名字没变。
下面是输出结果,其实mr也会自动排序,不过string按字典序排序了。
二、理论知识
字符串拼接,记得以前自己写过,现在拿出来看看,http://www.cnblogs.com/hxsyl/archive/2012/10/18/2729112.html
简单总结扩展如下:String是final的,不能改变也不能继承,因此在每次对 String 类型进行改变的时候其实都等同于生成了一个新的 String 对象,然后将指针指向新的 String 对象,所以经常改变内容的字符串最好不要用 String ,因为每次生成对象都会对系统性能产生影响,特别当内存中无引用对象多了以后, JVM 的 GC 就会开始工作,那速度是一定会相当慢的。
如果for循环1w次,这句 string += "hello";的过程相当于将原有的string变量指向的对象内容取出与"hello"作字符串相加操作再存进另一个新的String对象当中,再让string变量指向新生成的对象。反编译出的字节码文件可以很清楚地看出,每次循环会new出一个StringBuilder对象,然后进行append操作,最后通过toString方法返回String对象。也就是说这个循环执行完毕new出了10000个对象,试想一下,如果这些对象没有被回收,内存浪费不说,有可能重复使用赵成系统卡死。从上面还可以看出:string+="hello"的操作事实上会自动被JVM优化成:
StringBuilder str = new StringBuilder(string);
str.append("hello");
str.toString();
如果直接for循环里StringBuilder 的话会只是new一次。效率高。
而StringBuffer是线程安全的,多了synchronized关键字,也就是在多线程下会顺序读取换冲刺。
参考了这个http://blog.csdn.net/loveyaozu/article/details/47037957
三、实体类
收入相同的话按消费从低到高,否则收入从高到低。
package cn.app.hadoop.mr.sort; import java.io.DataInput; import java.io.DataOutput; import java.io.IOException; import java.math.BigDecimal; import org.apache.hadoop.io.WritableComparable; import org.apache.jasper.tagplugins.jstl.core.Out; //Writable是序列化接口 //泛型是InfoBean,就像比较学生信息一样,成绩,性别等 ,封装在了一个bean里 //不过发现WritableComparable 有了序列化和反序列化 public class InfoBean implements WritableComparable<InfoBean>{ private String account; //金钱类都需要BigDecimal,double顺势精度,不过不知道下边序列化咋写类型,所以先用double,估计writeUTF可以 private double income; private double expense; private double surplus; public String getAccount() { return account; } public void setAccount(String account) { this.account = account; } public double getIncome() { return income; } public void setIncome(double income) { this.income = income; } public double getExpense() { return expense; } public void setExpense(double expense) { this.expense = expense; } public double getSurplus() { return surplus; } public void setSurplus(double surplus) { this.surplus = surplus; } public void readFields(DataInput in) throws IOException { // TODO Auto-generated method stub this.account = in.readUTF(); this.income = in.readDouble(); this.expense = in.readDouble(); this.surplus = in.readDouble(); } public void write(DataOutput out) throws IOException { // TODO Auto-generated method stub out.writeUTF(account); out.writeDouble(income); out.writeDouble(expense); out.writeDouble(surplus); } public void set(String account,double income,double expense) { this.account = account; this.income = income; this.expense = expense; this.surplus = income - expense; } public InfoBean() { super(); // TODO Auto-generated constructor stub } @Override public String toString() { return "InfoBean [income=" + income + ", expense=" + expense + ", surplus=" + surplus + "]"; } public int compareTo(InfoBean o) { // TODO Auto-generated method stub if(this.income == o.getIncome()) { return this.expense>o.getExpense()?1:-1; }else { return this.income>o.getIncome()?-1:1; } } }
四、第一种实现
4.1 Mapper
//第一个处理文本的话一般是LongWritable 或者object //一行一行的文本是text //输出的key的手机号 定位Text //结果是DataBean 一定要实现Writable接口 public class InfoSortMapper extends Mapper<LongWritable, Text, Text, InfoBean> { private InfoBean v = new InfoBean(); private Text k = new Text(); public void map(LongWritable key, Text value, Context context) throws IOException, InterruptedException { String line = value.toString(); String[] fields = line.split("\t"); String account = fields[0]; double in = Double.parseDouble(fields[1]); double out = Double.parseDouble(fields[2]); //不用每次new 几遍不重写内存引用,也很站用资源 k.set(account); v.set(account, in, out); context.write(k, v); }
4.2 Reducer
public class InfoSortReducer extends Reducer<Text, InfoBean, Text, InfoBean> { //k就是key,不需要 private InfoBean v = new InfoBean(); public void reduce(Text key, Iterable<InfoBean> value, Context context) throws IOException, InterruptedException { // process values double incomeSum = 0; double expenseSum = 0; for (InfoBean o : value) { incomeSum += o.getIncome(); expenseSum += o.getExpense(); } v.set(key.toString(), incomeSum, expenseSum); //databean会自动调用toString context.write(key,v); } }
五、第二种实现
5.1 Mapper
//对 InfoBean 排序 k2就是他 public class SortMapper extends Mapper<LongWritable, Text, InfoBean, NullWritable> { private InfoBean k = new InfoBean(); public void map(LongWritable key, Text value, Context context) throws IOException, InterruptedException { String line = value.toString(); String[] fields = line.split("\t"); String account = fields[0]; double in = Double.parseDouble(fields[1]); double out = Double.parseDouble(fields[2]); //不用每次new 几遍不重写内存引用,也很站用资源 k.set(account, in, out); //value必须是NullWritable.get(),NullWritable不行,提示不是变量 context.write(k, NullWritable.get()); } }
5.2 Reducer
//对 InfoBean 排序 k2就是他 public class SortMapper extends Mapper<LongWritable, Text, InfoBean, NullWritable> { private InfoBean k = new InfoBean(); public void map(LongWritable key, Text value, Context context) throws IOException, InterruptedException { String line = value.toString(); String[] fields = line.split("\t"); String account = fields[0]; double in = Double.parseDouble(fields[1]); double out = Double.parseDouble(fields[2]); //不用每次new 几遍不重写内存引用,也很站用资源 k.set(account, in, out); //value必须是NullWritable.get(),NullWritable不行,提示不是变量 context.write(k, NullWritable.get()); } }
六、结束语
如果k2 v2和k4 v4,也就是mapp的输出和reducer的输出类型不一致的话必须在Main里也设置Mapper的输出,上面的第二种就是。
job.setMapOutputKeyClass(InfoBean.class); job.setMapOutputValueClass(NullWritable.class); job.setOutputKeyClass(Text.class); job.setOutputValueClass(InfoBean.class);
否则java里不报错,加上log4j后看到类型不匹配。