@Before
public void setUp() {
Configuration conf = new Configuration();
// Dispatcher that processes events inline
Dispatcher dispatcher = new InlineDispatcher();
dispatcher.register(RMNodeEventType.class, new TestRMNodeEventDispatcher());
RMContext context =
new RMContextImpl(dispatcher, null, null, null, null, null, null, null, null, null);
dispatcher.register(SchedulerEventType.class, new InlineDispatcher.EmptyEventHandler());
dispatcher.register(RMNodeEventType.class, new NodeEventDispatcher(context));
NMLivelinessMonitor nmLivelinessMonitor = new NMLivelinessMonitor(dispatcher);
nmLivelinessMonitor.init(conf);
nmLivelinessMonitor.start();
NodesListManager nodesListManager = new NodesListManager(context);
nodesListManager.init(conf);
RMContainerTokenSecretManager containerTokenSecretManager =
new RMContainerTokenSecretManager(conf);
containerTokenSecretManager.start();
NMTokenSecretManagerInRM nmTokenSecretManager = new NMTokenSecretManagerInRM(conf);
nmTokenSecretManager.start();
resourceTrackerService =
new ResourceTrackerService(
context,
nodesListManager,
nmLivelinessMonitor,
containerTokenSecretManager,
nmTokenSecretManager);
resourceTrackerService.init(conf);
resourceTrackerService.start();
}
private static StubbedJob createStubbedJob(
Configuration conf, Dispatcher dispatcher, int numSplits, AppContext appContext) {
JobID jobID = JobID.forName("job_1234567890000_0001");
JobId jobId = TypeConverter.toYarn(jobID);
if (appContext == null) {
appContext = mock(AppContext.class);
when(appContext.hasSuccessfullyUnregistered()).thenReturn(true);
}
StubbedJob job =
new StubbedJob(
jobId,
ApplicationAttemptId.newInstance(ApplicationId.newInstance(0, 0), 0),
conf,
dispatcher.getEventHandler(),
true,
"somebody",
numSplits,
appContext);
dispatcher.register(JobEventType.class, job);
EventHandler mockHandler = mock(EventHandler.class);
dispatcher.register(TaskEventType.class, mockHandler);
dispatcher.register(org.apache.hadoop.mapreduce.jobhistory.EventType.class, mockHandler);
dispatcher.register(JobFinishEvent.Type.class, mockHandler);
return job;
}
private static CommitterEventHandler createCommitterEventHandler(
Dispatcher dispatcher, OutputCommitter committer) {
final SystemClock clock = new SystemClock();
AppContext appContext = mock(AppContext.class);
when(appContext.getEventHandler()).thenReturn(dispatcher.getEventHandler());
when(appContext.getClock()).thenReturn(clock);
RMHeartbeatHandler heartbeatHandler =
new RMHeartbeatHandler() {
@Override
public long getLastHeartbeatTime() {
return clock.getTime();
}
@Override
public void runOnNextHeartbeat(Runnable callback) {
callback.run();
}
};
ApplicationAttemptId id =
ConverterUtils.toApplicationAttemptId("appattempt_1234567890000_0001_0");
when(appContext.getApplicationID()).thenReturn(id.getApplicationId());
when(appContext.getApplicationAttemptId()).thenReturn(id);
CommitterEventHandler handler =
new CommitterEventHandler(appContext, committer, heartbeatHandler);
dispatcher.register(CommitterEventType.class, handler);
return handler;
}
@SuppressWarnings("unchecked")
protected void internalRemoveFromClusterNodeLabels(Collection<String> labelsToRemove) {
// remove labels from nodes
for (Map.Entry<String, Host> nodeEntry : nodeCollections.entrySet()) {
Host host = nodeEntry.getValue();
if (null != host) {
host.labels.removeAll(labelsToRemove);
for (Node nm : host.nms.values()) {
if (nm.labels != null) {
nm.labels.removeAll(labelsToRemove);
}
}
}
}
// remove labels from node labels collection
for (String label : labelsToRemove) {
