use of org.apache.tez.dag.app.rm.container.AMContainerMap in project tez by apache.
the class TestContainerReuse method testReuseWithTaskSpecificLaunchCmdOption.
@Test(timeout = 10000l)
public void testReuseWithTaskSpecificLaunchCmdOption() throws IOException, InterruptedException, ExecutionException {
LOG.info("Test testReuseWithTaskSpecificLaunchCmdOption");
Configuration tezConf = new Configuration();
tezConf.setBoolean(TezConfiguration.TEZ_AM_CONTAINER_REUSE_ENABLED, true);
tezConf.setBoolean(TezConfiguration.TEZ_AM_CONTAINER_REUSE_RACK_FALLBACK_ENABLED, true);
tezConf.setLong(TezConfiguration.TEZ_AM_CONTAINER_REUSE_LOCALITY_DELAY_ALLOCATION_MILLIS, 0);
tezConf.setLong(TezConfiguration.TEZ_AM_CONTAINER_IDLE_RELEASE_TIMEOUT_MIN_MILLIS, 0);
tezConf.setLong(TezConfiguration.TEZ_AM_CONTAINER_IDLE_RELEASE_TIMEOUT_MAX_MILLIS, 0);
// Profile 3 tasks
tezConf.set(TezConfiguration.TEZ_TASK_SPECIFIC_LAUNCH_CMD_OPTS_LIST, "v1[1,3,4]");
tezConf.set(TezConfiguration.TEZ_TASK_SPECIFIC_LAUNCH_CMD_OPTS, "dir=/tmp/__VERTEX_NAME__/__TASK_INDEX__");
TaskSpecificLaunchCmdOption taskSpecificLaunchCmdOption = new TaskSpecificLaunchCmdOption(tezConf);
CapturingEventHandler eventHandler = new CapturingEventHandler();
TezDAGID dagID = TezDAGID.getInstance("0", 0, 0);
AMRMClient<CookieContainerRequest> rmClientCore = new AMRMClientForTest();
TezAMRMClientAsync<CookieContainerRequest> rmClient = spy(new AMRMClientAsyncForTest(rmClientCore, 100));
AppContext appContext = mock(AppContext.class);
doReturn(new Configuration(false)).when(appContext).getAMConf();
AMContainerMap amContainerMap = new AMContainerMap(mock(ContainerHeartbeatHandler.class), mock(TaskCommunicatorManagerInterface.class), new ContainerContextMatcher(), appContext);
AMNodeTracker amNodeTracker = new AMNodeTracker(eventHandler, appContext);
doReturn(amContainerMap).when(appContext).getAllContainers();
doReturn(amNodeTracker).when(appContext).getNodeTracker();
doReturn(DAGAppMasterState.RUNNING).when(appContext).getAMState();
doReturn(dagID).when(appContext).getCurrentDAGID();
doReturn(mock(ClusterInfo.class)).when(appContext).getClusterInfo();
// Use ContainerContextMatcher here. Otherwise it would not match the JVM options
TaskSchedulerManager taskSchedulerManagerReal = new TaskSchedulerManagerForTest(appContext, eventHandler, rmClient, new ContainerContextMatcher(), TezUtils.createUserPayloadFromConf(tezConf));
TaskSchedulerManager taskSchedulerManager = spy(taskSchedulerManagerReal);
taskSchedulerManager.init(tezConf);
taskSchedulerManager.start();
TaskSchedulerWithDrainableContext taskScheduler = (TaskSchedulerWithDrainableContext) ((TaskSchedulerManagerForTest) taskSchedulerManager).getSpyTaskScheduler();
TaskSchedulerContextDrainable drainableAppCallback = taskScheduler.getDrainableAppCallback();
AtomicBoolean drainNotifier = new AtomicBoolean(false);
taskScheduler.delayedContainerManager.drainedDelayedContainersForTest = drainNotifier;
Resource resource1 = Resource.newInstance(1024, 1);
String[] host1 = { "host1" };
String[] host2 = { "host2" };
String[] host3 = { "host3" };
String[] racks = { "/default-rack" };
Priority priority1 = Priority.newInstance(1);
TezVertexID vertexID1 = TezVertexID.getInstance(dagID, 1);
Map<String, LocalResource> localResources = new HashMap<String, LocalResource>();
String tsLaunchCmdOpts = taskSpecificLaunchCmdOption.getTaskSpecificOption("", "v1", 1);
/**
* Schedule 2 tasks (1 with additional launch-cmd option and another in normal mode).
