MAPREDUCE-4646. Fixed MR framework to send diagnostic information correctly to clients in case of failed jobs also. Contributed by Jason Lowe.
svn merge --ignore-ancestry -c 1383709 ../../trunk/ git-svn-id: https://svn.apache.org/repos/asf/hadoop/common/branches/branch-2@1383710 13f79535-47bb-0310-9956-ffa450edef68
This commit is contained in:
parent
ddf8691f03
commit
a5e50dfae9
|
@ -15,6 +15,9 @@ Release 2.0.3-alpha - Unreleased
|
||||||
MAPREDUCE-4607. Race condition in ReduceTask completion can result in Task
|
MAPREDUCE-4607. Race condition in ReduceTask completion can result in Task
|
||||||
being incorrectly failed. (Bikas Saha via tomwhite)
|
being incorrectly failed. (Bikas Saha via tomwhite)
|
||||||
|
|
||||||
|
MAPREDUCE-4646. Fixed MR framework to send diagnostic information correctly
|
||||||
|
to clients in case of failed jobs also. (Jason Lowe via vinodkv)
|
||||||
|
|
||||||
Release 2.0.2-alpha - 2012-09-07
|
Release 2.0.2-alpha - 2012-09-07
|
||||||
|
|
||||||
INCOMPATIBLE CHANGES
|
INCOMPATIBLE CHANGES
|
||||||
|
|
|
@ -582,17 +582,23 @@ public class JobImpl implements org.apache.hadoop.mapreduce.v2.app.job.Job,
|
||||||
String jobFile =
|
String jobFile =
|
||||||
remoteJobConfFile == null ? "" : remoteJobConfFile.toString();
|
remoteJobConfFile == null ? "" : remoteJobConfFile.toString();
|
||||||
|
|
||||||
|
StringBuilder diagsb = new StringBuilder();
|
||||||
|
for (String s : getDiagnostics()) {
|
||||||
|
diagsb.append(s).append("\n");
|
||||||
|
}
|
||||||
|
|
||||||
if (getState() == JobState.NEW) {
|
if (getState() == JobState.NEW) {
|
||||||
return MRBuilderUtils.newJobReport(jobId, jobName, username, state,
|
return MRBuilderUtils.newJobReport(jobId, jobName, username, state,
|
||||||
appSubmitTime, startTime, finishTime, setupProgress, 0.0f, 0.0f,
|
appSubmitTime, startTime, finishTime, setupProgress, 0.0f, 0.0f,
|
||||||
cleanupProgress, jobFile, amInfos, isUber);
|
cleanupProgress, jobFile, amInfos, isUber, diagsb.toString());
|
||||||
}
|
}
|
||||||
|
|
||||||
computeProgress();
|
computeProgress();
|
||||||
return MRBuilderUtils.newJobReport(jobId, jobName, username, state,
|
JobReport report = MRBuilderUtils.newJobReport(jobId, jobName, username,
|
||||||
appSubmitTime, startTime, finishTime, setupProgress,
|
state, appSubmitTime, startTime, finishTime, setupProgress,
|
||||||
this.mapProgress, this.reduceProgress,
|
this.mapProgress, this.reduceProgress,
|
||||||
cleanupProgress, jobFile, amInfos, isUber);
|
cleanupProgress, jobFile, amInfos, isUber, diagsb.toString());
|
||||||
|
return report;
|
||||||
} finally {
|
} finally {
|
||||||
readLock.unlock();
|
readLock.unlock();
|
||||||
}
|
}
|
||||||
|
|
|
@ -138,7 +138,7 @@ public class TestRMContainerAllocator {
|
||||||
Job mockJob = mock(Job.class);
|
Job mockJob = mock(Job.class);
|
||||||
when(mockJob.getReport()).thenReturn(
|
when(mockJob.getReport()).thenReturn(
|
||||||
MRBuilderUtils.newJobReport(jobId, "job", "user", JobState.RUNNING, 0,
|
