public virtual void TestTaskAttemptUnsuccessfulCompletionWithoutCounters0239()
{
Path histPath = new Path(GetType().GetClassLoader().GetResource("job_0.23.9-FAILED.jhist"
).GetFile());
JobHistoryParser parser = new JobHistoryParser(FileSystem.GetLocal(new Configuration
()), histPath);
JobHistoryParser.JobInfo jobInfo = parser.Parse();
Log.Info(" job info: " + jobInfo.GetJobname() + " " + jobInfo.GetFinishedMaps() +
" " + jobInfo.GetTotalMaps() + " " + jobInfo.GetJobId());
}
//History data is leisurely loaded when task level data is requested
/// <exception cref="System.IO.IOException"/>
protected internal virtual void LoadFullHistoryData(bool loadTasks, Path historyFileAbsolute
)
{
lock (this)
{
Log.Info("Loading history file: [" + historyFileAbsolute + "]");
if (this.jobInfo != null)
{
return;
}
if (historyFileAbsolute != null)
{
JobHistoryParser parser = null;
try
{
parser = new JobHistoryParser(historyFileAbsolute.GetFileSystem(conf), historyFileAbsolute
);
this.jobInfo = parser.Parse();
}
catch (IOException e)
{
throw new YarnRuntimeException("Could not load history file " + historyFileAbsolute
, e);
}
IOException parseException = parser.GetParseException();
if (parseException != null)
{
throw new YarnRuntimeException("Could not parse history file " + historyFileAbsolute
, parseException);
}
}
else
{
throw new IOException("History file not found");
}
if (loadTasks)
{
LoadAllTasks();
Log.Info("TaskInfo loaded");
}
}
}
public virtual void TestFailedJobHistoryWithoutDiagnostics()
{
Path histPath = new Path(GetType().GetClassLoader().GetResource("job_1393307629410_0001-1393307687476-user-Sleep+job-1393307723835-0-0-FAILED-default-1393307693920.jhist"
).GetFile());
FileSystem lfs = FileSystem.GetLocal(new Configuration());
FSDataInputStream fsdis = lfs.Open(histPath);
try
{
JobHistoryParser parser = new JobHistoryParser(fsdis);
JobHistoryParser.JobInfo info = parser.Parse();
NUnit.Framework.Assert.AreEqual("History parsed jobId incorrectly", info.GetJobId
(), JobID.ForName("job_1393307629410_0001"));
NUnit.Framework.Assert.AreEqual("Default diagnostics incorrect ", string.Empty, info
.GetErrorInfo());
}
finally
{
fsdis.Close();
}
}
public virtual void TestMultipleFailedTasks()
{
JobHistoryParser parser = new JobHistoryParser(Org.Mockito.Mockito.Mock <FSDataInputStream
>());
EventReader reader = Org.Mockito.Mockito.Mock <EventReader>();
AtomicInteger numEventsRead = new AtomicInteger(0);
// Hack!