* Container should not be reused in this case.
*/
// Vertex 1, Task 1, Attempt 1, host1
TezTaskAttemptID taID11 = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexID1, 1), 1);
TaskAttempt ta11 = mock(TaskAttempt.class);
AMSchedulerEventTALaunchRequest lrEvent1 = createLaunchRequestEvent(taID11, ta11, resource1, host1, racks, priority1, localResources, tsLaunchCmdOpts);
// Vertex 1, Task 2, Attempt 1, host1
TezTaskAttemptID taID12 = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexID1, 2), 1);
TaskAttempt ta12 = mock(TaskAttempt.class);
AMSchedulerEventTALaunchRequest lrEvent2 = createLaunchRequestEvent(taID12, ta12, resource1, host1, racks, priority1);
taskSchedulerManager.handleEvent(lrEvent1);
taskSchedulerManager.handleEvent(lrEvent2);
Container container1 = createContainer(1, "host1", resource1, priority1);
// One container allocated.
taskScheduler.onContainersAllocated(Collections.singletonList(container1));
TestTaskSchedulerHelpers.waitForDelayedDrainNotify(drainNotifier);
drainableAppCallback.drain();
verify(taskSchedulerManager).taskAllocated(eq(0), eq(ta11), any(Object.class), eq(container1));
// First task had profiling on. This container can not be reused further.
taskSchedulerManager.handleEvent(new AMSchedulerEventTAEnded(ta11, container1.getId(), TaskAttemptState.SUCCEEDED, null, null, 0));
drainableAppCallback.drain();
verifyDeAllocateTask(taskScheduler, ta11, true, null, null);
verify(taskSchedulerManager, times(0)).taskAllocated(eq(0), eq(ta12), any(Object.class), eq(container1));
verify(rmClient, times(1)).releaseAssignedContainer(eq(container1.getId()));
eventHandler.verifyInvocation(AMContainerEventStopRequest.class);
eventHandler.reset();
/**
* Schedule 2 tasks (both having different task specific JVM option).
* Container should not be reused.
*/
// Vertex 1, Task 3, Attempt 1, host2
tsLaunchCmdOpts = taskSpecificLaunchCmdOption.getTaskSpecificOption("", "v1", 3);
TezTaskAttemptID taID13 = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexID1, 3), 1);
TaskAttempt ta13 = mock(TaskAttempt.class);
AMSchedulerEventTALaunchRequest lrEvent3 = createLaunchRequestEvent(taID13, ta13, resource1, host2, racks, priority1, localResources, tsLaunchCmdOpts);
// Vertex 1, Task 4, Attempt 1, host2
tsLaunchCmdOpts = taskSpecificLaunchCmdOption.getTaskSpecificOption("", "v1", 4);
TezTaskAttemptID taID14 = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexID1, 4), 1);
TaskAttempt ta14 = mock(TaskAttempt.class);
AMSchedulerEventTALaunchRequest lrEvent4 = createLaunchRequestEvent(taID14, ta14, resource1, host2, racks, priority1, localResources, tsLaunchCmdOpts);
taskSchedulerManager.handleEvent(lrEvent3);
taskSchedulerManager.handleEvent(lrEvent4);
// Container started
Container container2 = createContainer(2, "host2", resource1, priority1);
taskScheduler.onContainersAllocated(Collections.singletonList(container2));
TestTaskSchedulerHelpers.waitForDelayedDrainNotify(drainNotifier);
drainableAppCallback.drain();
verify(taskSchedulerManager).taskAllocated(eq(0), eq(ta13), any(Object.class), eq(container2));
// Verify that the container can not be reused when profiling option is turned on
// Even for 2 tasks having same profiling option can have container reusability.
taskSchedulerManager.handleEvent(new AMSchedulerEventTAEnded(ta13, container2.getId(), TaskAttemptState.SUCCEEDED, null, null, 0));
drainableAppCallback.drain();
verifyDeAllocateTask(taskScheduler, ta13, true, null, null);
verify(taskSchedulerManager, times(0)).taskAllocated(eq(0), eq(ta14), any(Object.class), eq(container2));
verify(rmClient, times(1)).releaseAssignedContainer(eq(container2.getId()));
eventHandler.verifyInvocation(AMContainerEventStopRequest.class);
eventHandler.reset();
/**
* Schedule 2 tasks with same jvm profiling option.
