Sqoop源码分析(4) Sqoop中通过hadoop mapreduce从关系型数据库import数据分析

Sqoop源码分析(四) Sqoop中通过hadoop mapreduce从关系型数据库import数据分析

      本博客属原创文章，转载请务必注明出处:http://guoyunsky.iteye.com/blogs/1213966/         

       欢迎加入Hadoop超级群: 180941958

           Sqoop中一大亮点就是可以通过hadoop的mapreduce从关系型数据库中导入数据到HDFS,如此可以加快导入时间.一直想了解MapReduce,所以也仔细的阅读了下相关代码,整理成这篇博客.

            一.原理:

             Sqoop在import时,需要制定split-by参数.Sqoop根据不同的split-by参数值来进行切分,然后将切分出来的区域分配到不同map中.每个map中再处理数据库中获取的一行一行的值,写入到HDFS中.同时split-by根据不同的参数类型有不同的切分方法，如比较简单的int型，Sqoop会取最大和最小split-by字段值，然后根据传入的num-mappers来确定划分几个区域。比如select max(split_by),min(split-by) from得到的max(split-by)和min(split-by)分别为1000和1，而num-mappers为2的话，则会分成两个区域(1,500)和(501-100),同时也会分成2个sql给2个map去进行导入操作，分别为select XXX from table where split-by>=1 and split-by<500和select XXX from table where split-by>=501 and split-by<=1000.最后每个map各自获取各自SQL中的数据进行导入工作。

      

       二.mapreduce job所需要的各种参数在Sqoop中的实现

           1)InputFormatClass
              com.cloudera.sqoop.mapreduce.db.DataDrivenDBInputFormat
           2)OutputFormatClass
               1)TextFile
                   com.cloudera.sqoop.mapreduce.RawKeyTextOutputFormat
               2)SequenceFile
                   org.apache.hadoop.mapreduce.lib.output.SequenceFileOutputFormat 
               3)AvroDataFile
                   com.cloudera.sqoop.mapreduce.AvroOutputFormat
          3)Mapper
             1)TextFile
                 com.cloudera.sqoop.mapreduce.TextImportMapper
             2)SequenceFile
                 com.cloudera.sqoop.mapreduce.SequenceFileImportMapper
            3)AvroDataFile
                 com.cloudera.sqoop.mapreduce.AvroImportMapper
         4)taskNumbers
            1)mapred.map.tasks (对应num-mappers参数)
            2)job.setNumReduceTasks(0);

       这里以我命令行:import --connect jdbc:mysql://localhost/sqoop_datas  --username root --password 123456 --query "select sqoop_1.id as foo_id, sqoop_2.id as bar_id from sqoop_1 ,sqoop_2  WHERE $CONDITIONS" --target-dir /tmp/sqoop/foo2 -split-by sqoop_1.id   --hadoop-home=/home/guoyun/Downloads/hadoop-0.20.2-CDH3B4  --num-mappers 2

       注:红色部分参数，后接根据我的命令衍生的参数值

      1）设置Input

       DataDrivenImportJob.configureInputFormat(Job job, String tableName,String tableClassName, String splitByCol)
           a)DBConfiguration.configureDB(Configuration conf, String driverClass,
           String dbUrl, String userName, String passwd, Integer fetchSize)
                1).mapreduce.jdbc.driver.class com.mysql.jdbc.Driver
                2).mapreduce.jdbc.url  jdbc:mysql://localhost/sqoop_datas
                3).mapreduce.jdbc.username  root
                4).mapreduce.jdbc.password  123456
                5).mapreduce.jdbc.fetchsize -2147483648
            b)DataDrivenDBInputFormat.setInput(Job job,Class<? extends DBWritable> inputClass,
           String inputQuery, String inputBoundingQuery)
               1)job.setInputFormatClass(DBInputFormat.class);
               2)mapred.jdbc.input.bounding.query SELECT MIN(sqoop_1.id), MAX(sqoop_2.id) FROM (select sqoop_1.id as foo_id, sqoop_2.id as bar_id from sqoop_1 ,sqoop_2  WHERE  (1 = 1) ) AS t1
               3)job.setInputFormatClass(DataDrivenDBInputFormat.class);
               4)mapreduce.jdbc.input.orderby sqoop_1.id
           c)mapreduce.jdbc.input.class QueryResult
           d)sqoop.inline.lob.length.max 16777216
         3)job.setInputFormatClass(inputFormatClass); class com.cloudera.sqoop.mapreduce.db.DataDrivenDBInputFormat

   2）设置Output 

    ImportJobBase.configureOutputFormat(Job job, String tableName,String tableClassName)
             a)job.setOutputFormatClass(getOutputFormatClass());
             b)FileOutputFormat.setOutputCompressorClass(job, codecClass);
             c)SequenceFileOutputFormat.setOutputCompressionType(job,CompressionType.BLOCK);
             d)FileOutputFormat.setOutputPath(job, outputPath);

    3）设置Map

       DataDrivenImportJob.configureMapper(Job job, String tableName,String tableClassName)
              a) job.setOutputKeyClass(Text.class);
              b)job.setOutputValueClass(NullWritable.class);
              c)job.setMapperClass(com.cloudera.sqoop.mapreduce.TextImportMapper);

    4）设置task number

       JobBase.configureNumTasks(Job job)
              mapred.map.tasks 4
              job.setNumReduceTasks(0);

       三。大概流程

          1.读取要导入数据的表结构，生成运行类，默认是QueryResult，打成jar包，然后提交给Hadoop

          2.设置好job，主要也就是设置好以上二中的各个参数

          3.这里就由Hadoop来执行MapReduce来执行Import命令了，

             1）首先要对数据进行切分，也就是DataSplit

                 DataDrivenDBInputFormat.getSplits(JobContext job)

             2）切分好范围后，写入范围，以便读取

                DataDrivenDBInputFormat.write(DataOutput output) 

                这里是lowerBoundQuery and  upperBoundQuery

             3）读取以上2）写入的范围

                DataDrivenDBInputFormat.readFields(DataInput input)

             4）然后创建RecordReader从数据库中读取数据

                DataDrivenDBInputFormat.createRecordReader(InputSplit split,TaskAttemptContext context)

             5）创建Map

                TextImportMapper.setup(Context context)

             6）RecordReader一行一行从关系型数据库中读取数据，设置好Map的Key和Value，交给Map

                DBRecordReader.nextKeyValue() 

             7）运行map

                 TextImportMapper.map(LongWritable key, SqoopRecord val, Context context)

                  最后生成的Key是行数据，由QueryResult生成，Value是NullWritable.get()

  四.总结

       通过这些,我大概了解了MapReduce运行流程.但对于Sqoop这种切分方式感觉还是有很大的问题.比如这里根据ID范围来切分,如此切分出来的数据会很不平均,比如min(split-id)=1,max(split-id)=3000，交给三个map来处理。那么范围是(1-1000),(1001-2000),(2001-3000).而假如1001-2000是没有数据，已经被删除了。那么这个map就什么都不能做。而其他map却累的半死。如此就会拖累job的运行结果。我这里说的范围很小，比如有几十亿条数据交给几百个map去做。map一多，如果任务不均衡就会影响进度。看有没有更好的切分方式？比如取样？如此看来，写好map reduce也不简单!