[MAPREDUCE-2236] No task may execute due to an Integer overflow possibility

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Not A Problem
• Affects Version/s: 0.20.2

Environment

Linux, Hadoop 0.20.2

• Release Note: Cap maximum retry attempts to 100, to avoid an overflow possibility.
• Tags: Integer overflow, TaskInProgress
• Assignee: Harsh J
• Reporter: Harsh J
• Created: 29/Dec/10 19:39
• Updated: 28/Sep/15 21:10
• Resolved: 11/Dec/11 20:23

Description

If the attempts is configured to use Integer.MAX_VALUE, an overflow occurs inside TaskInProgress, and thereby no task is attempted by the cluster and the map tasks stay in pending state forever.

For example, here's a job driver that causes this:

import java.io.IOException;

import org.apache.hadoop.fs.FSDataOutputStream;
import org.apache.hadoop.fs.FileSystem;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.mapred.FileInputFormat;
import org.apache.hadoop.mapred.JobClient;
import org.apache.hadoop.mapred.JobConf;
import org.apache.hadoop.mapred.TextInputFormat;
import org.apache.hadoop.mapred.lib.IdentityMapper;
import org.apache.hadoop.mapred.lib.NullOutputFormat;


@SuppressWarnings("deprecation")
public class IntegerOverflow {

	/**
	 * @param args
	 * @throws IOException 
	 */
	@SuppressWarnings("deprecation")
	public static void main(String[] args) throws IOException {
		JobConf conf = new JobConf();
		
		Path inputPath = new Path("ignore");
		FileSystem fs = FileSystem.get(conf);
		if (!fs.exists(inputPath)) {
			FSDataOutputStream out = fs.create(inputPath);
			out.writeChars("Test");
			out.close();
		}
		
		conf.setInputFormat(TextInputFormat.class);
		conf.setOutputFormat(NullOutputFormat.class);
		FileInputFormat.addInputPath(conf, inputPath);
		
		conf.setMapperClass(IdentityMapper.class);
		conf.setNumMapTasks(1);
		// Problem inducing line follows.
		conf.setMaxMapAttempts(Integer.MAX_VALUE);
		
		// No reducer in this test, although setMaxReduceAttempts leads to the same problem.
		conf.setNumReduceTasks(0);
		
		JobClient.runJob(conf);
	}

}

The above code will not let any map task run. Additionally, a log would be created inside JobTracker logs with the following information that clearly shows the overflow:

2010-12-30 00:59:07,836 WARN org.apache.hadoop.mapred.TaskInProgress: Exceeded limit of -2147483648 (plus 0 killed) attempts for the tip 'task_201012300058_0001_m_000000'

The issue lies inside the TaskInProgress class (/o/a/h/mapred/TaskInProgress.java), at line 1018 (trunk), part of the getTaskToRun(String taskTracker) method.

  public Task getTaskToRun(String taskTracker) throws IOException {   
    // Create the 'taskid'; do not count the 'killed' tasks against the job!
    TaskAttemptID taskid = null;
    /* ============ THIS LINE v ====================================== */
    if (nextTaskId < (MAX_TASK_EXECS + maxTaskAttempts + numKilledTasks)) {
    /* ============ THIS LINE ^====================================== */
      // Make sure that the attempts are unqiue across restarts
      int attemptId = job.getNumRestarts() * NUM_ATTEMPTS_PER_RESTART + nextTaskId;
      taskid = new TaskAttemptID( id, attemptId);
      ++nextTaskId;
    } else {
      LOG.warn("Exceeded limit of " + (MAX_TASK_EXECS + maxTaskAttempts) +
              " (plus " + numKilledTasks + " killed)"  + 
              " attempts for the tip '" + getTIPId() + "'");
      return null;
    }

Since all three variables being added are integer in type, one of them being Integer.MAX_VALUE makes the condition fail with an overflow, thereby logging and returning a null as the result is negative.

One solution would be to make one of these variables into a long, so the addition does not overflow?