// run a job for which all the attempts simply fail.
/// <exception cref="System.IO.IOException"/>
private void TestFailedJob(string fileName, Type committer, string[] exclude)
{
JobConf jc = mr.CreateJobConf();
Path outDir = GetNewOutputDir();
ConfigureJob(jc, "fail job with abort()", 1, 0, outDir);
jc.SetMaxMapAttempts(1);
// set the job to fail
jc.SetMapperClass(typeof(UtilsForTests.FailMapper));
jc.SetOutputCommitter(committer);
JobClient jobClient = new JobClient(jc);
RunningJob job = jobClient.SubmitJob(jc);
JobID id = job.GetID();
job.WaitForCompletion();
NUnit.Framework.Assert.AreEqual("Job did not fail", JobStatus.Failed, job.GetJobState
());
if (fileName != null)
{
Path testFile = new Path(outDir, fileName);
NUnit.Framework.Assert.IsTrue("File " + testFile + " missing for failed job " + id
, fileSys.Exists(testFile));
}
// check if the files from the missing set exists
foreach (string ex in exclude)
{
Path file = new Path(outDir, ex);
NUnit.Framework.Assert.IsFalse("File " + file + " should not be present for failed job "
+ id, fileSys.Exists(file));
}
}
// runs a sample job as a user (ugi)
/// <exception cref="System.Exception"/>
internal virtual void RunJobAsUser(JobConf job, UserGroupInformation ugi)
{
RunningJob rj = ugi.DoAs(new _PrivilegedExceptionAction_66(job));
rj.WaitForCompletion();
NUnit.Framework.Assert.AreEqual("SUCCEEDED", JobStatus.GetJobRunState(rj.GetJobState
()));
}
/// <exception cref="System.IO.IOException"/>
public virtual void TestCommitFail()
{
Path inDir = new Path(rootDir, "./input");
Path outDir = new Path(rootDir, "./output");
JobConf jobConf = CreateJobConf();
jobConf.SetMaxMapAttempts(1);
jobConf.SetOutputCommitter(typeof(TestTaskCommit.CommitterWithCommitFail));
RunningJob rJob = UtilsForTests.RunJob(jobConf, inDir, outDir, 1, 0);
rJob.WaitForCompletion();
NUnit.Framework.Assert.AreEqual(JobStatus.Failed, rJob.GetJobState());
}
// Run a job that will be killed and wait until it completes
/// <exception cref="System.IO.IOException"/>
public static RunningJob RunJobKill(JobConf conf, Path inDir, Path outDir)
{
conf.SetJobName("test-job-kill");
conf.SetMapperClass(typeof(UtilsForTests.KillMapper));
conf.SetReducerClass(typeof(IdentityReducer));
RunningJob job = UtilsForTests.RunJob(conf, inDir, outDir);
long sleepCount = 0;
while (job.GetJobState() != JobStatus.Running)
{
try
{
if (sleepCount > 300)
{
// 30 seconds
throw new IOException("Job didn't finish in 30 seconds");
}
Sharpen.Thread.Sleep(100);
sleepCount++;
}
catch (Exception)
{
break;
}
}
job.KillJob();
sleepCount = 0;
while (job.CleanupProgress() == 0.0f)
{
try
{
if (sleepCount > 2000)
{
// 20 seconds
throw new IOException("Job cleanup didn't start in 20 seconds");
}
Sharpen.Thread.Sleep(10);
sleepCount++;
}
catch (Exception)
{
break;
}
}
return(job);
}
// run a job which gets stuck in mapper and kill it.
/// <exception cref="System.IO.IOException"/>
private void TestKilledJob(string fileName, Type committer, string[] exclude)
{
JobConf jc = mr.CreateJobConf();
Path outDir = GetNewOutputDir();
ConfigureJob(jc, "kill job with abort()", 1, 0, outDir);
// set the job to wait for long
jc.SetMapperClass(typeof(UtilsForTests.KillMapper));
jc.SetOutputCommitter(committer);
JobClient jobClient = new JobClient(jc);
RunningJob job = jobClient.SubmitJob(jc);
JobID id = job.GetID();
Counters counters = job.GetCounters();
// wait for the map to be launched
while (true)
{
if (counters.GetCounter(JobCounter.TotalLaunchedMaps) == 1)
{
break;
}
Log.Info("Waiting for a map task to be launched");
UtilsForTests.WaitFor(100);
counters = job.GetCounters();
}
job.KillJob();
// kill the job
job.WaitForCompletion();
// wait for the job to complete
NUnit.Framework.Assert.AreEqual("Job was not killed", JobStatus.Killed, job.GetJobState
());
if (fileName != null)
{
Path testFile = new Path(outDir, fileName);
NUnit.Framework.Assert.IsTrue("File " + testFile + " missing for job " + id, fileSys
.Exists(testFile));
}
// check if the files from the missing set exists
foreach (string ex in exclude)
{
Path file = new Path(outDir, ex);
NUnit.Framework.Assert.IsFalse("File " + file + " should not be present for killed job "
+ id, fileSys.Exists(file));
}
}
//Starts the job in a thread. It also starts the taskKill/tasktrackerKill
//threads.
/// <exception cref="System.Exception"/>
private void RunTest(JobClient jc, Configuration conf, string jobClass, string[]
args, ReliabilityTest.KillTaskThread killTaskThread, ReliabilityTest.KillTrackerThread
killTrackerThread)
{
Sharpen.Thread t = new _Thread_202(this, conf, jobClass, args, "Job Test");
t.SetDaemon(true);
t.Start();
JobStatus[] jobs;
//get the job ID. This is the job that we just submitted
while ((jobs = jc.JobsToComplete()).Length == 0)
{
Log.Info("Waiting for the job " + jobClass + " to start");
Sharpen.Thread.Sleep(1000);
}
JobID jobId = ((JobID)jobs[jobs.Length - 1].GetJobID());
RunningJob rJob = jc.GetJob(jobId);
if (rJob.IsComplete())
{
Log.Error("The last job returned by the querying JobTracker is complete :" + rJob
.GetJobID() + " .Exiting the test");
System.Environment.Exit(-1);
}
while (rJob.GetJobState() == JobStatus.Prep)
{
Log.Info("JobID : " + jobId + " not started RUNNING yet");
Sharpen.Thread.Sleep(1000);
rJob = jc.GetJob(jobId);
}
if (killTaskThread != null)
{
killTaskThread.SetRunningJob(rJob);
killTaskThread.Start();
killTaskThread.Join();
Log.Info("DONE WITH THE TASK KILL/FAIL TESTS");
}
if (killTrackerThread != null)
{
killTrackerThread.SetRunningJob(rJob);
killTrackerThread.Start();
killTrackerThread.Join();
Log.Info("DONE WITH THE TESTS TO DO WITH LOST TASKTRACKERS");
}
t.Join();
}