public void testJobRetire() throws Exception {
MiniMRCluster mr = null;
try {
JobConf conf = new JobConf();
mr = startCluster(conf, 1);
JobConf jobConf = mr.createJobConf();
JobTracker jobtracker = mr.getJobTrackerRunner().getJobTracker();
Path inDir = new Path(testDir, "input1");
Path outDir = new Path(testDir, "output1");
JobID id1 = validateJobRetire(jobConf, inDir, outDir, jobtracker);
outDir = new Path(testDir, "output2");
JobID id2 = validateJobRetire(jobConf, inDir, outDir, jobtracker);
assertNull("Job not removed from cache", jobtracker.getJobStatus(id1));
assertEquals("Total job in cache not correct", 1, jobtracker.getAllJobs().length);
} finally {
if (mr != null) {
mr.shutdown();
}
FileUtil.fullyDelete(new File(testDir.toString()));
}
}
/** (Mock)Test JobTracker.removeJobTasks() which is called only when the job retires. */
public void testJobRemoval() throws Exception {
MiniMRCluster mr = null;
try {
JobConf conf = new JobConf();
mr = startCluster(conf, 0);
JobTracker jobtracker = mr.getJobTrackerRunner().getJobTracker();
// test map task removal
testRemoveJobTasks(jobtracker, conf, TaskType.MAP);
// test reduce task removal
testRemoveJobTasks(jobtracker, conf, TaskType.REDUCE);
// test job setup removal
testRemoveJobTasks(jobtracker, conf, TaskType.JOB_SETUP);
// test job cleanup removal
testRemoveJobTasks(jobtracker, conf, TaskType.JOB_CLEANUP);
} finally {
if (mr != null) {
mr.shutdown();
}
// cleanup
FileUtil.fullyDelete(new File(testDir.toString()));
}
}
/** Check refreshNodes for decommissioning blacklisted nodes. */
public void testBlacklistedNodeDecommissioning() throws Exception {
LOG.info("Testing blacklisted node decommissioning");
MiniMRCluster mr = null;
JobTracker jt = null;
try {
// start mini mr
JobConf jtConf = new JobConf();
jtConf.set("mapred.max.tracker.blacklists", "1");
mr = new MiniMRCluster(0, 0, 2, "file:///", 1, null, null, null, jtConf);
jt = mr.getJobTrackerRunner().getJobTracker();
assertEquals("Trackers not up", 2, jt.taskTrackers().size());
// validate the total tracker count
assertEquals(
"Active tracker count mismatch", 2, jt.getClusterStatus(false).getTaskTrackers());
// validate blacklisted count
assertEquals(
"Blacklisted tracker count mismatch",
0,
jt.getClusterStatus(false).getBlacklistedTrackers());
// run a failing job to blacklist the tracker
JobConf jConf = mr.createJobConf();
jConf.set("mapred.max.tracker.failures", "1");
jConf.setJobName("test-job-fail-once");
jConf.setMapperClass(FailOnceMapper.class);
jConf.setReducerClass(IdentityReducer.class);
jConf.setNumMapTasks(1);
jConf.setNumReduceTasks(0);
RunningJob job =
UtilsForTests.runJob(jConf, new Path(TEST_DIR, "in"), new Path(TEST_DIR, "out"));
job.waitForCompletion();
// validate the total tracker count
assertEquals(
"Active tracker count mismatch", 1, jt.getClusterStatus(false).getTaskTrackers());
// validate blacklisted count
assertEquals(
"Blacklisted tracker count mismatch",
1,
jt.getClusterStatus(false).getBlacklistedTrackers());
// find the blacklisted tracker
String trackerName = null;
for (TaskTrackerStatus status : jt.taskTrackers()) {
if (jt.isBlacklisted(status.getTrackerName())) {
trackerName = status.getTrackerName();
break;
}
}
// get the hostname
String hostToDecommission = JobInProgress.convertTrackerNameToHostName(trackerName);
LOG.info("Decommissioning tracker " + hostToDecommission);
// decommission the node
HashSet<String> decom = new HashSet<String>(1);
decom.add(hostToDecommission);
jt.decommissionNodes(decom);
// validate
// check the cluster status and tracker size
assertEquals(
"Tracker is not lost upon host decommissioning",
1,
jt.getClusterStatus(false).getTaskTrackers());
assertEquals(
"Blacklisted tracker count incorrect in cluster status " + "after decommissioning",
0,
jt.getClusterStatus(false).getBlacklistedTrackers());
assertEquals("Tracker is not lost upon host decommissioning", 1, jt.taskTrackers().size());
} finally {
if (mr != null) {
mr.shutdown();
mr = null;
jt = null;
FileUtil.fullyDelete(new File(TEST_DIR.toString()));
}
}
}
/**
* Test job retire with tasks that report their *first* status only after the job retires. Steps :
* - Start a mini-mr cluster with 1 task-tracker having only map slots. Note that this
* task-tracker will take care of setup/cleanup and the map tasks. - Submit a job with 1 map task
* and 1 reduce task - Wait for the job to finish the map task - Start a 2nd tracker that waits
* for a long time after contacting the JT. - Wait for the 2nd tracker to get stuck - Kill the job
* - Wait for the job to retire - Check if the tip mappings are cleaned up.
