Heap error when using custom RecordReader with large file

Issue

I've written a custom file reader to not split my input files as they are large gzipped files and I want my first mapper job to simply gunzip them. I followed the example in 'Hadoop The Definitive Guide' but I get a heap error when trying to read in to the BytesWritable. I believe this is because the byte array is of size 85713669, but I'm not sure how to overcome this issue.

Here is the code:

public class WholeFileRecordReader extends RecordReader<NullWritable, BytesWritable> {

private FileSplit fileSplit;
private Configuration conf;
private BytesWritable value = new BytesWritable();
private boolean processed = false;

@Override
public void close() throws IOException {
    // do nothing
}

@Override
public NullWritable getCurrentKey() throws IOException,
        InterruptedException {
    return NullWritable.get();
}

@Override
public BytesWritable getCurrentValue() throws IOException,
        InterruptedException {
    return value;
}

@Override
public float getProgress() throws IOException, InterruptedException {
    return processed ? 1.0f : 0.0f;
}

@Override
public void initialize(InputSplit split, TaskAttemptContext context)
        throws IOException, InterruptedException {
    this.fileSplit = (FileSplit) split;
    this.conf = context.getConfiguration();
}

@Override
public boolean nextKeyValue() throws IOException, InterruptedException {
    if (!processed) {
        byte[] contents = new byte[(int) fileSplit.getLength()];
        Path file = fileSplit.getPath();
        FileSystem fs = file.getFileSystem(conf);
        FSDataInputStream in = null;
        try {
            in = fs.open(file);
            IOUtils.readFully(in, contents, 0, contents.length);
            value.set(contents, 0, contents.length);
        } finally {
            IOUtils.closeStream(in);
        }
        processed = true;
        return true;
    }
    return false;
}

}

Solution

In general you can not load whole file into memory of Java VM. You should find some streaming solution to process large files - read data chunk by chunk and save the results w/o fixing in memory whole data set
This specific task - unzip is probably not suited for the MR since there is no logical division of data into records.
Please also note that hadoop is handling gzip automatically - your input stream will be already decompressed.

Answered By - David Gruzman