TaskType taskType = TaskType.Map;
TaskID[] tids = new TaskID[2];
JobID jid = new JobID("1", 1);
tids[0] = new TaskID(jid, taskType, 0);
tids[1] = new TaskID(jid, taskType, 1);
Org.Mockito.Mockito.When(reader.GetNextEvent()).ThenAnswer(new _Answer_842(numEventsRead
, tids, taskType, jid));
// send two task start and two task fail events for tasks 0 and 1
JobHistoryParser.JobInfo info = parser.Parse(reader);
NUnit.Framework.Assert.IsTrue("Task 0 not implicated", info.GetErrorInfo().Contains
(tids[0].ToString()));
}
/// <exception cref="System.Exception"/>
public virtual void TestDiagnosticsForKilledJob()
{
Log.Info("STARTING testDiagnosticsForKilledJob");
try
{
Configuration conf = new Configuration();
conf.SetClass(CommonConfigurationKeysPublic.NetTopologyNodeSwitchMappingImplKey,
typeof(TestJobHistoryParsing.MyResolver), typeof(DNSToSwitchMapping));
RackResolver.Init(conf);
MRApp app = new TestJobHistoryParsing.MRAppWithHistoryWithJobKilled(2, 1, true, this
.GetType().FullName, true);
app.Submit(conf);
Org.Apache.Hadoop.Mapreduce.V2.App.Job.Job job = app.GetContext().GetAllJobs().Values
.GetEnumerator().Next();
JobId jobId = job.GetID();
app.WaitForState(job, JobState.Killed);
// make sure all events are flushed
app.WaitForState(Service.STATE.Stopped);
JobHistory jobHistory = new JobHistory();
jobHistory.Init(conf);
HistoryFileManager.HistoryFileInfo fileInfo = jobHistory.GetJobFileInfo(jobId);
JobHistoryParser parser;
JobHistoryParser.JobInfo jobInfo;
lock (fileInfo)
{
Path historyFilePath = fileInfo.GetHistoryFile();
FSDataInputStream @in = null;
FileContext fc = null;
try
{
fc = FileContext.GetFileContext(conf);
@in = fc.Open(fc.MakeQualified(historyFilePath));
}
catch (IOException ioe)
{
Log.Info("Can not open history file: " + historyFilePath, ioe);
throw (new Exception("Can not open History File"));
}
parser = new JobHistoryParser(@in);
jobInfo = parser.Parse();
}
Exception parseException = parser.GetParseException();
NUnit.Framework.Assert.IsNull("Caught an expected exception " + parseException, parseException
);
IList <string> originalDiagnostics = job.GetDiagnostics();
string historyError = jobInfo.GetErrorInfo();
NUnit.Framework.Assert.IsTrue("No original diagnostics for a failed job", originalDiagnostics
!= null && !originalDiagnostics.IsEmpty());
NUnit.Framework.Assert.IsNotNull("No history error info for a failed job ", historyError
);
foreach (string diagString in originalDiagnostics)
{
NUnit.Framework.Assert.IsTrue(historyError.Contains(diagString));
}
NUnit.Framework.Assert.IsTrue("No killed message in diagnostics", historyError.Contains
(JobImpl.JobKilledDiag));
}
finally
{
Log.Info("FINISHED testDiagnosticsForKilledJob");
}
}
/// <exception cref="System.Exception"/>
public virtual void TestHistoryParsingForFailedAttempts()
{
Log.Info("STARTING testHistoryParsingForFailedAttempts");
try
{
Configuration conf = new Configuration();
conf.SetClass(CommonConfigurationKeysPublic.NetTopologyNodeSwitchMappingImplKey,
typeof(TestJobHistoryParsing.MyResolver), typeof(DNSToSwitchMapping));
RackResolver.Init(conf);
MRApp app = new TestJobHistoryParsing.MRAppWithHistoryWithFailedAttempt(2, 1, true
, this.GetType().FullName, true);
app.Submit(conf);
Org.Apache.Hadoop.Mapreduce.V2.App.Job.Job job = app.GetContext().GetAllJobs().Values
.GetEnumerator().Next();
JobId jobId = job.GetID();
app.WaitForState(job, JobState.Succeeded);
// make sure all events are flushed