labelCollections.remove(label);
}
// create event to remove labels
if (null != dispatcher) {
dispatcher.getEventHandler().handle(new RemoveClusterNodeLabels(labelsToRemove));
}
LOG.info("Remove labels: [" + StringUtils.join(labelsToRemove.iterator(), ",") + "]");
}
@SuppressWarnings("unchecked")
public void addToCluserNodeLabels(Collection<NodeLabel> labels) throws IOException {
if (!nodeLabelsEnabled) {
LOG.error(NODE_LABELS_NOT_ENABLED_ERR);
throw new IOException(NODE_LABELS_NOT_ENABLED_ERR);
}
if (null == labels || labels.isEmpty()) {
return;
}
List<NodeLabel> newLabels = new ArrayList<NodeLabel>();
normalizeNodeLabels(labels);
// do a check before actual adding them, will throw exception if any of them
// doesn't meet label name requirement
for (NodeLabel label : labels) {
checkAndThrowLabelName(label.getName());
}
for (NodeLabel label : labels) {
// shouldn't overwrite it to avoid changing the Label.resource
if (this.labelCollections.get(label.getName()) == null) {
this.labelCollections.put(label.getName(), new RMNodeLabel(label));
newLabels.add(label);
}
}
if (null != dispatcher && !newLabels.isEmpty()) {
dispatcher.getEventHandler().handle(new StoreNewClusterNodeLabels(newLabels));
}
LOG.info("Add labels: [" + StringUtils.join(labels.iterator(), ",") + "]");
}
/** Create and initialize (but don't start) a single job. */
protected Job createJob(Configuration conf) {
// create single job
Job newJob = new JobInAppMaster(jobId, conf, context);
((RunningAppContext) context).jobs.put(newJob.getID(), newJob);
dispatcher.register(JobFinishEvent.Type.class, createJobFinishEventHandler());
return newJob;
}
/** Create and initialize (but don't start) a single dag. */
protected DAG createDAG(DAGPlan dagPB) {
TezDAGID dagId = new TezDAGID(appAttemptID.getApplicationId(), dagCounter.incrementAndGet());
Iterator<PlanKeyValuePair> iter = dagPB.getDagKeyValues().getConfKeyValuesList().iterator();
Configuration dagConf = new Configuration(amConf);
while (iter.hasNext()) {
PlanKeyValuePair keyValPair = iter.next();
dagConf.set(keyValPair.getKey(), keyValPair.getValue());
}
// create single dag
DAG newDag =
new DAGImpl(
dagId,
dagConf,
dagPB,
dispatcher.getEventHandler(),
taskAttemptListener,
jobTokenSecretManager,
fsTokens,
clock,
currentUser.getShortUserName(),
taskHeartbeatHandler,
context);
return newDag;
} // end createDag()
@SuppressWarnings({"fallthrough", "unchecked"})
private void finished() {
ApplicationId applicationId = containerId.getApplicationAttemptId().getApplicationId();
switch (getContainerState()) {
case EXITED_WITH_SUCCESS:
metrics.endRunningContainer();
metrics.completedContainer();
NMAuditLogger.logSuccess(
user,
AuditConstants.FINISH_SUCCESS_CONTAINER,
"ContainerImpl",
applicationId,
containerId);
break;
case EXITED_WITH_FAILURE:
if (wasLaunched) {
metrics.endRunningContainer();
}
// fall through
case LOCALIZATION_FAILED:
metrics.failedContainer();
NMAuditLogger.logFailure(
user,
AuditConstants.FINISH_FAILED_CONTAINER,
"ContainerImpl",
"Container failed with state: " + getContainerState(),
applicationId,
containerId);
break;
case CONTAINER_CLEANEDUP_AFTER_KILL:
if (wasLaunched) {
metrics.endRunningContainer();
}
// fall through
case NEW:
metrics.killedContainer();
NMAuditLogger.logSuccess(
user,
AuditConstants.FINISH_KILLED_CONTAINER,
"ContainerImpl",
applicationId,
containerId);
}
metrics.releaseContainer(this.resource);
// Inform the application
@SuppressWarnings("rawtypes")
EventHandler eventHandler = dispatcher.getEventHandler();
