/**
* Validates the path, verifying that it contains the {@link Constants#HEADER} or {@link
* Constants#HEADER_FT} and a hostname:port specified.
*
* @param path the path to be verified
* @param tachyonConf the instance of {@link tachyon.conf.TachyonConf} to be used
* @return the verified path in a form like tachyon://host:port/dir. If only the "/dir" or "dir"
* part is provided, the host and port are retrieved from property, tachyon.master.hostname
* and tachyon.master.port, respectively.
* @throws IOException if the given path is not valid
*/
public static String validatePath(String path, TachyonConf tachyonConf) throws IOException {
if (path.startsWith(Constants.HEADER) || path.startsWith(Constants.HEADER_FT)) {
if (!path.contains(":")) {
throw new IOException(
"Invalid Path: "
+ path
+ ". Use "
+ Constants.HEADER
+ "host:port/ ,"
+ Constants.HEADER_FT
+ "host:port/"
+ " , or /file");
} else {
return path;
}
} else {
String hostname = NetworkAddressUtils.getConnectHost(ServiceType.MASTER_RPC, tachyonConf);
int port = tachyonConf.getInt(Constants.MASTER_PORT);
if (tachyonConf.getBoolean(Constants.ZOOKEEPER_ENABLED)) {
return PathUtils.concatPath(Constants.HEADER_FT + hostname + ":" + port, path);
}
return PathUtils.concatPath(Constants.HEADER + hostname + ":" + port, path);
}
}
/**
* Starts the Tachyon worker server.
*
* @throws Exception if the workers fail to start
*/
public void start() throws Exception {
// NOTE: the order to start different services is sensitive. If you change it, do it cautiously.
// Start serving metrics system, this will not block
mWorkerMetricsSystem.start();
// Start serving the web server, this will not block
// Requirement: metrics system started so we could add the metrics servlet to the web server
// Consequence: when starting webserver, the webport will be updated.
mWebServer.addHandler(mWorkerMetricsSystem.getServletHandler());
mWebServer.startWebServer();
// Set updated net address for this worker in context
// Requirement: RPC, web, and dataserver ports are updated
// Consequence: create a NetAddress object and set it into WorkerContext
mNetAddress =
new NetAddress(
NetworkAddressUtils.getConnectHost(ServiceType.WORKER_RPC, mTachyonConf),
mTachyonConf.getInt(Constants.WORKER_RPC_PORT),
getDataLocalPort(),
mTachyonConf.getInt(Constants.WORKER_WEB_PORT));
WorkerContext.setWorkerNetAddress(mNetAddress);
// Start each worker
// Requirement: NetAddress set in WorkerContext, so block worker can initialize BlockMasterSync
// Consequence: worker id is granted
startWorkers();
LOG.info("Started worker with id {}", WorkerIdRegistry.getWorkerId());
mIsServingRPC = true;
// Start serving RPC, this will block
LOG.info("Tachyon Worker version {} started @ {}", Version.VERSION, mWorkerAddress);
mThriftServer.serve();
LOG.info("Tachyon Worker version {} ended @ {}", Version.VERSION, mWorkerAddress);
}
private LocalTachyonMaster(final String tachyonHome) throws IOException {
mTachyonHome = tachyonHome;
TachyonConf tachyonConf = MasterContext.getConf();
mHostname = NetworkAddressUtils.getConnectHost(ServiceType.MASTER_RPC, tachyonConf);
// To start the UFS either for integration or unit test. If it targets the unit test, UFS is
// setup over the local file system (see also {@link LocalFilesystemCluster} - under folder of
// "mTachyonHome/tachyon*". Otherwise, it starts some distributed file system cluster e.g.,
// miniDFSCluster (see also {@link tachyon.LocalMiniDFScluster} and setup the folder like
// "hdfs://xxx:xxx/tachyon*".