* Container should be reused.
*/
tezConf.set(TezConfiguration.TEZ_TASK_SPECIFIC_LAUNCH_CMD_OPTS_LIST, "v1[1,2,3,5,6]");
tezConf.set(TezConfiguration.TEZ_TASK_SPECIFIC_LAUNCH_CMD_OPTS, "dummyOpts");
taskSpecificLaunchCmdOption = new TaskSpecificLaunchCmdOption(tezConf);
// Vertex 1, Task 5, Attempt 1, host3
TezTaskAttemptID taID15 = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexID1, 3), 1);
TaskAttempt ta15 = mock(TaskAttempt.class);
AMSchedulerEventTALaunchRequest lrEvent5 = createLaunchRequestEvent(taID15, ta15, resource1, host3, racks, priority1, localResources, taskSpecificLaunchCmdOption.getTaskSpecificOption("", "v1", 5));
// Vertex 1, Task 6, Attempt 1, host3
tsLaunchCmdOpts = taskSpecificLaunchCmdOption.getTaskSpecificOption("", "v1", 4);
TezTaskAttemptID taID16 = TezTaskAttemptID.getInstance(TezTaskID.getInstance(vertexID1, 4), 1);
TaskAttempt ta16 = mock(TaskAttempt.class);
AMSchedulerEventTALaunchRequest lrEvent6 = createLaunchRequestEvent(taID16, ta16, resource1, host3, racks, priority1, localResources, taskSpecificLaunchCmdOption.getTaskSpecificOption("", "v1", 6));
// Container started
Container container3 = createContainer(3, "host3", resource1, priority1);
taskSchedulerManager.handleEvent(lrEvent5);
taskSchedulerManager.handleEvent(lrEvent6);
taskScheduler.onContainersAllocated(Collections.singletonList(container3));
TestTaskSchedulerHelpers.waitForDelayedDrainNotify(drainNotifier);
drainableAppCallback.drain();
verify(taskSchedulerManager).taskAllocated(eq(0), eq(ta15), any(Object.class), eq(container3));
// Ensure task 6 (of vertex 1) is allocated to same container
taskSchedulerManager.handleEvent(new AMSchedulerEventTAEnded(ta15, container3.getId(), TaskAttemptState.SUCCEEDED, null, null, 0));
drainableAppCallback.drain();
verifyDeAllocateTask(taskScheduler, ta15, true, null, null);
verify(taskSchedulerManager).taskAllocated(eq(0), eq(ta16), any(Object.class), eq(container3));
eventHandler.reset();
taskScheduler.shutdown();
taskSchedulerManager.close();
}
use of org.apache.tez.dag.app.rm.container.AMContainerMap in project tez by apache.
the class TestAMNodeTracker method testMultipleAMNodeIDs.
@Test(timeout = 10000)
public void testMultipleAMNodeIDs() {
AppContext appContext = mock(AppContext.class);
Configuration conf = new Configuration(false);
conf.setInt(TezConfiguration.TEZ_AM_MAX_TASK_FAILURES_PER_NODE, 2);
TestEventHandler handler = new TestEventHandler();
AMNodeTracker amNodeTracker = new AMNodeTracker(handler, appContext);
doReturn(amNodeTracker).when(appContext).getNodeTracker();
AMContainerMap amContainerMap = mock(AMContainerMap.class);
TaskSchedulerManager taskSchedulerManager = mock(TaskSchedulerManager.class);
dispatcher.register(AMNodeEventType.class, amNodeTracker);
dispatcher.register(AMContainerEventType.class, amContainerMap);
dispatcher.register(AMSchedulerEventType.class, taskSchedulerManager);
amNodeTracker.init(conf);
amNodeTracker.start();
try {
amNodeTracker.nodeSeen(new ExtendedNodeId(NodeId.newInstance("host", 2222), "uuid1"), 0);
amNodeTracker.nodeSeen(new ExtendedNodeId(NodeId.newInstance("host", 2222), "uuid1"), 0);
amNodeTracker.nodeSeen(new ExtendedNodeId(NodeId.newInstance("host", 2222), "uuid2"), 0);
amNodeTracker.nodeSeen(new ExtendedNodeId(NodeId.newInstance("host", 2222), "uuid2"), 0);
assertEquals(2, amNodeTracker.getNumNodes(0));
} finally {
amNodeTracker.stop();
}
}
use of org.apache.tez.dag.app.rm.container.AMContainerMap in project tez by apache.
the class TestAMNodeTracker method testNodeSelfBlacklistAlternateScheduler2.
