/**
* All parameters are obtained from the {@link GlobalConfiguration}, which must be loaded prior to
* instantiating the task manager.
*/
public TaskManager(ExecutionMode executionMode) throws Exception {
if (executionMode == null) {
throw new NullPointerException("Execution mode must not be null.");
}
RevisionInformation rev = JobManagerUtils.getRevisionInformation();
LOG.info(
"Starting Stratosphere TaskManager "
+ "(Version: "
+ JobManagerUtils.getVersion()
+ ", "
+ "Rev:"
+ rev.commitId
+ ", "
+ "Date:"
+ rev.commitDate
+ ")");
try {
LOG.info(
"TaskManager started as user "
+ UserGroupInformation.getCurrentUser().getShortUserName());
} catch (Throwable t) {
LOG.error("Cannot determine user group information.", t);
}
LOG.info("Execution mode: " + executionMode);
// IMPORTANT! At this point, the GlobalConfiguration must have been read!
final InetSocketAddress jobManagerAddress;
{
LOG.info("Reading location of job manager from configuration");
final String address =
GlobalConfiguration.getString(ConfigConstants.JOB_MANAGER_IPC_ADDRESS_KEY, null);
final int port =
GlobalConfiguration.getInteger(
ConfigConstants.JOB_MANAGER_IPC_PORT_KEY,
ConfigConstants.DEFAULT_JOB_MANAGER_IPC_PORT);
if (address == null) {
throw new Exception("Job manager address not configured in the GlobalConfiguration.");
}
// Try to convert configured address to {@link InetAddress}
try {
final InetAddress tmpAddress = InetAddress.getByName(address);
jobManagerAddress = new InetSocketAddress(tmpAddress, port);
} catch (UnknownHostException e) {
LOG.fatal("Could not resolve JobManager host name.");
throw new Exception("Could not resolve JobManager host name: " + e.getMessage(), e);
}
LOG.info("Connecting to JobManager at: " + jobManagerAddress);
}
// Create RPC connection to the JobManager
try {
this.jobManager =
RPC.getProxy(JobManagerProtocol.class, jobManagerAddress, NetUtils.getSocketFactory());
} catch (IOException e) {
LOG.fatal("Could not connect to the JobManager: " + e.getMessage(), e);
throw new Exception("Failed to initialize connection to JobManager: " + e.getMessage(), e);
}
int ipcPort = GlobalConfiguration.getInteger(ConfigConstants.TASK_MANAGER_IPC_PORT_KEY, -1);
int dataPort = GlobalConfiguration.getInteger(ConfigConstants.TASK_MANAGER_DATA_PORT_KEY, -1);
if (ipcPort == -1) {
ipcPort = getAvailablePort();
}
if (dataPort == -1) {
dataPort = getAvailablePort();
}
// Determine our own public facing address and start the server
{
final InetAddress taskManagerAddress;
try {
taskManagerAddress = getTaskManagerAddress(jobManagerAddress);
} catch (Exception e) {
throw new RuntimeException(
"The TaskManager failed to determine its own network address.", e);
}
this.localInstanceConnectionInfo =
new InstanceConnectionInfo(taskManagerAddress, ipcPort, dataPort);
LOG.info("TaskManager connection information:" + this.localInstanceConnectionInfo);
// Start local RPC server
try {
this.taskManagerServer =
RPC.getServer(this, taskManagerAddress.getHostAddress(), ipcPort, IPC_HANDLER_COUNT);
this.taskManagerServer.start();
} catch (IOException e) {
LOG.fatal("Failed to start TaskManager server. " + e.getMessage(), e);
throw new Exception("Failed to start taskmanager server. " + e.getMessage(), e);
}
}
// Try to create local stub of the global input split provider
try {
this.globalInputSplitProvider =
RPC.getProxy(
InputSplitProviderProtocol.class, jobManagerAddress, NetUtils.getSocketFactory());
} catch (IOException e) {
LOG.fatal(e.getMessage(), e);
throw new Exception(
"Failed to initialize connection to global input split provider: " + e.getMessage(), e);
}
// Try to create local stub for the lookup service
try {
this.lookupService =
RPC.getProxy(ChannelLookupProtocol.class, jobManagerAddress, NetUtils.getSocketFactory());
} catch (IOException e) {
LOG.fatal(e.getMessage(), e);
throw new Exception("Failed to initialize channel lookup protocol. " + e.getMessage(), e);
}
// Try to create local stub for the accumulators
try {
this.accumulatorProtocolProxy =
RPC.getProxy(AccumulatorProtocol.class, jobManagerAddress, NetUtils.getSocketFactory());
} catch (IOException e) {
LOG.fatal("Failed to initialize accumulator protocol: " + e.getMessage(), e);
throw new Exception("Failed to initialize accumulator protocol: " + e.getMessage(), e);
}
// Load profiler if it should be used
