@Override
public synchronized void start() {
this.containerAllocator = new RMContainerAllocator(this.context);
((Service) this.containerAllocator).init(getConfig());
((Service) this.containerAllocator).start();
super.start();
}
@Override
public synchronized void start() {
this.containerLauncher = new ContainerLauncherImpl(context);
((Service) this.containerLauncher).init(getConfig());
((Service) this.containerLauncher).start();
super.start();
}
@Override
public void start() {
// NodeManager is the last service to start, so NodeId is available.
this.nodeId = this.context.getNodeId();
String httpBindAddressStr =
getConfig()
.get(YarnConfiguration.NM_WEBAPP_ADDRESS, YarnConfiguration.DEFAULT_NM_WEBAPP_ADDRESS);
InetSocketAddress httpBindAddress =
NetUtils.createSocketAddr(
httpBindAddressStr,
YarnConfiguration.DEFAULT_NM_WEBAPP_PORT,
YarnConfiguration.NM_WEBAPP_ADDRESS);
try {
// this.hostName = InetAddress.getLocalHost().getCanonicalHostName();
this.httpPort = httpBindAddress.getPort();
// Registration has to be in start so that ContainerManager can get the
// perNM tokens needed to authenticate ContainerTokens.
registerWithRM();
super.start();
startStatusUpdater();
} catch (Exception e) {
throw new AvroRuntimeException(e);
}
}
@Override
public synchronized void stop() {
try {
accepted.close().awaitUninterruptibly(10, TimeUnit.SECONDS);
ServerBootstrap bootstrap = new ServerBootstrap(selector);
bootstrap.releaseExternalResources();
pipelineFact.destroy();
localFS.close();
} catch (Throwable t) {
LOG.error(t);
} finally {
super.stop();
}
}
@Override
public synchronized void init(Configuration conf) {
this.rmAddress =
conf.get(
YarnConfiguration.RM_RESOURCE_TRACKER_ADDRESS,
YarnConfiguration.DEFAULT_RM_RESOURCE_TRACKER_ADDRESS);
this.heartBeatInterval =
conf.getLong(
YarnConfiguration.NM_TO_RM_HEARTBEAT_INTERVAL_MS,
YarnConfiguration.DEFAULT_NM_TO_RM_HEARTBEAT_INTERVAL_MS);
int memoryMb = conf.getInt(YarnConfiguration.NM_PMEM_MB, YarnConfiguration.DEFAULT_NM_PMEM_MB);
this.totalResource = recordFactory.newRecordInstance(Resource.class);
this.totalResource.setMemory(memoryMb);
metrics.addResource(totalResource);
super.init(conf);
}
@Override
public void start() {
eventHandlingThread =
new Thread(
new Runnable() {
@Override
public void run() {
JobHistoryEvent event = null;
while (!stopped && !Thread.currentThread().isInterrupted()) {
// Log the size of the history-event-queue every so often.
if (eventCounter != 0 && eventCounter % 1000 == 0) {
eventCounter = 0;
LOG.info("Size of the JobHistory event queue is " + eventQueue.size());
} else {
eventCounter++;
}
try {
event = eventQueue.take();
} catch (InterruptedException e) {
LOG.info("EventQueue take interrupted. Returning");
return;
}
// If an event has been removed from the queue. Handle it.
// The rest of the queue is handled via stop()
// Clear the interrupt status if it's set before calling handleEvent
// and set it if it was set before calling handleEvent.
// Interrupts received from other threads during handleEvent cannot be
// dealth with - Shell.runCommand() ignores them.