app.WaitForState(Service.STATE.Stopped);
JobHistory jobHistory = new JobHistory();
jobHistory.Init(conf);
HistoryFileManager.HistoryFileInfo fileInfo = jobHistory.GetJobFileInfo(jobId);
JobHistoryParser parser;
JobHistoryParser.JobInfo jobInfo;
lock (fileInfo)
{
Path historyFilePath = fileInfo.GetHistoryFile();
FSDataInputStream @in = null;
FileContext fc = null;
try
{
fc = FileContext.GetFileContext(conf);
@in = fc.Open(fc.MakeQualified(historyFilePath));
}
catch (IOException ioe)
{
Log.Info("Can not open history file: " + historyFilePath, ioe);
throw (new Exception("Can not open History File"));
}
parser = new JobHistoryParser(@in);
jobInfo = parser.Parse();
}
Exception parseException = parser.GetParseException();
NUnit.Framework.Assert.IsNull("Caught an expected exception " + parseException, parseException
);
int noOffailedAttempts = 0;
IDictionary <TaskID, JobHistoryParser.TaskInfo> allTasks = jobInfo.GetAllTasks();
foreach (Task task in job.GetTasks().Values)
{
JobHistoryParser.TaskInfo taskInfo = allTasks[TypeConverter.FromYarn(task.GetID()
)];
foreach (TaskAttempt taskAttempt in task.GetAttempts().Values)
{
JobHistoryParser.TaskAttemptInfo taskAttemptInfo = taskInfo.GetAllTaskAttempts()[
TypeConverter.FromYarn((taskAttempt.GetID()))];
// Verify rack-name for all task attempts
NUnit.Framework.Assert.AreEqual("rack-name is incorrect", taskAttemptInfo.GetRackname
(), RackName);
if (taskAttemptInfo.GetTaskStatus().Equals("FAILED"))
{
noOffailedAttempts++;
}
}
}
NUnit.Framework.Assert.AreEqual("No of Failed tasks doesn't match.", 2, noOffailedAttempts
);
}
finally
{
Log.Info("FINISHED testHistoryParsingForFailedAttempts");
}
}
/// <exception cref="System.Exception"/>
private void CheckHistoryParsing(int numMaps, int numReduces, int numSuccessfulMaps
)
{
Configuration conf = new Configuration();
conf.Set(MRJobConfig.UserName, Runtime.GetProperty("user.name"));
long amStartTimeEst = Runtime.CurrentTimeMillis();
conf.SetClass(CommonConfigurationKeysPublic.NetTopologyNodeSwitchMappingImplKey,
typeof(TestJobHistoryParsing.MyResolver), typeof(DNSToSwitchMapping));
RackResolver.Init(conf);
MRApp app = new TestJobHistoryEvents.MRAppWithHistory(numMaps, numReduces, true,
this.GetType().FullName, true);
app.Submit(conf);
Org.Apache.Hadoop.Mapreduce.V2.App.Job.Job job = app.GetContext().GetAllJobs().Values
.GetEnumerator().Next();
JobId jobId = job.GetID();
Log.Info("JOBID is " + TypeConverter.FromYarn(jobId).ToString());
app.WaitForState(job, JobState.Succeeded);
// make sure all events are flushed
app.WaitForState(Service.STATE.Stopped);
string jobhistoryDir = JobHistoryUtils.GetHistoryIntermediateDoneDirForUser(conf);
FileContext fc = null;
try
{
fc = FileContext.GetFileContext(conf);
}
catch (IOException ioe)
{
Log.Info("Can not get FileContext", ioe);
throw (new Exception("Can not get File Context"));
}
if (numMaps == numSuccessfulMaps)
{
string summaryFileName = JobHistoryUtils.GetIntermediateSummaryFileName(jobId);
Path summaryFile = new Path(jobhistoryDir, summaryFileName);
string jobSummaryString = GetJobSummary(fc, summaryFile);
NUnit.Framework.Assert.IsNotNull(jobSummaryString);
NUnit.Framework.Assert.IsTrue(jobSummaryString.Contains("resourcesPerMap=100"));
NUnit.Framework.Assert.IsTrue(jobSummaryString.Contains("resourcesPerReduce=100")
);
IDictionary <string, string> jobSummaryElements = new Dictionary <string, string>();