MRBuilderUtils.newJobReport(jobId, "job", "user", JobState.RUNNING, 0,
|
||||||
0, 0, 0, 0, 0, 0, "jobfile", null, false));
|
0, 0, 0, 0, 0, 0, "jobfile", null, false, ""));
|
||||||
MyContainerAllocator allocator = new MyContainerAllocator(rm, conf,
|
MyContainerAllocator allocator = new MyContainerAllocator(rm, conf,
|
||||||
appAttemptId, mockJob);
|
appAttemptId, mockJob);
|
||||||
|
|
||||||
|
@ -215,7 +215,7 @@ public class TestRMContainerAllocator {
|
||||||
Job mockJob = mock(Job.class);
|
Job mockJob = mock(Job.class);
|
||||||
when(mockJob.getReport()).thenReturn(
|
when(mockJob.getReport()).thenReturn(
|
||||||
MRBuilderUtils.newJobReport(jobId, "job", "user", JobState.RUNNING, 0,
|
MRBuilderUtils.newJobReport(jobId, "job", "user", JobState.RUNNING, 0,
|
||||||
0, 0, 0, 0, 0, 0, "jobfile", null, false));
|
0, 0, 0, 0, 0, 0, "jobfile", null, false, ""));
|
||||||
MyContainerAllocator allocator = new MyContainerAllocator(rm, conf,
|
MyContainerAllocator allocator = new MyContainerAllocator(rm, conf,
|
||||||
appAttemptId, mockJob);
|
appAttemptId, mockJob);
|
||||||
|
|
||||||
|
@ -281,7 +281,7 @@ public class TestRMContainerAllocator {
|
||||||
Job mockJob = mock(Job.class);
|
Job mockJob = mock(Job.class);
|
||||||
when(mockJob.getReport()).thenReturn(
|
when(mockJob.getReport()).thenReturn(
|
||||||
MRBuilderUtils.newJobReport(jobId, "job", "user", JobState.RUNNING, 0,
|
MRBuilderUtils.newJobReport(jobId, "job", "user", JobState.RUNNING, 0,
|
||||||
0, 0, 0, 0, 0, 0, "jobfile", null, false));
|
0, 0, 0, 0, 0, 0, "jobfile", null, false, ""));
|
||||||
MyContainerAllocator allocator = new MyContainerAllocator(rm, conf,
|
MyContainerAllocator allocator = new MyContainerAllocator(rm, conf,
|
||||||
appAttemptId, mockJob);
|
appAttemptId, mockJob);
|
||||||
|
|
||||||
|
@ -723,7 +723,7 @@ public class TestRMContainerAllocator {
|
||||||
Job mockJob = mock(Job.class);
|
Job mockJob = mock(Job.class);
|
||||||
when(mockJob.getReport()).thenReturn(
|
when(mockJob.getReport()).thenReturn(
|
||||||
MRBuilderUtils.newJobReport(jobId, "job", "user", JobState.RUNNING, 0,
|
MRBuilderUtils.newJobReport(jobId, "job", "user", JobState.RUNNING, 0,
|
||||||
0, 0, 0, 0, 0, 0, "jobfile", null, false));
|
0, 0, 0, 0, 0, 0, "jobfile", null, false, ""));
|
||||||
MyContainerAllocator allocator = new MyContainerAllocator(rm, conf,
|
MyContainerAllocator allocator = new MyContainerAllocator(rm, conf,
|
||||||
appAttemptId, mockJob);
|
appAttemptId, mockJob);
|
||||||
|
|
||||||
|
@ -827,7 +827,7 @@ public class TestRMContainerAllocator {
|
||||||
Job mockJob = mock(Job.class);
|
Job mockJob = mock(Job.class);
|
||||||
when(mockJob.getReport()).thenReturn(
|
when(mockJob.getReport()).thenReturn(
|
||||||
MRBuilderUtils.newJobReport(jobId, "job", "user", JobState.RUNNING, 0,
|
MRBuilderUtils.newJobReport(jobId, "job", "user", JobState.RUNNING, 0,
|
||||||
0, 0, 0, 0, 0, 0, "jobfile", null, false));
|
0, 0, 0, 0, 0, 0, "jobfile", null, false, ""));
|
||||||