synchronized (lock) {
boolean isInterrupted = Thread.interrupted();
handleEvent(event);
if (isInterrupted) {
Thread.currentThread().interrupt();
}
}
}
}
});
eventHandlingThread.start();
super.start();
}
// TODO change AbstractService to throw InterruptedException
@Override
public synchronized void start() {
Configuration conf = getConfig();
ServerBootstrap bootstrap = new ServerBootstrap(selector);
try {
pipelineFact = new HttpPipelineFactory(conf);
} catch (Exception ex) {
throw new RuntimeException(ex);
}
bootstrap.setPipelineFactory(pipelineFact);
port = conf.getInt(ConfVars.PULLSERVER_PORT.varname, ConfVars.PULLSERVER_PORT.defaultIntVal);
Channel ch = bootstrap.bind(new InetSocketAddress(port));
accepted.add(ch);
port = ((InetSocketAddress) ch.getLocalAddress()).getPort();
conf.set(ConfVars.PULLSERVER_PORT.varname, Integer.toString(port));
pipelineFact.PullServer.setPort(port);
LOG.info(getName() + " listening on port " + port);
super.start();
sslFileBufferSize =
conf.getInt(SUFFLE_SSL_FILE_BUFFER_SIZE_KEY, DEFAULT_SUFFLE_SSL_FILE_BUFFER_SIZE);
}
@Override
public synchronized void init(Configuration conf) {
try {
manageOsCache = conf.getBoolean(SHUFFLE_MANAGE_OS_CACHE, DEFAULT_SHUFFLE_MANAGE_OS_CACHE);
readaheadLength = conf.getInt(SHUFFLE_READAHEAD_BYTES, DEFAULT_SHUFFLE_READAHEAD_BYTES);
ThreadFactory bossFactory =
new ThreadFactoryBuilder().setNameFormat("PullServerAuxService Netty Boss #%d").build();
ThreadFactory workerFactory =
new ThreadFactoryBuilder().setNameFormat("PullServerAuxService Netty Worker #%d").build();
selector =
new NioServerSocketChannelFactory(
Executors.newCachedThreadPool(bossFactory),
Executors.newCachedThreadPool(workerFactory));
localFS = new LocalFileSystem();
super.init(new Configuration(conf));
} catch (Throwable t) {
LOG.error(t);
}
}
@Override
public synchronized void stop() {
((Service) this.containerLauncher).stop();
super.stop();
}
@Override
public synchronized void stop() {
// Interrupt the updater.
this.isStopped = true;
super.stop();
}
@Override
public void stop() {
LOG.info(
"Stopping JobHistoryEventHandler. "
+ "Size of the outstanding queue size is "
+ eventQueue.size());
stopped = true;
// do not interrupt while event handling is in progress
synchronized (lock) {
if (eventHandlingThread != null) eventHandlingThread.interrupt();
}
try {
if (eventHandlingThread != null) eventHandlingThread.join();
} catch (InterruptedException ie) {
LOG.info("Interruped Exception while stopping", ie);
}
// Cancel all timers - so that they aren't invoked during or after
// the metaInfo object is wrapped up.
for (MetaInfo mi : fileMap.values()) {
try {
mi.shutDownTimer();
} catch (IOException e) {
LOG.info(
"Exception while cancelling delayed flush timer. "
+ "Likely caused by a failed flush "
+ e.getMessage());
}
}
// write all the events remaining in queue
Iterator<JobHistoryEvent> it = eventQueue.iterator();
while (it.hasNext()) {
JobHistoryEvent ev = it.next();
LOG.info("In stop, writing event " + ev.getType());
handleEvent(ev);
}
// Process JobUnsuccessfulCompletionEvent for jobIds which still haven't
// closed their event writers
Iterator<JobId> jobIt = fileMap.keySet().iterator();
if (isSignalled) {
while (jobIt.hasNext()) {
JobId toClose = jobIt.next();
MetaInfo mi = fileMap.get(toClose);
if (mi != null && mi.isWriterActive()) {
LOG.warn("Found jobId " + toClose + " to have not been closed. Will close");
// Create a JobFinishEvent so that it is written to the job history
JobUnsuccessfulCompletionEvent jucEvent =
new JobUnsuccessfulCompletionEvent(
TypeConverter.fromYarn(toClose),
System.currentTimeMillis(),
context.getJob(toClose).getCompletedMaps(),
context.getJob(toClose).getCompletedReduces(),
JobState.KILLED.toString());
JobHistoryEvent jfEvent = new JobHistoryEvent(toClose, jucEvent);
// Bypass the queue mechanism which might wait. Call the method directly
handleEvent(jfEvent);
}
}
}
// close all file handles
for (MetaInfo mi : fileMap.values()) {
try {
mi.closeWriter();
} catch (IOException e) {
LOG.info("Exception while closing file " + e.getMessage());
}
}
LOG.info("Stopped JobHistoryEventHandler. super.stop()");
super.stop();
}
/* (non-Javadoc)
* @see org.apache.hadoop.yarn.service.AbstractService#init(org.
* apache.hadoop.conf.Configuration)
* Initializes the FileSystem and Path objects for the log and done directories.
* Creates these directories if they do not already exist.