eventHandler.handle(new ApplicationContainerFinishedEvent(containerId));
// Remove the container from the resource-monitor
eventHandler.handle(new ContainerStopMonitoringEvent(containerId));
// Tell the logService too
eventHandler.handle(new LogHandlerContainerFinishedEvent(containerId, exitCode));
}
@SuppressWarnings("unchecked") // dispatcher not typed
public void cleanup() {
Map<LocalResourceVisibility, Collection<LocalResourceRequest>> rsrc =
new HashMap<LocalResourceVisibility, Collection<LocalResourceRequest>>();
if (!publicRsrcs.isEmpty()) {
rsrc.put(LocalResourceVisibility.PUBLIC, publicRsrcs);
}
if (!privateRsrcs.isEmpty()) {
rsrc.put(LocalResourceVisibility.PRIVATE, privateRsrcs);
}
if (!appRsrcs.isEmpty()) {
rsrc.put(LocalResourceVisibility.APPLICATION, appRsrcs);
}
dispatcher.getEventHandler().handle(new ContainerLocalizationCleanupEvent(this, rsrc));
}
@Override
protected void serviceStart() throws Exception {
if (nodeLabelsEnabled) {
initNodeLabelStore(getConfig());
}
// init dispatcher only when service start, because recover will happen in
// service init, we don't want to trigger any event handling at that time.
initDispatcher(getConfig());
if (null != dispatcher) {
dispatcher.register(NodeLabelsStoreEventType.class, new ForwardingEventHandler());
}
startDispatcher();
}
@SuppressWarnings("unchecked")
@Override
public void serviceStart() throws Exception {
// start all the components
startServices();
super.serviceStart();
this.state = DAGAppMasterState.IDLE;
// metrics system init is really init & start.
// It's more test friendly to put it here.
DefaultMetricsSystem.initialize("DAGAppMaster");
this.appsStartTime = clock.getTime();
AMStartedEvent startEvent =
new AMStartedEvent(appAttemptID, startTime, appsStartTime, appSubmitTime);
dispatcher.getEventHandler().handle(new DAGHistoryEvent(startEvent));
this.lastDAGCompletionTime = clock.getTime();
if (!isSession) {
startDAG();
} else {
LOG.info("In Session mode. Waiting for DAG over RPC");
this.dagSubmissionTimer = new Timer(true);
this.dagSubmissionTimer.scheduleAtFixedRate(
new TimerTask() {
@Override
public void run() {
checkAndHandleSessionTimeout();
}
},
sessionTimeoutInterval,
sessionTimeoutInterval / 10);
}
}
@SuppressWarnings("unchecked")
protected void internalUpdateLabelsOnNodes(
Map<NodeId, Set<String>> nodeToLabels, NodeLabelUpdateOperation op) throws IOException {
// do update labels from nodes
Map<NodeId, Set<String>> newNMToLabels = new HashMap<NodeId, Set<String>>();
Set<String> oldLabels;
for (Entry<NodeId, Set<String>> entry : nodeToLabels.entrySet()) {
NodeId nodeId = entry.getKey();
Set<String> labels = entry.getValue();
createHostIfNonExisted(nodeId.getHost());
if (nodeId.getPort() == WILDCARD_PORT) {
Host host = nodeCollections.get(nodeId.getHost());
switch (op) {
case REMOVE:
removeNodeFromLabels(nodeId, labels);
host.labels.removeAll(labels);
for (Node node : host.nms.values()) {
if (node.labels != null) {
node.labels.removeAll(labels);
}
removeNodeFromLabels(node.nodeId, labels);
}
break;
case ADD:
addNodeToLabels(nodeId, labels);
host.labels.addAll(labels);
for (Node node : host.nms.values()) {
if (node.labels != null) {
node.labels.addAll(labels);
}
addNodeToLabels(node.nodeId, labels);
}
break;
case REPLACE:
replaceNodeForLabels(nodeId, host.labels, labels);
host.labels.clear();
host.labels.addAll(labels);
for (Node node : host.nms.values()) {
replaceNodeForLabels(node.nodeId, node.labels, labels);
node.labels = null;
}
break;
default:
break;
}
newNMToLabels.put(nodeId, host.labels);
} else {