@Test(timeout = 10000)
public void testNodeSelfBlacklistAlternateScheduler2() {
AppContext appContext = mock(AppContext.class);
Configuration conf = new Configuration(false);
conf.setInt(TezConfiguration.TEZ_AM_MAX_TASK_FAILURES_PER_NODE, 2);
TestEventHandler handler = new TestEventHandler();
AMNodeTracker amNodeTracker = new AMNodeTracker(handler, appContext);
doReturn(amNodeTracker).when(appContext).getNodeTracker();
AMContainerMap amContainerMap = mock(AMContainerMap.class);
TaskSchedulerManager taskSchedulerManager = mock(TaskSchedulerManager.class);
dispatcher.register(AMNodeEventType.class, amNodeTracker);
dispatcher.register(AMContainerEventType.class, amContainerMap);
dispatcher.register(AMSchedulerEventType.class, taskSchedulerManager);
amNodeTracker.init(conf);
amNodeTracker.start();
try {
// This should not affect the blacklisting behaviour
for (int i = 0; i < 100; i++) {
amNodeTracker.nodeSeen(NodeId.newInstance("fakenode" + i, 3333), 0);
}
_testNodeSelfBlacklist(amNodeTracker, handler, 1);
assertFalse(amNodeTracker.isBlacklistingIgnored(0));
} finally {
amNodeTracker.stop();
}
}
use of org.apache.tez.dag.app.rm.container.AMContainerMap in project tez by apache.
the class TestAMNodeTracker method testSingleNodeNotBlacklistedAlternateScheduler.
@Test(timeout = 5000)
public void testSingleNodeNotBlacklistedAlternateScheduler() {
AppContext appContext = mock(AppContext.class);
Configuration conf = new Configuration(false);
conf.setInt(TezConfiguration.TEZ_AM_MAX_TASK_FAILURES_PER_NODE, 2);
conf.setBoolean(TezConfiguration.TEZ_AM_NODE_BLACKLISTING_ENABLED, true);
conf.setInt(TezConfiguration.TEZ_AM_NODE_BLACKLISTING_IGNORE_THRESHOLD, 33);
TestEventHandler handler = new TestEventHandler();
AMNodeTracker amNodeTracker = new AMNodeTracker(handler, appContext);
doReturn(amNodeTracker).when(appContext).getNodeTracker();
AMContainerMap amContainerMap = mock(AMContainerMap.class);
TaskSchedulerManager taskSchedulerManager = mock(TaskSchedulerManager.class);
dispatcher.register(AMNodeEventType.class, amNodeTracker);
dispatcher.register(AMContainerEventType.class, amContainerMap);
dispatcher.register(AMSchedulerEventType.class, taskSchedulerManager);
amNodeTracker.init(conf);
amNodeTracker.start();
_testSingleNodeNotBlacklisted(amNodeTracker, handler, 1);
}
use of org.apache.tez.dag.app.rm.container.AMContainerMap in project tez by apache.
the class TestAMNodeTracker method testSingleNodeNotBlacklisted.
@Test(timeout = 5000)
public void testSingleNodeNotBlacklisted() {
AppContext appContext = mock(AppContext.class);
Configuration conf = new Configuration(false);
conf.setInt(TezConfiguration.TEZ_AM_MAX_TASK_FAILURES_PER_NODE, 2);
conf.setBoolean(TezConfiguration.TEZ_AM_NODE_BLACKLISTING_ENABLED, true);
conf.setInt(TezConfiguration.TEZ_AM_NODE_BLACKLISTING_IGNORE_THRESHOLD, 33);
TestEventHandler handler = new TestEventHandler();
AMNodeTracker amNodeTracker = new AMNodeTracker(handler, appContext);
doReturn(amNodeTracker).when(appContext).getNodeTracker();
AMContainerMap amContainerMap = mock(AMContainerMap.class);
TaskSchedulerManager taskSchedulerManager = mock(TaskSchedulerManager.class);
dispatcher.register(AMNodeEventType.class, amNodeTracker);
dispatcher.register(AMContainerEventType.class, amContainerMap);
dispatcher.register(AMSchedulerEventType.class, taskSchedulerManager);
amNodeTracker.init(conf);
amNodeTracker.start();
_testSingleNodeNotBlacklisted(amNodeTracker, handler, 0);
}
Aggregations