*/
public void testJobRetireWithUnreportedTasks() throws Exception {
MiniMRCluster mr = null;
try {
JobConf conf = new JobConf();
// set the num-map-slots to 1 so that no reduce tasks but setup/cleanup
// can run on it
conf.setInt("mapred.tasktracker.map.tasks.maximum", 1);
conf.setInt("mapred.tasktracker.reduce.tasks.maximum", 0);
mr = startCluster(conf, 1);
JobTracker jobtracker = mr.getJobTrackerRunner().getJobTracker();
RunningJob job =
UtilsForTests.runJob(
mr.createJobConf(), new Path(testDir, "in-1"), new Path(testDir, "out-1"), 1, 1);
JobID id = JobID.downgrade(job.getID());
JobInProgress jip = jobtracker.getJob(id);
// wait 100 secs for the job to complete its map task
for (int i = 0; i < 1000 && jip.finishedMaps() < 1; i++) {
UtilsForTests.waitFor(100);
}
assertEquals(jip.finishedMaps(), 1);
// start a tracker that will wait
LOG.info("Adding a waiting tracker");
TaskTrackerRunner testTrackerRunner =
mr.new TaskTrackerRunner(1, 1, null, mr.createJobConf()) {
@Override
TaskTracker createTaskTracker(JobConf conf) throws InterruptedException, IOException {
return new WaitingTaskTracker(conf);
}
};
mr.addTaskTracker(testTrackerRunner);
LOG.info("Waiting tracker added");
WaitingTaskTracker testTT = (WaitingTaskTracker) testTrackerRunner.getTaskTracker();
// wait 100 secs for the newly started task-tracker to join
for (int i = 0; i < 1000 && jobtracker.taskTrackers().size() < 2; i++) {
UtilsForTests.waitFor(100);
}
assertEquals(jobtracker.taskTrackers().size(), 2);
LOG.info("Cluster is now up with 2 trackers");
// stop the test-tt as its no longer required
mr.stopTaskTracker(mr.getTaskTrackerID(testTT.getName()));
// 1 reduce task should be scheduled
assertEquals("TestTT contacted but no reduce task scheduled on it", 1, jip.runningReduces());
// kill the job
LOG.info("Killing job " + id);
job.killJob();
// check if the reduce task attempt status is missing
TaskInProgress tip = jip.getTasks(TaskType.REDUCE)[0];
assertNull(tip.getTaskStatus(tip.getAllTaskAttemptIDs()[0]));
// wait for the job to retire
waitTillRetire(id, jobtracker);
// check the taskidToTIPMap
for (TaskAttemptID tid : jobtracker.taskidToTIPMap.keySet()) {
LOG.info("TaskidToTIP : " + tid);
}
assertEquals("'taskid' to TIP mapping still exists", 0, jobtracker.taskidToTIPMap.size());
} finally {
if (mr != null) {
mr.shutdown();
}
// cleanup
FileUtil.fullyDelete(new File(testDir.toString()));
}
}