StringTokenizer strToken = new StringTokenizer(jobSummaryString, ",");
while (strToken.HasMoreTokens())
{
string keypair = strToken.NextToken();
jobSummaryElements[keypair.Split("=")[0]] = keypair.Split("=")[1];
}
NUnit.Framework.Assert.AreEqual("JobId does not match", jobId.ToString(), jobSummaryElements
["jobId"]);
NUnit.Framework.Assert.AreEqual("JobName does not match", "test", jobSummaryElements
["jobName"]);
NUnit.Framework.Assert.IsTrue("submitTime should not be 0", long.Parse(jobSummaryElements
["submitTime"]) != 0);
NUnit.Framework.Assert.IsTrue("launchTime should not be 0", long.Parse(jobSummaryElements
["launchTime"]) != 0);
NUnit.Framework.Assert.IsTrue("firstMapTaskLaunchTime should not be 0", long.Parse
(jobSummaryElements["firstMapTaskLaunchTime"]) != 0);
NUnit.Framework.Assert.IsTrue("firstReduceTaskLaunchTime should not be 0", long.Parse
(jobSummaryElements["firstReduceTaskLaunchTime"]) != 0);
NUnit.Framework.Assert.IsTrue("finishTime should not be 0", long.Parse(jobSummaryElements
["finishTime"]) != 0);
NUnit.Framework.Assert.AreEqual("Mismatch in num map slots", numSuccessfulMaps, System.Convert.ToInt32
(jobSummaryElements["numMaps"]));
NUnit.Framework.Assert.AreEqual("Mismatch in num reduce slots", numReduces, System.Convert.ToInt32
(jobSummaryElements["numReduces"]));
NUnit.Framework.Assert.AreEqual("User does not match", Runtime.GetProperty("user.name"
), jobSummaryElements["user"]);
NUnit.Framework.Assert.AreEqual("Queue does not match", "default", jobSummaryElements
["queue"]);
NUnit.Framework.Assert.AreEqual("Status does not match", "SUCCEEDED", jobSummaryElements
["status"]);
}
JobHistory jobHistory = new JobHistory();
jobHistory.Init(conf);
HistoryFileManager.HistoryFileInfo fileInfo = jobHistory.GetJobFileInfo(jobId);
JobHistoryParser.JobInfo jobInfo;
long numFinishedMaps;
lock (fileInfo)
{
Path historyFilePath = fileInfo.GetHistoryFile();
FSDataInputStream @in = null;
Log.Info("JobHistoryFile is: " + historyFilePath);
try
{
@in = fc.Open(fc.MakeQualified(historyFilePath));
}
catch (IOException ioe)
{
Log.Info("Can not open history file: " + historyFilePath, ioe);
throw (new Exception("Can not open History File"));
}
JobHistoryParser parser = new JobHistoryParser(@in);
EventReader realReader = new EventReader(@in);
EventReader reader = Org.Mockito.Mockito.Mock <EventReader>();
if (numMaps == numSuccessfulMaps)
{
reader = realReader;
}
else
{
AtomicInteger numFinishedEvents = new AtomicInteger(0);
// Hack!
Org.Mockito.Mockito.When(reader.GetNextEvent()).ThenAnswer(new _Answer_257(realReader
, numFinishedEvents, numSuccessfulMaps));
}
jobInfo = parser.Parse(reader);
numFinishedMaps = ComputeFinishedMaps(jobInfo, numMaps, numSuccessfulMaps);
if (numFinishedMaps != numMaps)
{
Exception parseException = parser.GetParseException();
NUnit.Framework.Assert.IsNotNull("Didn't get expected parse exception", parseException
);
}
}
NUnit.Framework.Assert.AreEqual("Incorrect username ", Runtime.GetProperty("user.name"
), jobInfo.GetUsername());
NUnit.Framework.Assert.AreEqual("Incorrect jobName ", "test", jobInfo.GetJobname(
));
NUnit.Framework.Assert.AreEqual("Incorrect queuename ", "default", jobInfo.GetJobQueueName
());
NUnit.Framework.Assert.AreEqual("incorrect conf path", "test", jobInfo.GetJobConfPath
());
NUnit.Framework.Assert.AreEqual("incorrect finishedMap ", numSuccessfulMaps, numFinishedMaps
);
NUnit.Framework.Assert.AreEqual("incorrect finishedReduces ", numReduces, jobInfo
