Apache
/
hadoop
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

MAPREDUCE-6489. Fail fast rogue tasks that write too much to local disk. Contributed by Maysam Yabandeh

This commit is contained in:
Jason Lowe 2015-10-21 14:10:28 +00:00
parent cb19552fb1
commit be72ed1c75
5 changed files with 164 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
hadoop-mapreduce-project/CHANGES.txt
View File

@ -127,6 +127,9 @@ Release 2.8.0 - UNRELEASED
MAPREDUCE-6479. Add missing mapred job command options in mapreduce
document. (nijel via aajisaka)
MAPREDUCE-6489. Fail fast rogue tasks that write too much to local disk
(Maysam Yabandeh via jlowe)
OPTIMIZATIONS
MAPREDUCE-6376. Add avro binary support for jhist files (Ray Chiang via

72
hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-core/src/main/java/org/apache/hadoop/mapred/Task.java
View File

@ -42,6 +42,7 @@ import org.apache.hadoop.conf.Configurable;
import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.FileSystem;
import org.apache.hadoop.fs.LocalDirAllocator;
import org.apache.hadoop.fs.LocalFileSystem;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.fs.FileSystem.Statistics;
import org.apache.hadoop.io.BytesWritable;
@ -63,6 +64,7 @@ import org.apache.hadoop.mapreduce.lib.reduce.WrappedReducer;
import org.apache.hadoop.mapreduce.task.ReduceContextImpl;
import org.apache.hadoop.yarn.util.ResourceCalculatorProcessTree;
import org.apache.hadoop.net.NetUtils;
import org.apache.hadoop.util.ExitUtil;
import org.apache.hadoop.util.Progress;
import org.apache.hadoop.util.Progressable;
import org.apache.hadoop.util.ReflectionUtils;
@ -729,11 +731,49 @@ abstract public class Task implements Writable, Configurable {
} else {
return split;
}
}
/**
* The communication thread handles communication with the parent (Task Tracker).
* It sends progress updates if progress has been made or if the task needs to
* let the parent know that it's alive. It also pings the parent to see if it's alive.
}
/**
* exception thrown when the task exceeds some configured limits.
*/
public class TaskLimitException extends IOException {
public TaskLimitException(String str) {
super(str);
}
}
/**
* check the counters to see whether the task has exceeded any configured
* limits.
* @throws TaskLimitException
*/
protected void checkTaskLimits() throws TaskLimitException {
// check the limit for writing to local file system
long limit = conf.getLong(MRJobConfig.TASK_LOCAL_WRITE_LIMIT_BYTES,
MRJobConfig.DEFAULT_TASK_LOCAL_WRITE_LIMIT_BYTES);
if (limit >= 0) {
Counters.Counter localWritesCounter = null;
try {
LocalFileSystem localFS = FileSystem.getLocal(conf);
localWritesCounter = counters.findCounter(localFS.getScheme(),
FileSystemCounter.BYTES_WRITTEN);
} catch (IOException e) {
LOG.warn("Could not get LocalFileSystem BYTES_WRITTEN counter");
}
if (localWritesCounter != null
&& localWritesCounter.getCounter() > limit) {
throw new TaskLimitException("too much write to local file system." +
" current value is " + localWritesCounter.getCounter() +
" the limit is " + limit);
}
}
}
/**
* The communication thread handles communication with the parent (Task
* Tracker). It sends progress updates if progress has been made or if
* the task needs to let the parent know that it's alive. It also pings
* the parent to see if it's alive.
*/
public void run() {
final int MAX_RETRIES = 3;
@ -763,8 +803,9 @@ abstract public class Task implements Writable, Configurable {
if (sendProgress) {
// we need to send progress update
updateCounters();
checkTaskLimits();
taskStatus.statusUpdate(taskProgress.get(),
taskProgress.toString(),
taskProgress.toString(),
counters);
taskFound = umbilical.statusUpdate(taskId, taskStatus);
taskStatus.clearStatus();
@ -782,10 +823,21 @@ abstract public class Task implements Writable, Configurable {
System.exit(66);
}
sendProgress = resetProgressFlag();
sendProgress = resetProgressFlag();
remainingRetries = MAX_RETRIES;
}
catch (Throwable t) {
} catch (TaskLimitException e) {
String errMsg = "Task exceeded the limits: " +
StringUtils.stringifyException(e);
LOG.fatal(errMsg);
try {
umbilical.fatalError(taskId, errMsg);
} catch (IOException ioe) {
LOG.fatal("Failed to update failure diagnosis", ioe);
}
LOG.fatal("Killing " + taskId);
resetDoneFlag();
ExitUtil.terminate(69);
} catch (Throwable t) {
LOG.info("Communication exception: " + StringUtils.stringifyException(t));
remainingRetries -=1;
if (remainingRetries == 0) {
@ -1043,7 +1095,7 @@ abstract public class Task implements Writable, Configurable {
TaskReporter reporter
) throws IOException, InterruptedException {
LOG.info("Task:" + taskId + " is done."
+ " And is in the process of committing");
+ " And is in the process of committing");
updateCounters();
boolean commitRequired = isCommitRequired();

5
hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-core/src/main/java/org/apache/hadoop/mapreduce/MRJobConfig.java
View File