if (EnumSet.of(NodeLabelUpdateOperation.ADD, NodeLabelUpdateOperation.REPLACE)
.contains(op)) {
// Add and replace
createNodeIfNonExisted(nodeId);
Node nm = getNMInNodeSet(nodeId);
switch (op) {
case ADD:
addNodeToLabels(nodeId, labels);
if (nm.labels == null) {
nm.labels = new HashSet<String>();
}
nm.labels.addAll(labels);
break;
case REPLACE:
oldLabels = getLabelsByNode(nodeId);
replaceNodeForLabels(nodeId, oldLabels, labels);
if (nm.labels == null) {
nm.labels = new HashSet<String>();
}
nm.labels.clear();
nm.labels.addAll(labels);
break;
default:
break;
}
newNMToLabels.put(nodeId, nm.labels);
} else {
// remove
removeNodeFromLabels(nodeId, labels);
Node nm = getNMInNodeSet(nodeId);
if (nm.labels != null) {
nm.labels.removeAll(labels);
newNMToLabels.put(nodeId, nm.labels);
}
}
}
}
if (null != dispatcher && !isDistributedNodeLabelConfiguration) {
// In case of DistributedNodeLabelConfiguration, no need to save the the
// NodeLabels Mapping to the back-end store, as on RM restart/failover
// NodeLabels are collected from NM through Register/Heartbeat again
dispatcher.getEventHandler().handle(new UpdateNodeToLabelsMappingsEvent(newNMToLabels));
}
// shows node->labels we added
LOG.info(op.name() + " labels on nodes:");
for (Entry<NodeId, Set<String>> entry : newNMToLabels.entrySet()) {
LOG.info(
" NM="
+ entry.getKey()
+ ", labels=["
+ StringUtils.join(entry.getValue().iterator(), ",")
+ "]");
}
}
@Override
public void init(final Configuration conf) {
// Get all needed security tokens.
downloadTokensAndSetupUGI(conf);
// Initialize application context,name,attemptId,jobId
context = new RunningAppContext();
appName = conf.get(DragonJobConfig.JOB_NAME, "<missing app name>");
conf.setInt(DragonJobConfig.APPLICATION_ATTEMPT_ID, appAttemptId.getAttemptId());
jobId =
JobId.newJobId(appAttemptId.getApplicationId(), appAttemptId.getApplicationId().getId());
// service to hand out event
dispatcher = createDispatcher();
addIfService(dispatcher);
// service to handle requests from JobClient
clientService = createClientService(context);
addIfService(clientService);
// service to log job history events
EventHandler<JobHistoryEvent> historyHandlerService = createJobHistoryHandler(context);
dispatcher.register(
org.apache.hadoop.realtime.jobhistory.EventType.class, historyHandlerService);
// Initialize the JobEventDispatcher
this.jobEventDispatcher = new JobEventDispatcher();
// register the event dispatchers
dispatcher.register(JobEventType.class, jobEventDispatcher);
dispatcher.register(TaskEventType.class, new TaskEventDispatcher());
dispatcher.register(TaskAttemptEventType.class, new TaskAttemptEventDispatcher());
// service to handle requests to TaskUmbilicalProtocol
childService = createChildService(context);
addIfService(childService);
dispatcher.register(ChildExecutionEventType.class, childService);
// service to allocate containers from RM (if non-uber) or to fake it (uber)
containerAllocator = createContainerAllocator(clientService, context);
addIfService(containerAllocator);
dispatcher.register(ContainerAllocator.EventType.class, containerAllocator);
// corresponding service to launch allocated containers via NodeManager
containerLauncher = createContainerLauncher(context);
addIfService(containerLauncher);
dispatcher.register(ContainerLauncher.EventType.class, containerLauncher);
// dragon zookeeper service
zkService = new DragonZKService(context);
addIfService(zkService);
dispatcher.register(ZKEventType.class, zkService);
// Add the JobHistoryEventHandler last so that it is properly stopped first.