.GetFinishedReduces());
NUnit.Framework.Assert.AreEqual("incorrect uberized ", job.IsUber(), jobInfo.GetUberized
());
IDictionary <TaskID, JobHistoryParser.TaskInfo> allTasks = jobInfo.GetAllTasks();
int totalTasks = allTasks.Count;
NUnit.Framework.Assert.AreEqual("total number of tasks is incorrect ", (numMaps
+ numReduces), totalTasks);
// Verify aminfo
NUnit.Framework.Assert.AreEqual(1, jobInfo.GetAMInfos().Count);
NUnit.Framework.Assert.AreEqual(MRApp.NmHost, jobInfo.GetAMInfos()[0].GetNodeManagerHost
());
JobHistoryParser.AMInfo amInfo = jobInfo.GetAMInfos()[0];
NUnit.Framework.Assert.AreEqual(MRApp.NmPort, amInfo.GetNodeManagerPort());
NUnit.Framework.Assert.AreEqual(MRApp.NmHttpPort, amInfo.GetNodeManagerHttpPort()
);
NUnit.Framework.Assert.AreEqual(1, amInfo.GetAppAttemptId().GetAttemptId());
NUnit.Framework.Assert.AreEqual(amInfo.GetAppAttemptId(), amInfo.GetContainerId()
.GetApplicationAttemptId());
NUnit.Framework.Assert.IsTrue(amInfo.GetStartTime() <= Runtime.CurrentTimeMillis(
) && amInfo.GetStartTime() >= amStartTimeEst);
ContainerId fakeCid = MRApp.NewContainerId(-1, -1, -1, -1);
// Assert at taskAttempt level
foreach (JobHistoryParser.TaskInfo taskInfo in allTasks.Values)
{
int taskAttemptCount = taskInfo.GetAllTaskAttempts().Count;
NUnit.Framework.Assert.AreEqual("total number of task attempts ", 1, taskAttemptCount
);
JobHistoryParser.TaskAttemptInfo taInfo = taskInfo.GetAllTaskAttempts().Values.GetEnumerator
().Next();
NUnit.Framework.Assert.IsNotNull(taInfo.GetContainerId());
// Verify the wrong ctor is not being used. Remove after mrv1 is removed.
NUnit.Framework.Assert.IsFalse(taInfo.GetContainerId().Equals(fakeCid));
}
// Deep compare Job and JobInfo
foreach (Task task in job.GetTasks().Values)
{
JobHistoryParser.TaskInfo taskInfo_1 = allTasks[TypeConverter.FromYarn(task.GetID
())];
NUnit.Framework.Assert.IsNotNull("TaskInfo not found", taskInfo_1);
foreach (TaskAttempt taskAttempt in task.GetAttempts().Values)
{
JobHistoryParser.TaskAttemptInfo taskAttemptInfo = taskInfo_1.GetAllTaskAttempts(
)[TypeConverter.FromYarn((taskAttempt.GetID()))];
NUnit.Framework.Assert.IsNotNull("TaskAttemptInfo not found", taskAttemptInfo);
NUnit.Framework.Assert.AreEqual("Incorrect shuffle port for task attempt", taskAttempt
.GetShufflePort(), taskAttemptInfo.GetShufflePort());
if (numMaps == numSuccessfulMaps)
{
NUnit.Framework.Assert.AreEqual(MRApp.NmHost, taskAttemptInfo.GetHostname());
NUnit.Framework.Assert.AreEqual(MRApp.NmPort, taskAttemptInfo.GetPort());
// Verify rack-name
NUnit.Framework.Assert.AreEqual("rack-name is incorrect", taskAttemptInfo.GetRackname
(), RackName);
}
}
}
// test output for HistoryViewer
TextWriter stdps = System.Console.Out;
try
{
Runtime.SetOut(new TextWriter(outContent));
HistoryViewer viewer;
lock (fileInfo)
{
viewer = new HistoryViewer(fc.MakeQualified(fileInfo.GetHistoryFile()).ToString()
, conf, true);
}
viewer.Print();
foreach (JobHistoryParser.TaskInfo taskInfo_1 in allTasks.Values)
{
string test = (taskInfo_1.GetTaskStatus() == null ? string.Empty : taskInfo_1.GetTaskStatus
()) + " " + taskInfo_1.GetTaskType() + " task list for " + taskInfo_1.GetTaskId(
).GetJobID();
NUnit.Framework.Assert.IsTrue(outContent.ToString().IndexOf(test) > 0);
NUnit.Framework.Assert.IsTrue(outContent.ToString().IndexOf(taskInfo_1.GetTaskId(
).ToString()) > 0);
}
}
finally
{
Runtime.SetOut(stdps);
}
}