MyContainerAllocator allocator =
|
MyContainerAllocator allocator =
|
||||||
new MyContainerAllocator(rm, conf, appAttemptId, mockJob);
|
new MyContainerAllocator(rm, conf, appAttemptId, mockJob);
|
||||||
|
|
||||||
|
@ -993,7 +993,7 @@ public class TestRMContainerAllocator {
|
||||||
Job mockJob = mock(Job.class);
|
Job mockJob = mock(Job.class);
|
||||||
when(mockJob.getReport()).thenReturn(
|
when(mockJob.getReport()).thenReturn(
|
||||||
MRBuilderUtils.newJobReport(jobId, "job", "user", JobState.RUNNING, 0,
|
MRBuilderUtils.newJobReport(jobId, "job", "user", JobState.RUNNING, 0,
|
||||||
0, 0, 0, 0, 0, 0, "jobfile", null, false));
|
0, 0, 0, 0, 0, 0, "jobfile", null, false, ""));
|
||||||
MyContainerAllocator allocator = new MyContainerAllocator(rm, conf,
|
MyContainerAllocator allocator = new MyContainerAllocator(rm, conf,
|
||||||
appAttemptId, mockJob);
|
appAttemptId, mockJob);
|
||||||
|
|
||||||
|
@ -1445,7 +1445,7 @@ public class TestRMContainerAllocator {
|
||||||
Job job = mock(Job.class);
|
Job job = mock(Job.class);
|
||||||
when(job.getReport()).thenReturn(
|
when(job.getReport()).thenReturn(
|
||||||
MRBuilderUtils.newJobReport(jobId, "job", "user", JobState.RUNNING, 0,
|
MRBuilderUtils.newJobReport(jobId, "job", "user", JobState.RUNNING, 0,
|
||||||
0, 0, 0, 0, 0, 0, "jobfile", null, false));
|
0, 0, 0, 0, 0, 0, "jobfile", null, false, ""));
|
||||||
doReturn(10).when(job).getTotalMaps();
|
doReturn(10).when(job).getTotalMaps();
|
||||||
doReturn(10).when(job).getTotalReduces();
|
doReturn(10).when(job).getTotalReduces();
|
||||||
doReturn(0).when(job).getCompletedMaps();
|
doReturn(0).when(job).getCompletedMaps();
|
||||||
|
|
|
@ -45,11 +45,14 @@ import org.apache.hadoop.mapreduce.v2.api.records.JobId;
|
||||||
import org.apache.hadoop.mapreduce.v2.api.records.JobState;
|
import org.apache.hadoop.mapreduce.v2.api.records.JobState;
|
||||||
import org.apache.hadoop.mapreduce.v2.api.records.TaskId;
|
import org.apache.hadoop.mapreduce.v2.api.records.TaskId;
|
||||||
import org.apache.hadoop.mapreduce.v2.app.job.Task;
|
import org.apache.hadoop.mapreduce.v2.app.job.Task;
|
||||||
|
import org.apache.hadoop.mapreduce.v2.app.job.event.JobDiagnosticsUpdateEvent;
|
||||||
import org.apache.hadoop.mapreduce.v2.app.job.event.JobEvent;
|
import org.apache.hadoop.mapreduce.v2.app.job.event.JobEvent;
|
||||||
|
import org.apache.hadoop.mapreduce.v2.app.job.event.JobEventType;
|
||||||
import org.apache.hadoop.mapreduce.v2.app.job.impl.JobImpl.InitTransition;
|
import org.apache.hadoop.mapreduce.v2.app.job.impl.JobImpl.InitTransition;
|
||||||
import org.apache.hadoop.mapreduce.v2.app.job.impl.JobImpl.JobNoTasksCompletedTransition;
|
import org.apache.hadoop.mapreduce.v2.app.job.impl.JobImpl.JobNoTasksCompletedTransition;
|
||||||
import org.apache.hadoop.mapreduce.v2.app.metrics.MRAppMetrics;
|
import org.apache.hadoop.mapreduce.v2.app.metrics.MRAppMetrics;
|
||||||
import org.apache.hadoop.security.UserGroupInformation;
|