*/
@Override
public void init(Configuration conf) {
String stagingDirStr = null;
String doneDirStr = null;
String userDoneDirStr = null;
try {
stagingDirStr = JobHistoryUtils.getConfiguredHistoryStagingDirPrefix(conf);
doneDirStr = JobHistoryUtils.getConfiguredHistoryIntermediateDoneDirPrefix(conf);
userDoneDirStr = JobHistoryUtils.getHistoryIntermediateDoneDirForUser(conf);
} catch (IOException e) {
LOG.error("Failed while getting the configured log directories", e);
throw new YarnException(e);
}
// Check for the existence of the history staging dir. Maybe create it.
try {
stagingDirPath = FileSystem.get(conf).makeQualified(new Path(stagingDirStr));
stagingDirFS = FileSystem.get(stagingDirPath.toUri(), conf);
mkdir(
stagingDirFS,
stagingDirPath,
new FsPermission(JobHistoryUtils.HISTORY_STAGING_DIR_PERMISSIONS));
} catch (IOException e) {
LOG.error(
"Failed while checking for/creating history staging path: [" + stagingDirPath + "]", e);
throw new YarnException(e);
}
// Check for the existence of intermediate done dir.
Path doneDirPath = null;
try {
doneDirPath = FileSystem.get(conf).makeQualified(new Path(doneDirStr));
doneDirFS = FileSystem.get(doneDirPath.toUri(), conf);
// This directory will be in a common location, or this may be a cluster
// meant for a single user. Creating based on the conf. Should ideally be
// created by the JobHistoryServer or as part of deployment.
if (!doneDirFS.exists(doneDirPath)) {
if (JobHistoryUtils.shouldCreateNonUserDirectory(conf)) {
LOG.info(
"Creating intermediate history logDir: ["
+ doneDirPath
+ "] + based on conf. Should ideally be created by the JobHistoryServer: "
+ MRJobConfig.MR_AM_CREATE_JH_INTERMEDIATE_BASE_DIR);
mkdir(
doneDirFS,
doneDirPath,
new FsPermission(
JobHistoryUtils.HISTORY_INTERMEDIATE_DONE_DIR_PERMISSIONS.toShort()));
// TODO Temporary toShort till new FsPermission(FsPermissions)
// respects
// sticky
} else {
String message =
"Not creating intermediate history logDir: ["
+ doneDirPath
+ "] based on conf: "
+ MRJobConfig.MR_AM_CREATE_JH_INTERMEDIATE_BASE_DIR
+ ". Either set to true or pre-create this directory with"
+ " appropriate permissions";
LOG.error(message);
throw new YarnException(message);
}
}
} catch (IOException e) {
LOG.error(
"Failed checking for the existance of history intermediate "
+ "done directory: ["
+ doneDirPath
+ "]");
throw new YarnException(e);
}
// Check/create user directory under intermediate done dir.
try {
doneDirPrefixPath = FileSystem.get(conf).makeQualified(new Path(userDoneDirStr));
mkdir(
doneDirFS,
doneDirPrefixPath,
new FsPermission(JobHistoryUtils.HISTORY_INTERMEDIATE_USER_DIR_PERMISSIONS));
} catch (IOException e) {
LOG.error(
"Error creating user intermediate history done directory: [ " + doneDirPrefixPath + "]",
e);
throw new YarnException(e);
}
// Maximum number of unflushed completion-events that can stay in the queue
// before flush kicks in.
maxUnflushedCompletionEvents =
conf.getInt(
MRJobConfig.MR_AM_HISTORY_MAX_UNFLUSHED_COMPLETE_EVENTS,
MRJobConfig.DEFAULT_MR_AM_HISTORY_MAX_UNFLUSHED_COMPLETE_EVENTS);
// We want to cut down flushes after job completes so as to write quicker,
// so we increase maxUnflushedEvents post Job completion by using the
// following multiplier.
postJobCompletionMultiplier =
conf.getInt(
MRJobConfig.MR_AM_HISTORY_JOB_COMPLETE_UNFLUSHED_MULTIPLIER,
MRJobConfig.DEFAULT_MR_AM_HISTORY_JOB_COMPLETE_UNFLUSHED_MULTIPLIER);
// Max time until which flush doesn't take place.
flushTimeout =
conf.getLong(
MRJobConfig.MR_AM_HISTORY_COMPLETE_EVENT_FLUSH_TIMEOUT_MS,
MRJobConfig.DEFAULT_MR_AM_HISTORY_COMPLETE_EVENT_FLUSH_TIMEOUT_MS);
minQueueSizeForBatchingFlushes =
conf.getInt(
MRJobConfig.MR_AM_HISTORY_USE_BATCHED_FLUSH_QUEUE_SIZE_THRESHOLD,
MRJobConfig.DEFAULT_MR_AM_HISTORY_USE_BATCHED_FLUSH_QUEUE_SIZE_THRESHOLD);
super.init(conf);
}
