将输出发送到多个目录的hadoop方法

Question

2013
    01
    02
    ..

2012
    01
    02
    ..

等等。

在任何时候，我只能获得长达12个月的数据，因此，我使用MultipleOutputs类在驱动程序中使用以下函数创建12个输出：

public void createOutputs(){
    Calendar c = Calendar.getInstance();
    String monthStr, pathStr;

    // Create multiple outputs for last 12 months
    // TODO make 12 configurable
    for(int i = 0; i < 12; ++i ){
        //Get month and add 1 as month is 0 based index
        int month = c.get(Calendar.MONTH)+1; 
        //Add leading 0
        monthStr = month > 10 ? "" + month : "0" + month ;  
        // Generate path string in the format 2013/03/etl
        pathStr = c.get(Calendar.YEAR) + "" + monthStr + "etl";
        // Add the named output
        MultipleOutputs.addNamedOutput(config, pathStr );  
        // Move to previous month
        c.add(Calendar.MONTH, -1); 
    }
}

在reducer中，我添加了一个cleanup函数，将生成的输出移动到适当的目录。

protected void cleanup(Context context) throws IOException, InterruptedException {
        // Custom function to recursively process data
        moveFiles (FileSystem.get(new Configuration()), new Path("/MyOutputPath"));
}

null

MyMapReduce: filepath:hdfs://localhost:8020/dev/test
MyMapReduce: filepath:hdfs://localhost:8020/dev/test/_logs
MyMapReduce: filepath:hdfs://localhost:8020/dev/test/_logs/history/job_201310301015_0224_1383763613843_371979_HtmlEtl
MyMapReduce: filepath:hdfs://localhost:8020/dev/test/_temporary
MyMapReduce: filepath:hdfs://localhost:8020/dev/test/_temporary/_attempt_201310301015_0224_r_000000_0
MyMapReduce: filepath:hdfs://localhost:8020/dev/test/_temporary/_attempt_201310301015_0224_r_000000_0/201307etl-r-00000
MyMapReduce: filepath:hdfs://localhost:8020/dev/test/_temporary/_attempt_201310301015_0224_r_000000_0/part-r-00000

Answer 1

你不应该需要第二份工作。我目前正在使用MultipleOutputs在我的一个程序中创建大量的输出目录。尽管有30多个目录，但我只能使用几个MultipleOutputs对象。这是因为您可以在写入时设置输出目录，所以只有在需要时才可以确定。如果要以不同的格式输出（例如，键:text.class，值:text.class，键:text.class，值：intwritable.class），实际上只需要多个namedOutput

设置：

MultipleOutputs.addNamedOutput(job, "Output", TextOutputFormat.class, Text.class, Text.class);

减速机的设置：

mout = new MultipleOutputs<Text, Text>(context);

String key; //set to whatever output key will be
String value; //set to whatever output value will be
String outputFileName; //set to absolute path to file where this should write

mout.write("Output",new Text(key),new Text(value),outputFileName);

int year;//extract year from data
int month;//extract month from data
String baseFileName; //parent directory to all outputs from this job
String outputFileName = baseFileName + "/" + year + "/" + month;

mout.write("Output",new Text(key),new Text(value),outputFileName);

Base
    2013
        01
        02
        03
        ...
    2012
        01
        ...
    ...