import org.apache.hadoop.security.UserGroupInformation;
|
||||||
|
import org.apache.hadoop.yarn.SystemClock;
|
||||||
import org.apache.hadoop.yarn.api.records.ApplicationAttemptId;
|
import org.apache.hadoop.yarn.api.records.ApplicationAttemptId;
|
||||||
import org.apache.hadoop.yarn.event.EventHandler;
|
import org.apache.hadoop.yarn.event.EventHandler;
|
||||||
import org.apache.hadoop.yarn.util.Records;
|
import org.apache.hadoop.yarn.util.Records;
|
||||||
|
@ -172,6 +175,8 @@ public class TestJobImpl {
|
||||||
t.testCheckJobCompleteSuccess();
|
t.testCheckJobCompleteSuccess();
|
||||||
t.testCheckJobCompleteSuccessFailed();
|
t.testCheckJobCompleteSuccessFailed();
|
||||||
t.testCheckAccess();
|
t.testCheckAccess();
|
||||||
|
t.testReportDiagnostics();
|
||||||
|
t.testUberDecision();
|
||||||
}
|
}
|
||||||
|
|
||||||
@Test
|
@Test
|
||||||
|
@ -241,6 +246,41 @@ public class TestJobImpl {
|
||||||
Assert.assertTrue(job5.checkAccess(ugi1, null));
|
Assert.assertTrue(job5.checkAccess(ugi1, null));
|
||||||
Assert.assertTrue(job5.checkAccess(ugi2, null));
|
Assert.assertTrue(job5.checkAccess(ugi2, null));
|
||||||
}
|
}
|
||||||
|
|
||||||
|
@Test
|
||||||
|
public void testReportDiagnostics() throws Exception {
|
||||||
|
JobID jobID = JobID.forName("job_1234567890000_0001");
|
||||||
|
JobId jobId = TypeConverter.toYarn(jobID);
|
||||||
|
final String diagMsg = "some diagnostic message";
|
||||||
|
final JobDiagnosticsUpdateEvent diagUpdateEvent =
|
||||||
|
new JobDiagnosticsUpdateEvent(jobId, diagMsg);
|
||||||
|
MRAppMetrics mrAppMetrics = MRAppMetrics.create();
|
||||||
|
JobImpl job = new JobImpl(jobId, Records
|
||||||
|
.newRecord(ApplicationAttemptId.class), new Configuration(),
|
||||||
|
mock(EventHandler.class),
|
||||||
|
null, mock(JobTokenSecretManager.class), null,
|
||||||
|
new SystemClock(), null,
|
||||||
|
mrAppMetrics, mock(OutputCommitter.class),
|
||||||
|
true, null, 0, null, null);
|
||||||
|
job.handle(diagUpdateEvent);
|
||||||
|
String diagnostics = job.getReport().getDiagnostics();
|
||||||
|
Assert.assertNotNull(diagnostics);
|
||||||
|
Assert.assertTrue(diagnostics.contains(diagMsg));
|
||||||
|
|
||||||
|
job = new JobImpl(jobId, Records
|
||||||
|
.newRecord(ApplicationAttemptId.class), new Configuration(),
|
||||||
|
mock(EventHandler.class),
|
||||||
|
null, mock(JobTokenSecretManager.class), null,
|
||||||
|
new SystemClock(), null,
|
||||||
|
mrAppMetrics, mock(OutputCommitter.class),
|
||||||
|
true, null, 0, null, null);
|
||||||
|
job.handle(new JobEvent(jobId, JobEventType.JOB_KILL));
|
||||||
|
job.handle(diagUpdateEvent);
|
||||||
|
diagnostics = job.getReport().getDiagnostics();
|
||||||
|
Assert.assertNotNull(diagnostics);
|
||||||
|
Assert.assertTrue(diagnostics.contains(diagMsg));
|
||||||
|
}
|
||||||
|
|
||||||
@Test
|
@Test
|
||||||
public void testUberDecision() throws Exception {
|
public void testUberDecision() throws Exception {
|
||||||
|
|
||||||
|
|
|
@ -67,7 +67,7 @@ public class MRBuilderUtils {
|
||||||