if (GlobalConfiguration.getBoolean(ProfilingUtils.ENABLE_PROFILING_KEY, false)) {
final String profilerClassName =
GlobalConfiguration.getString(
ProfilingUtils.TASKMANAGER_CLASSNAME_KEY,
"eu.stratosphere.nephele.profiling.impl.TaskManagerProfilerImpl");
this.profiler =
ProfilingUtils.loadTaskManagerProfiler(
profilerClassName, jobManagerAddress.getAddress(), this.localInstanceConnectionInfo);
if (this.profiler == null) {
LOG.error("Cannot find class name for the profiler.");
} else {
LOG.info("Profiling of jobs is enabled.");
}
} else {
this.profiler = null;
LOG.info("Profiling of jobs is disabled.");
}
// Get the directory for storing temporary files
final String[] tmpDirPaths =
GlobalConfiguration.getString(
ConfigConstants.TASK_MANAGER_TMP_DIR_KEY,
ConfigConstants.DEFAULT_TASK_MANAGER_TMP_PATH)
.split(",|" + File.pathSeparator);
checkTempDirs(tmpDirPaths);
final int pageSize =
GlobalConfiguration.getInteger(
ConfigConstants.TASK_MANAGER_NETWORK_BUFFER_SIZE_KEY,
ConfigConstants.DEFAULT_TASK_MANAGER_NETWORK_BUFFER_SIZE);
// Initialize network buffer pool
int numBuffers =
GlobalConfiguration.getInteger(
ConfigConstants.TASK_MANAGER_NETWORK_NUM_BUFFERS_KEY,
ConfigConstants.DEFAULT_TASK_MANAGER_NETWORK_NUM_BUFFERS);
int bufferSize =
GlobalConfiguration.getInteger(
ConfigConstants.TASK_MANAGER_NETWORK_BUFFER_SIZE_KEY,
ConfigConstants.DEFAULT_TASK_MANAGER_NETWORK_BUFFER_SIZE);
// Initialize the channel manager
try {
NetworkConnectionManager networkConnectionManager = null;
switch (executionMode) {
case LOCAL:
networkConnectionManager = new LocalConnectionManager();
break;
case CLUSTER:
int numInThreads =
GlobalConfiguration.getInteger(
ConfigConstants.TASK_MANAGER_NET_NUM_IN_THREADS_KEY,
ConfigConstants.DEFAULT_TASK_MANAGER_NET_NUM_IN_THREADS);
int numOutThreads =
GlobalConfiguration.getInteger(
ConfigConstants.TASK_MANAGER_NET_NUM_OUT_THREADS_KEY,
ConfigConstants.DEFAULT_TASK_MANAGER_NET_NUM_OUT_THREADS);
int lowWaterMark =
GlobalConfiguration.getInteger(
ConfigConstants.TASK_MANAGER_NET_NETTY_LOW_WATER_MARK,
ConfigConstants.DEFAULT_TASK_MANAGER_NET_NETTY_LOW_WATER_MARK);
int highWaterMark =
GlobalConfiguration.getInteger(
ConfigConstants.TASK_MANAGER_NET_NETTY_HIGH_WATER_MARK,
ConfigConstants.DEFAULT_TASK_MANAGER_NET_NETTY_HIGH_WATER_MARK);
networkConnectionManager =
new NettyConnectionManager(
localInstanceConnectionInfo.address(),
localInstanceConnectionInfo.dataPort(),
bufferSize,
numInThreads,
numOutThreads,
lowWaterMark,
highWaterMark);
break;
}
channelManager =
new ChannelManager(
lookupService,
localInstanceConnectionInfo,
numBuffers,
bufferSize,
networkConnectionManager);
} catch (IOException ioe) {
LOG.error(StringUtils.stringifyException(ioe));
throw new Exception("Failed to instantiate channel manager. " + ioe.getMessage(), ioe);
}
{
HardwareDescription resources = HardwareDescriptionFactory.extractFromSystem();
int slots = GlobalConfiguration.getInteger(ConfigConstants.TASK_MANAGER_NUM_TASK_SLOTS, -1);
if (slots == -1) {
slots = Hardware.getNumberCPUCores();
} else if (slots <= 0) {
throw new Exception("Illegal value for the number of task slots: " + slots);
}
this.numberOfSlots = slots;
// Check whether the memory size has been explicitly configured. if so that overrides the
// default mechanism
// of taking as much as is mentioned in the hardware description
long memorySize =
GlobalConfiguration.getInteger(ConfigConstants.TASK_MANAGER_MEMORY_SIZE_KEY, -1);
if (memorySize > 0) {
// manually configured memory size. override the value in the hardware config
resources =
HardwareDescriptionFactory.construct(
resources.getNumberOfCPUCores(),
resources.getSizeOfPhysicalMemory(),
memorySize * 1024L * 1024L);
}
this.hardwareDescription = resources;
// Initialize the memory manager
LOG.info(
"Initializing memory manager with "
+ (resources.getSizeOfFreeMemory() >>> 20)
+ " megabytes of memory. "
+ "Page size is "
+ pageSize
+ " bytes.");
try {
@SuppressWarnings("unused")
final boolean lazyAllocation =
GlobalConfiguration.getBoolean(
ConfigConstants.TASK_MANAGER_MEMORY_LAZY_ALLOCATION_KEY,
ConfigConstants.DEFAULT_TASK_MANAGER_MEMORY_LAZY_ALLOCATION);
this.memoryManager =
new DefaultMemoryManager(resources.getSizeOfFreeMemory(), this.numberOfSlots, pageSize);
} catch (Throwable t) {
LOG.fatal(
"Unable to initialize memory manager with "
+ (resources.getSizeOfFreeMemory() >>> 20)
