[ Skip to the content ]

Institute of Formal and Applied Linguistics Wiki

[ Back to the navigation ]

You are here: start » courses » mapreduce-tutorial » step-29

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revision Previous revision
Next revision 		Previous revision
Next revision Both sides next revision
courses:mapreduce-tutorial:step-29 [2012/01/29 17:40]
straka 		courses:mapreduce-tutorial:step-29 [2012/01/30 00:53]
straka
Line 7: Line 7:
 We start by creating ''FileAsPathInputFormat'', which reads any file, splits it and for each split return exactly one input pair (file_path, start-length) with types (''Text'', ''Text''), where ''file_path'' is path to the file and ''start-length'' is a string containing two dash-separated numbers: start offset of the split and length of the split. We start by creating ''FileAsPathInputFormat'', which reads any file, splits it and for each split return exactly one input pair (file_path, start-length) with types (''Text'', ''Text''), where ''file_path'' is path to the file and ''start-length'' is a string containing two dash-separated numbers: start offset of the split and length of the split.
  
-When +When implementing new input format, we must 
 +  * decide whether the input files are splittable. Usually uncompressed files are splittable and compressed files are not splittable, with the exception of ''SequenceFile'', which is always splittable. 
 +  * implement [[http://hadoop.apache.org/common/docs/r1.0.0/api/org/apache/hadoop/mapreduce/RecordReader.html|RecordReader<K, V>]]. The ''RecordReader'' is the one doing the real work -- it is given a file split and it reads (key, value) pairs of types (K, V), until there are any.
  
 +Our ''FileAsPathInputFormat'' is simple -- we allow splitting of uncompressed file and the ''RecordReader'' reads exactly one input pair.
 <code java> <code java>
-public static class FileAsPathInputFormat extends FileInputFormat<Text, Text> {+public class FileAsPathInputFormat extends FileInputFormat<Text, Text> { 
 +  // Helper class, which does the actual work -- produce the (path, offset-length) input pair.
   public static class FileAsPathRecordReader extends RecordReader<Text, Text> {   public static class FileAsPathRecordReader extends RecordReader<Text, Text> {
     private Path file;     private Path file;
Line 39: Line 43:
   }   }
              
 +  // Use the helper class as a RecordReader in out file format.
   public RecordReader<Text, Text> createRecordReader(InputSplit split, TaskAttemptContext context) {   public RecordReader<Text, Text> createRecordReader(InputSplit split, TaskAttemptContext context) {
     return new FileAsPathRecordReader();     return new FileAsPathRecordReader();
   }      }   
              
 +  // Allow splitting uncompressed files.
   protected boolean isSplittable(JobContext context, Path filename) {   protected boolean isSplittable(JobContext context, Path filename) {
     CompressionCodec codec = new CompressionCodecFactory(context.getConfiguration()).getCodec(filename);     CompressionCodec codec = new CompressionCodecFactory(context.getConfiguration()).getCodec(filename);
Line 52: Line 58:
 ===== WholeFileInputFormat ===== ===== WholeFileInputFormat =====
  
-We start by creating ''WholeFileInputFormat'', which reads any file and return exactly one input pair (input_path, file_content) with types (''Text'', ''BytesWritable''). The format does not allow file splitting -- each file will be processed by exactly one mapper. +Next we create ''WhileFileInputFormat'', which for each file return exactly one input pair (input_path, file_content) with types (''Text'', ''BytesWritable''). The format does not allow file splitting -- each file will be processed by exactly one mapper.
- +
-The main functionality lays in ''WholeFileRecordReader'', a subclass of [[http://hadoop.apache.org/common/docs/r1.0.0/api/org/apache/hadoop/mapreduce/RecordReader.html|RecordReader<Text, BytesWritable]].+
  
 <code java> <code java>
Line 62: Line 66:
     private Path file;     private Path file;
     int length;     int length;
-    private boolean value_read; 
     private Text key;     private Text key;
     private BytesWritable value;     private BytesWritable value;
Line 73: Line 76:
       key = null;       key = null;
       value = null;       value = null;
-      value_read = false; 
  
       FileSystem fs = file.getFileSystem(context.getConfiguration());       FileSystem fs = file.getFileSystem(context.getConfiguration());
Line 85: Line 87:
  
     public boolean nextKeyValue() throws IOException {     public boolean nextKeyValue() throws IOException {
-      if (value_read) return false;+      if (key != null) return false;
  
       byte[] data = new byte[length];       byte[] data = new byte[length];
Line 92: Line 94:
       key = new Text(file.toString());       key = new Text(file.toString());
       value = new BytesWritable(data);       value = new BytesWritable(data);
-      value_read = true; 
  
       return true;       return true;
Line 99: Line 100:
     public Text getCurrentKey() { return key; }     public Text getCurrentKey() { return key; }
     public BytesWritable getCurrentValue() { return value; }     public BytesWritable getCurrentValue() { return value; }
-    public float getProgress() { return value_read ? 0 : 1; }+    public float getProgress() { return key == null ? 0 : 1; }
     public synchronized void close() throws IOException { if (in != null) { in.close(); in = null; } }     public synchronized void close() throws IOException { if (in != null) { in.close(); in = null; } }
   }   }
Line 113: Line 114:
   }   }
 } }
- 
 </code> </code>
  
 +===== Exercise: ParagraphTextInputFormat =====
 +
 +Implement ''ParagraphTextInputFormat''. This format reads plain text files and splits it into //paragraphs//. A paragraph consists of lines, all of which are nonempty, and different paragraphs are separated by at least one empty line. The ''ParagraphTextInputFormat'' reads one paragraph at a time and return its first line as key and the rest of lines as values.
 +
 +The ''ParagraphTextInputFormat'' should allow splitting of uncompressed files. Be careful to properly implement reading paragraphs which are on split boundary. The easiest way of doing so is the following:
 +  * if the offset of the split is 0, start reading at the beginning of the split. If the offset of the split is larger than 0, start reading at the offset and ignore first paragraph found.
 +  * read all paragraphs that start before the end of the split boundary, even if they end after the split boundary. //If a paragraph starts just after the current split (i.e., on the split boundary), read it too.//
 +It is simple to verify that with these rules, all paragraphs are read exactly once.

[ Back to the navigation ] [ Back to the content ]