// This will guarantee that all history-events are flushed before AM goes
// ahead with shutdown.
// Note: Even though JobHistoryEventHandler is started last, if any
// component creates a JobHistoryEvent in the meanwhile, it will be just be
// queued inside the JobHistoryEventHandler
addIfService(historyHandlerService);
super.init(conf);
}
@Override
public void serviceInit(final Configuration conf) throws Exception {
this.state = DAGAppMasterState.INITED;
this.amConf = conf;
conf.setBoolean(Dispatcher.DISPATCHER_EXIT_ON_ERROR_KEY, true);
downloadTokensAndSetupUGI(conf);
context = new RunningAppContext(conf);
clientHandler = new DAGClientHandler();
dispatcher = createDispatcher();
addIfService(dispatcher, false);
clientRpcServer = new DAGClientServer(clientHandler);
addIfService(clientRpcServer, true);
taskHeartbeatHandler = createTaskHeartbeatHandler(context, conf);
addIfService(taskHeartbeatHandler, true);
containerHeartbeatHandler = createContainerHeartbeatHandler(context, conf);
addIfService(containerHeartbeatHandler, true);
// service to handle requests to TaskUmbilicalProtocol
taskAttemptListener =
createTaskAttemptListener(context, taskHeartbeatHandler, containerHeartbeatHandler);
addIfService(taskAttemptListener, true);
containers = new AMContainerMap(containerHeartbeatHandler, taskAttemptListener, context);
addIfService(containers, true);
dispatcher.register(AMContainerEventType.class, containers);
nodes = new AMNodeMap(dispatcher.getEventHandler(), context);
addIfService(nodes, true);
dispatcher.register(AMNodeEventType.class, nodes);
// service to do the task cleanup
taskCleaner = createTaskCleaner(context);
addIfService(taskCleaner, true);
this.dagEventDispatcher = new DagEventDispatcher();
this.vertexEventDispatcher = new VertexEventDispatcher();
// register the event dispatchers
dispatcher.register(DAGAppMasterEventType.class, new DAGAppMasterEventHandler());
dispatcher.register(DAGEventType.class, dagEventDispatcher);
dispatcher.register(VertexEventType.class, vertexEventDispatcher);
dispatcher.register(TaskEventType.class, new TaskEventDispatcher());
dispatcher.register(TaskAttemptEventType.class, new TaskAttemptEventDispatcher());
dispatcher.register(TaskCleaner.EventType.class, taskCleaner);
taskSchedulerEventHandler =
new TaskSchedulerEventHandler(context, clientRpcServer, dispatcher.getEventHandler());
addIfService(taskSchedulerEventHandler, true);
dispatcher.register(AMSchedulerEventType.class, taskSchedulerEventHandler);
addIfServiceDependency(taskSchedulerEventHandler, clientRpcServer);
containerLauncher = createContainerLauncher(context);
addIfService(containerLauncher, true);
dispatcher.register(NMCommunicatorEventType.class, containerLauncher);
historyEventHandler = new HistoryEventHandler(context);
addIfService(historyEventHandler, true);
dispatcher.register(HistoryEventType.class, historyEventHandler);
this.sessionTimeoutInterval =
1000
* amConf.getInt(
TezConfiguration.TEZ_SESSION_AM_DAG_SUBMIT_TIMEOUT_SECS,
TezConfiguration.TEZ_SESSION_AM_DAG_SUBMIT_TIMEOUT_SECS_DEFAULT);
initServices(conf);
super.serviceInit(conf);
}
@SuppressWarnings("unchecked")
private void sendEvent(Event<?> event) {
dispatcher.getEventHandler().handle(event);
}