Question

A sqoop query generates a java file that contains a class that contains the code to get access in mapreduce to the columns data for each row. (the Sqoop import was done in text without the –as-sequencefile option, and with 1 line per record and commas between the columns)
But how do we actually use it?

I found a public method parse() in this class that takes Text as an input and populates all the members of the class , so to practice I modified the wordcount application to convert a line of text from the TextInputFormat in the mapper into an instnace of the class generated by sqoop. But that causes an “unreported exception.com.cloudera.sqoop.lib.RecordParser.ParseError; must be caught or declared to be thrown” when I call the parse() method.

Can it be done this way or is a custom InputFormat necessary to populate the class with the data from each record ?

Answer 1

Ok this seems obvious once you find out but as a java beginner this can take time.

First configure your project:
just add the sqoop generated .java file in your source folder.
I use eclipse to import it in my class source folder.

Then just make sure you configured your project’s java build path correctly:

Add the following jar files in the project’s properties/java build path/libraries/add external jar:
(for hadoop cdh4+) :

/usr/lib/hadoop/hadoop-common.jar 
/usr/lib/hadoop-[version]-mapreduce/hadoop-core.jar
/usr/lib/sqoop/sqoop-[sqoop-version]-cdh[cdh-version].jar

Then adapt your mapreduce source code:
First configure it:

public int run(String [] args) throws exception
{
 Job job = new Job(getConf());
 job.setJarByClass(YourClass.class);
 job.setMapperClass(SqoopImportMap.class);
 job.setReducerClass(SqoopImprtReduce.class);

 FileInputFormat.addInputPath((job,"hdfs_path_to_your_sqoop_imported_file"));
 FileOutputFormat.setOutputPath((job,"hdfs_output_path"));

 // I simply use text as output for the mapper but it can be any class you designed
 // as long as you implement it as a Writable
 job.setMapOutputKeyClass(Text.Class);
 job.setMapOutputValueClass(Text.Class);

 job.setOutputKeyClass(Text.Class);
 job.setOutputValueClass(Text.Class);
 ...

Now configure your mapper class.
Let’s assume your sqoop imported java file is called Sqimp.java:
and the table you imported had the following columns: id, name, age
your mapper class should look like this:

 public static class SqoopImportMap
 extends Mapper<LongWritable, Text, Text, Text> 
 {

 public void map(LongWritable k, Text v, Context context)
 {
  Sqimp s = new Sqimp(); 
  try
  {
  // this is where the code generated by sqoop is used.
  // it automatically casts one line of the imported data into an instance of the generated class, 
  // to let you access the data inside the columns easily
   s.parse(v);
  } 
  catch(ParseError pe) {// do something if there is an error.}

  try
  {
   // now the imported data is accessible:
   // e.g
   if (s.age>30)
   {
    // submit the selected data to the mapper's output as a key value pair.
    context.write(new Text(s.age),new Text(s.id));
   }
  }
  catch(Exception ex)
  {//do something about the error}
 }
}