String userName, JobState state, long submitTime, long startTime, long finishTime,
|
String userName, JobState state, long submitTime, long startTime, long finishTime,
|
||||||
float setupProgress, float mapProgress, float reduceProgress,
|
float setupProgress, float mapProgress, float reduceProgress,
|
||||||
float cleanupProgress, String jobFile, List<AMInfo> amInfos,
|
float cleanupProgress, String jobFile, List<AMInfo> amInfos,
|
||||||
boolean isUber) {
|
boolean isUber, String diagnostics) {
|
||||||
JobReport report = Records.newRecord(JobReport.class);
|
JobReport report = Records.newRecord(JobReport.class);
|
||||||
report.setJobId(jobId);
|
report.setJobId(jobId);
|
||||||
report.setJobName(jobName);
|
report.setJobName(jobName);
|
||||||
|
@ -83,6 +83,7 @@ public class MRBuilderUtils {
|
||||||
report.setJobFile(jobFile);
|
report.setJobFile(jobFile);
|
||||||
report.setAMInfos(amInfos);
|
report.setAMInfos(amInfos);
|
||||||
report.setIsUber(isUber);
|
report.setIsUber(isUber);
|
||||||
|
report.setDiagnostics(diagnostics);
|
||||||
return report;
|
return report;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
|
@ -219,7 +219,8 @@ public class TestClientServiceDelegate {
|
||||||
GetJobReportResponse jobReportResponse1 = mock(GetJobReportResponse.class);
|
GetJobReportResponse jobReportResponse1 = mock(GetJobReportResponse.class);
|
||||||
when(jobReportResponse1.getJobReport()).thenReturn(
|
when(jobReportResponse1.getJobReport()).thenReturn(
|
||||||
MRBuilderUtils.newJobReport(jobId, "jobName-firstGen", "user",
|
MRBuilderUtils.newJobReport(jobId, "jobName-firstGen", "user",
|
||||||
JobState.RUNNING, 0, 0, 0, 0, 0, 0, 0, "anything", null, false));
|
JobState.RUNNING, 0, 0, 0, 0, 0, 0, 0, "anything", null,
|
||||||
|
false, ""));
|
||||||
|
|
||||||
// First AM returns a report with jobName firstGen and simulates AM shutdown
|
// First AM returns a report with jobName firstGen and simulates AM shutdown
|
||||||
// on second invocation.
|
// on second invocation.
|
||||||
|
@ -231,7 +232,8 @@ public class TestClientServiceDelegate {
|
||||||
GetJobReportResponse jobReportResponse2 = mock(GetJobReportResponse.class);
|
GetJobReportResponse jobReportResponse2 = mock(GetJobReportResponse.class);
|
||||||
when(jobReportResponse2.getJobReport()).thenReturn(
|
when(jobReportResponse2.getJobReport()).thenReturn(
|
||||||
MRBuilderUtils.newJobReport(jobId, "jobName-secondGen", "user",
|
MRBuilderUtils.newJobReport(jobId, "jobName-secondGen", "user",
|
||||||
JobState.RUNNING, 0, 0, 0, 0, 0, 0, 0, "anything", null, false));
|
JobState.RUNNING, 0, 0, 0, 0, 0, 0, 0, "anything", null,
|
||||||
|
false, ""));
|
||||||
|
|
||||||
// Second AM generation returns a report with jobName secondGen
|
// Second AM generation returns a report with jobName secondGen
|
||||||
MRClientProtocol secondGenAMProxy = mock(MRClientProtocol.class);
|
MRClientProtocol secondGenAMProxy = mock(MRClientProtocol.class);
|
||||||
|
|
Loading…
Reference in New Issue