Time for action – handling dirty data by using skip mode

Let's see skip mode in action by writing a MapReduce job that receives the data that causes it to fail:

1. Save the following Ruby script as gendata.rb:

   File.open("skipdata.txt", "w") do |file|
     3.times do
       500000.times{file.write("A valid record\n")}
       5.times{file.write("skiptext\n")}
     end
     500000.times{file.write("A valid record\n")}
   End

2. Run the script:

   $ ruby gendata.rb 
   

3. Check the size of the generated file and its number of lines:

   $ ls -lh skipdata.txt
   -rw-rw-r-- 1 hadoop hadoop 29M 2011-12-17 01:53 skipdata.txt
   ~$ cat skipdata.txt | wc -l
   2000015
   

4. Copy the file onto HDFS:

   $ hadoop fs -put skipdata.txt skipdata.txt
   

5. Add the following property definition to mapred-site.xml:

   <property>
   <name>mapred.skip.map.max.skip.records</name>
   <value5</value>
   </property>

6. Check the value set for mapred.max.map.task.failures and set it to 20 if it is lower.

7. Save the following Java file as SkipData.java:

   import java...