public void testBasicSpeculation(boolean withProgress) throws Exception {
DAG dag = DAG.create("test");
Vertex vA = Vertex.create("A", ProcessorDescriptor.create("Proc.class"), 5);
dag.addVertex(vA);
MockTezClient tezClient = createTezSession();
DAGClient dagClient = tezClient.submitDAG(dag);
DAGImpl dagImpl = (DAGImpl) mockApp.getContext().getCurrentDAG();
TezVertexID vertexId = TezVertexID.getInstance(dagImpl.getID(), 0);
// original attempt is killed and speculative one is successful
TezTaskAttemptID killedTaId =
TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexId, 0), 0);
TezTaskAttemptID successTaId =
TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexId, 0), 1);
mockLauncher.updateProgress(withProgress);
// cause speculation trigger
mockLauncher.setStatusUpdatesForTask(killedTaId, 100);
mockLauncher.startScheduling(true);
dagClient.waitForCompletion();
Assert.assertEquals(DAGStatus.State.SUCCEEDED, dagClient.getDAGStatus(null).getState());
Task task = dagImpl.getTask(killedTaId.getTaskID());
Assert.assertEquals(2, task.getAttempts().size());
Assert.assertEquals(successTaId, task.getSuccessfulAttempt().getID());
TaskAttempt killedAttempt = task.getAttempt(killedTaId);
Joiner.on(",").join(killedAttempt.getDiagnostics()).contains("Killed as speculative attempt");
Assert.assertEquals(
TaskAttemptTerminationCause.TERMINATED_EFFECTIVE_SPECULATION,
killedAttempt.getTerminationCause());
if (withProgress) {
// without progress updates occasionally more than 1 task speculates
Assert.assertEquals(
1, task.getCounters().findCounter(TaskCounter.NUM_SPECULATIONS).getValue());
Assert.assertEquals(
1, dagImpl.getAllCounters().findCounter(TaskCounter.NUM_SPECULATIONS).getValue());
org.apache.tez.dag.app.dag.Vertex v = dagImpl.getVertex(killedTaId.getTaskID().getVertexID());
Assert.assertEquals(
1, v.getAllCounters().findCounter(TaskCounter.NUM_SPECULATIONS).getValue());
}
tezClient.stop();
}
@Override
public TaskReport getReport() {
// TODO TEZPB This is broken. Records will not work without the PBImpl, which
// is in a different package.
TaskReport report = Records.newRecord(TaskReport.class);
readLock.lock();
try {
report.setTaskId(taskId);
report.setStartTime(getLaunchTime());
report.setFinishTime(getFinishTime());
report.setTaskState(getState());
report.setProgress(getProgress());
for (TaskAttempt attempt : attempts.values()) {
if (TaskAttemptState.RUNNING.equals(attempt.getState())) {
report.addRunningAttempt(attempt.getID());
}
}
report.setSuccessfulAttempt(successfulAttempt);
for (TaskAttempt att : attempts.values()) {
String prefix = "AttemptID:" + att.getID() + " Info:";
for (CharSequence cs : att.getDiagnostics()) {
report.addDiagnostics(prefix + cs);
}
}
// Add a copy of counters as the last step so that their lifetime on heap
// is as small as possible.
report.setCounters(getCounters());
return report;
} finally {
readLock.unlock();
}
}