@ -49,6 +49,11 @@ public interface MRJobConfig {
public static final String TASK_CLEANUP_NEEDED = "mapreduce.job.committer.task.cleanup.needed";
public static final String TASK_LOCAL_WRITE_LIMIT_BYTES =
"mapreduce.task.local-fs.write-limit.bytes";
// negative values disable the limit
public static final long DEFAULT_TASK_LOCAL_WRITE_LIMIT_BYTES = -1;
public static final String TASK_PROGRESS_REPORT_INTERVAL =
"mapreduce.task.progress-report.interval";
/** The number of milliseconds between progress reports. */

12
hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-core/src/main/resources/mapred-default.xml
View File

@ -1781,4 +1781,16 @@
</description>
</property>
<property>
<name>mapreduce.task.local-fs.write-limit.bytes</name>
<value>-1</value>
<description>Limit on the byte written to the local file system by each task.
This limit only applies to writes that go through the Hadoop filesystem APIs
within the task process (i.e.: writes that will update the local filesystem's
BYTES_WRITTEN counter). It does not cover other writes such as logging,
sideband writes from subprocesses (e.g.: streaming jobs), etc.
Negative values disable the limit.
default is -1</description>
</property>
</configuration>

83
hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-core/src/test/java/org/apache/hadoop/mapred/TestTaskProgressReporter.java
View File

@ -19,15 +19,28 @@
package org.apache.hadoop.mapred;
import java.io.IOException;
import java.util.Random;
import org.apache.hadoop.fs.FSDataOutputStream;
import org.apache.hadoop.fs.FileSystem;
import org.apache.hadoop.fs.LocalFileSystem;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.ipc.ProtocolSignature;
import org.apache.hadoop.mapred.SortedRanges.Range;
import org.apache.hadoop.mapreduce.MRJobConfig;
import org.apache.hadoop.util.ExitUtil;
import org.junit.Assert;
import org.junit.Test;
public class TestTaskProgressReporter {
private static int statusUpdateTimes = 0;
// set to true if the thread is existed with ExitUtil.terminate
volatile boolean threadExited = false;
final static int LOCAL_BYTES_WRITTEN = 1024;
private FakeUmbilical fakeUmbilical = new FakeUmbilical();
private static class DummyTask extends Task {
@ -120,13 +133,22 @@ public class TestTaskProgressReporter {
}
private class DummyTaskReporter extends Task.TaskReporter {
volatile boolean taskLimitIsChecked = false;
public DummyTaskReporter(Task task) {
task.super(task.getProgress(), fakeUmbilical);
}
@Override
public void setProgress(float progress) {
super.setProgress(progress);
}
@Override
protected void checkTaskLimits() throws TaskLimitException {
taskLimitIsChecked = true;
super.checkTaskLimits();
}
}
@Test (timeout=10000)
@ -144,4 +166,63 @@ public class TestTaskProgressReporter {
t.join();
Assert.assertEquals(statusUpdateTimes, 2);
}
}
@Test(timeout=10000)
public void testBytesWrittenRespectingLimit() throws Exception {
// add 1024 to the limit to account for writes not controlled by the test
testBytesWrittenLimit(LOCAL_BYTES_WRITTEN + 1024, false);
}
@Test(timeout=10000)
public void testBytesWrittenExceedingLimit() throws Exception {
testBytesWrittenLimit(LOCAL_BYTES_WRITTEN - 1, true);
}
/**
* This is to test the limit on BYTES_WRITTEN. The test is limited in that
* the check is done only once at the first loop of TaskReport#run.
* @param limit the limit on BYTES_WRITTEN in local file system
* @param failFast should the task fail fast with such limit?
* @throws Exception
*/
public void testBytesWrittenLimit(long limit, boolean failFast)
throws Exception {
ExitUtil.disableSystemExit();
threadExited = false;
Thread.UncaughtExceptionHandler h = new Thread.UncaughtExceptionHandler() {
public void uncaughtException(Thread th, Throwable ex) {
System.out.println("Uncaught exception: " + ex);
if (ex instanceof ExitUtil.ExitException) {
threadExited = true;
}
}
};
JobConf conf = new JobConf();
// To disable task reporter sleeping
conf.getLong(MRJobConfig.TASK_PROGRESS_REPORT_INTERVAL, 0);
conf.setLong(MRJobConfig.TASK_LOCAL_WRITE_LIMIT_BYTES, limit);
LocalFileSystem localFS = FileSystem.getLocal(conf);
Path tmpPath = new Path("/tmp/testBytesWrittenLimit-tmpFile-"
+ new Random(System.currentTimeMillis()).nextInt());
FSDataOutputStream out = localFS.create(tmpPath, true);
out.write(new byte[LOCAL_BYTES_WRITTEN]);
out.close();
Task task = new DummyTask();
task.setConf(conf);
DummyTaskReporter reporter = new DummyTaskReporter(task);
Thread t = new Thread(reporter);
t.setUncaughtExceptionHandler(h);
reporter.setProgressFlag();
t.start();
while (!reporter.taskLimitIsChecked) {
Thread.yield();
}
task.setTaskDone();
reporter.resetDoneFlag();
t.join();
Assert.assertEquals(failFast, threadExited);
}
}