@Override
public void run() {
long lastHeartbeatMs = System.currentTimeMillis();
Command cmd = null;
while (!mStop) {
long diff = System.currentTimeMillis() - lastHeartbeatMs;
if (diff < WorkerConf.get().TO_MASTER_HEARTBEAT_INTERVAL_MS) {
LOG.debug("Heartbeat process takes {} ms.", diff);
CommonUtils.sleepMs(LOG, WorkerConf.get().TO_MASTER_HEARTBEAT_INTERVAL_MS - diff);
} else {
LOG.error("Heartbeat process takes " + diff + " ms.");
}
try {
cmd = mWorkerStorage.heartbeat();
lastHeartbeatMs = System.currentTimeMillis();
} catch (IOException e) {
LOG.error(e.getMessage(), e);
mWorkerStorage.resetMasterClient();
CommonUtils.sleepMs(LOG, Constants.SECOND_MS);
cmd = null;
if (System.currentTimeMillis() - lastHeartbeatMs >= WorkerConf.get().HEARTBEAT_TIMEOUT_MS) {
throw new RuntimeException(
"Timebeat timeout " + (System.currentTimeMillis() - lastHeartbeatMs) + "ms");
}
}
if (cmd != null) {
switch (cmd.mCommandType) {
case Unknown:
LOG.error("Unknown command: " + cmd);
break;
case Nothing:
LOG.debug("Nothing command: {}", cmd);
break;
case Register:
LOG.info("Register command: " + cmd);
mWorkerStorage.register();
break;
case Free:
mWorkerStorage.freeBlocks(cmd.mData);
LOG.info("Free command: " + cmd);
break;
case Delete:
LOG.info("Delete command: " + cmd);
break;
default:
throw new RuntimeException("Un-recognized command from master " + cmd.toString());
}
}
mWorkerStorage.checkStatus();
}
}
private void swapoutOrphanBlocksFileTestUtil(int filesize) throws Exception {
int fid = TestUtils.createByteFile(mTfs, "/xyz", WriteType.MUST_CACHE, filesize);
long bid = mTfs.getBlockId(fid, 0);
mLocalTachyonCluster.stopWorker();
mTfs.delete(fid, true);
WorkerStorage ws =
new WorkerStorage(mMasterAddress, mWorkerAddress, mWorkerDataFolder, WORKER_CAPACITY_BYTES);
File orphanblock = new File(ws.getUnderfsOrphansFolder() + "/" + bid);
Assert.assertFalse(
"Orphan block file isn't deleted from workerDataFolder",
new File(mWorkerDataFolder + "/" + bid).exists());
Assert.assertTrue("UFS hasn't the orphan block file ", orphanblock.exists());
Assert.assertTrue("Orpahblock file size is changed", orphanblock.length() == filesize);
}
/**
* @param masterAddress The TachyonMaster's address.
* @param workerAddress This TachyonWorker's address.
* @param dataPort This TachyonWorker's data server's port
* @param selectorThreads The number of selector threads of the worker's thrift server
* @param acceptQueueSizePerThreads The accept queue size per thread of the worker's thrift server
* @param workerThreads The number of threads of the worker's thrift server
* @param dataFolder This TachyonWorker's local folder's path
* @param memoryCapacityBytes The maximum memory space this TachyonWorker can use, in bytes
*/
private TachyonWorker(
InetSocketAddress masterAddress,
InetSocketAddress workerAddress,
int dataPort,
int selectorThreads,
int acceptQueueSizePerThreads,
int workerThreads,
String dataFolder,
long memoryCapacityBytes) {
CommonConf.assertValidPort(masterAddress);
CommonConf.assertValidPort(workerAddress);
CommonConf.assertValidPort(dataPort);
mMasterAddress = masterAddress;
mWorkerStorage =
new WorkerStorage(mMasterAddress, dataFolder, memoryCapacityBytes, mExecutorService);
mWorkerServiceHandler = new WorkerServiceHandler(mWorkerStorage);
// Extract the port from the generated socket.
// When running tests, its great to use port '0' so the system will figure out what port to use
// (any random free port).
// In a production or any real deployment setup, port '0' should not be used as it will make
// deployment more complicated.
InetSocketAddress dataAddress = new InetSocketAddress(workerAddress.getHostName(), dataPort);
BlocksLocker blockLocker = new BlocksLocker(mWorkerStorage, Users.DATASERVER_USER_ID);
mDataServer = createDataServer(dataAddress, blockLocker);
mDataPort = mDataServer.getPort();
mHeartbeatThread = new Thread(this);
try {
LOG.info("Tachyon Worker version " + Version.VERSION + " tries to start @ " + workerAddress);
WorkerService.Processor<WorkerServiceHandler> processor =
new WorkerService.Processor<WorkerServiceHandler>(mWorkerServiceHandler);
mServerTNonblockingServerSocket = new TNonblockingServerSocket(workerAddress);
mPort = NetworkUtils.getPort(mServerTNonblockingServerSocket);
mServer =
new TThreadedSelectorServer(
new TThreadedSelectorServer.Args(mServerTNonblockingServerSocket)
.processor(processor)
.selectorThreads(selectorThreads)
.acceptQueueSizePerThread(acceptQueueSizePerThreads)
.workerThreads(workerThreads));
} catch (TTransportException e) {
LOG.error(e.getMessage(), e);
throw Throwables.propagate(e);
}
mWorkerAddress =
new NetAddress(workerAddress.getAddress().getCanonicalHostName(), mPort, mDataPort);
mWorkerStorage.initialize(mWorkerAddress);
}
/**
* Stop this TachyonWorker. Stop all the threads belong to this TachyonWorker.
*
* @throws IOException
* @throws InterruptedException
*/
public void stop() throws IOException, InterruptedException {
mStop = true;
mWorkerStorage.stop();
mDataServer.close();
mServer.stop();
mServerTNonblockingServerSocket.close();
mExecutorService.shutdown();
while (!mDataServer.isClosed() || mServer.isServing() || mHeartbeatThread.isAlive()) {
// TODO The reason to stop and close again is due to some issues in Thrift.
mServer.stop();
mServerTNonblockingServerSocket.close();
CommonUtils.sleepMs(null, 100);
}
mHeartbeatThread.join();
}