+ " megabytes of memory.",
t);
throw new Exception("Unable to initialize memory manager.", t);
}
}
this.ioManager = new IOManager(tmpDirPaths);
this.heartbeatThread =
new Thread() {
@Override
public void run() {
runHeartbeatLoop();
}
};
this.heartbeatThread.setName("Heartbeat Thread");
this.heartbeatThread.start();
// --------------------------------------------------------------------
// Memory Usage
// --------------------------------------------------------------------
final MemoryMXBean memoryMXBean = ManagementFactory.getMemoryMXBean();
final List<GarbageCollectorMXBean> gcMXBeans = ManagementFactory.getGarbageCollectorMXBeans();
LOG.info(getMemoryUsageStatsAsString(memoryMXBean));
boolean startMemoryUsageLogThread =
GlobalConfiguration.getBoolean(
ConfigConstants.TASK_MANAGER_DEBUG_MEMORY_USAGE_START_LOG_THREAD,
ConfigConstants.DEFAULT_TASK_MANAGER_DEBUG_MEMORY_USAGE_START_LOG_THREAD);
if (startMemoryUsageLogThread && LOG.isDebugEnabled()) {
final int logIntervalMs =
GlobalConfiguration.getInteger(
ConfigConstants.TASK_MANAGER_DEBUG_MEMORY_USAGE_LOG_INTERVAL_MS,
ConfigConstants.DEFAULT_TASK_MANAGER_DEBUG_MEMORY_USAGE_LOG_INTERVAL_MS);
new Thread(
new Runnable() {
@Override
public void run() {
try {
while (!isShutDown()) {
Thread.sleep(logIntervalMs);
LOG.debug(getMemoryUsageStatsAsString(memoryMXBean));
LOG.debug(getGarbageCollectorStatsAsString(gcMXBeans));
}
} catch (InterruptedException e) {
LOG.warn("Unexpected interruption of memory usage logger thread.");
}
}
})
.start();
}
}
/**
* Creates a new {@link ClusterInstance} object to manage instances that can be executed on that
* host.
*
* @param instanceConnectionInfo the connection information for the instance
* @param hardwareDescription the hardware description provided by the new instance
* @return a new {@link ClusterInstance} object or <code>null</code> if the cluster instance could
* not be created
*/
private ClusterInstance createNewHost(
final InstanceConnectionInfo instanceConnectionInfo,
final HardwareDescription hardwareDescription) {
// Check if there is a user-defined instance type for this IP address
InstanceType instanceType =
this.ipToInstanceTypeMapping.get(instanceConnectionInfo.getAddress());
if (instanceType != null) {
LOG.info(
"Found user-defined instance type for cluster instance with IP "
+ instanceConnectionInfo.getAddress()
+ ": "
+ instanceType);
} else {
instanceType = matchHardwareDescriptionWithInstanceType(hardwareDescription);
if (instanceType != null) {
LOG.info(
"Hardware profile of cluster instance with IP "
+ instanceConnectionInfo.getAddress()
+ " matches with instance type "
+ instanceType);
} else {
LOG.error("No matching instance type, cannot create cluster instance");
return null;
}
}
// Try to match new host with a stub host from the existing topology
String instanceName = instanceConnectionInfo.getHostName();
NetworkNode parentNode = this.networkTopology.getRootNode();
NetworkNode currentStubNode = null;
// Try to match new host using the host name
while (true) {
currentStubNode = this.networkTopology.getNodeByName(instanceName);
if (currentStubNode != null) {
break;
}
final int pos = instanceName.lastIndexOf('.');
if (pos == -1) {
break;
}
/*
* If host name is reported as FQDN, iterative remove parts
* of the domain name until a match occurs or no more dots
* can be found in the host name.
*/
instanceName = instanceName.substring(0, pos);
}
// Try to match the new host using the IP address
if (currentStubNode == null) {
instanceName = instanceConnectionInfo.getAddress().toString();
instanceName = instanceName.replaceAll("/", ""); // Remove any / characters
currentStubNode = this.networkTopology.getNodeByName(instanceName);
}
if (currentStubNode != null) {
/*
* The instance name will be the same as the one of the stub node. That way
* the stub now will be removed from the network topology and replaced be
* the new node.
*/
if (currentStubNode.getParentNode() != null) {
parentNode = currentStubNode.getParentNode();
}
// Remove the stub node from the tree
currentStubNode.remove();
}
LOG.info(
"Creating instance of type "
+ instanceType
+ " for "
+ instanceConnectionInfo
+ ", parent is "
+ parentNode.getName());
final ClusterInstance host =
new ClusterInstance(
instanceConnectionInfo,
instanceType,
parentNode,
this.networkTopology,
hardwareDescription);
return host;
}