mUnderFSCluster = UnderFileSystemCluster.get(mTachyonHome + "/dfs", tachyonConf);
mUnderFSFolder = mUnderFSCluster.getUnderFilesystemAddress() + "/tachyon_underfs_folder";
// To setup the journalFolder under either local file system or distributed ufs like
// miniDFSCluster
mJournalFolder = mUnderFSCluster.getUnderFilesystemAddress() + "/journal";
UnderFileSystemUtils.mkdirIfNotExists(mJournalFolder, tachyonConf);
String[] masterServiceNames =
new String[] {
Constants.BLOCK_MASTER_SERVICE_NAME,
Constants.FILE_SYSTEM_MASTER_SERVICE_NAME,
Constants.RAW_TABLE_MASTER_SERVICE_NAME,
};
for (String masterServiceName : masterServiceNames) {
UnderFileSystemUtils.mkdirIfNotExists(
PathUtils.concatPath(mJournalFolder, masterServiceName), tachyonConf);
}
UnderFileSystemUtils.touch(
mJournalFolder + "/_format_" + System.currentTimeMillis(), tachyonConf);
tachyonConf.set(Constants.MASTER_JOURNAL_FOLDER, mJournalFolder);
tachyonConf.set(Constants.UNDERFS_ADDRESS, mUnderFSFolder);
tachyonConf.set(Constants.MASTER_MIN_WORKER_THREADS, "1");
tachyonConf.set(Constants.MASTER_MAX_WORKER_THREADS, "100");
// If tests fail to connect they should fail early rather than using the default ridiculously
// high retries
tachyonConf.set(Constants.MASTER_RETRY_COUNT, "3");
// Since tests are always running on a single host keep the resolution timeout low as otherwise
// people running with strange network configurations will see very slow tests
tachyonConf.set(Constants.HOST_RESOLUTION_TIMEOUT_MS, "250");
tachyonConf.set(Constants.WEB_THREAD_COUNT, "1");
tachyonConf.set(
Constants.WEB_RESOURCES,
PathUtils.concatPath(System.getProperty("user.dir"), "../servers/src/main/webapp"));
mTachyonMaster = TachyonMaster.Factory.createMaster();
// Reset the master port
tachyonConf.set(Constants.MASTER_PORT, Integer.toString(getRPCLocalPort()));
Runnable runMaster =
new Runnable() {
@Override
public void run() {
try {
mTachyonMaster.start();
} catch (Exception e) {
throw new RuntimeException(e + " \n Start Master Error \n" + e.getMessage(), e);
}
}
};
mMasterThread = new Thread(runMaster);
}
/**
* Creates a new instance of {@link BlockWorker}.
*
* @throws ConnectionFailedException if network connection failed
* @throws IOException for other exceptions
*/
public BlockWorker() throws IOException, ConnectionFailedException {
super(
Executors.newFixedThreadPool(
4, ThreadFactoryUtils.build("block-worker-heartbeat-%d", true)));
mTachyonConf = WorkerContext.getConf();
mStartTimeMs = System.currentTimeMillis();
// Setup MasterClientBase
mBlockMasterClient =
new BlockMasterClient(
NetworkAddressUtils.getConnectAddress(ServiceType.MASTER_RPC, mTachyonConf),
mTachyonConf);
mFileSystemMasterClient =
new FileSystemMasterClient(
NetworkAddressUtils.getConnectAddress(ServiceType.MASTER_RPC, mTachyonConf),
mTachyonConf);
// Set up BlockDataManager
WorkerSource workerSource = new WorkerSource();
mBlockDataManager =
new BlockDataManager(
workerSource, mBlockMasterClient, mFileSystemMasterClient, new TieredBlockStore());
// Setup metrics collection
mWorkerMetricsSystem = new MetricsSystem("worker", mTachyonConf);
workerSource.registerGauges(mBlockDataManager);
mWorkerMetricsSystem.registerSource(workerSource);
// Setup DataServer
mDataServer =
DataServer.Factory.create(
NetworkAddressUtils.getBindAddress(ServiceType.WORKER_DATA, mTachyonConf),
mBlockDataManager,
mTachyonConf);
// Reset data server port
mTachyonConf.set(Constants.WORKER_DATA_PORT, Integer.toString(mDataServer.getPort()));
// Setup RPC Server
mServiceHandler = new BlockWorkerClientServiceHandler(mBlockDataManager);
mThriftServerSocket = createThriftServerSocket();
mPort = NetworkAddressUtils.getThriftPort(mThriftServerSocket);
// Reset worker RPC port
mTachyonConf.set(Constants.WORKER_RPC_PORT, Integer.toString(mPort));
mThriftServer = createThriftServer();
// Setup web server
mWebServer =
new WorkerUIWebServer(
ServiceType.WORKER_WEB,
NetworkAddressUtils.getBindAddress(ServiceType.WORKER_WEB, mTachyonConf),
mBlockDataManager,
NetworkAddressUtils.getConnectAddress(ServiceType.WORKER_RPC, mTachyonConf),
mStartTimeMs,
mTachyonConf);
mWorkerMetricsSystem.start();
// Add the metrics servlet to the web server, this must be done after the metrics system starts
mWebServer.addHandler(mWorkerMetricsSystem.getServletHandler());
mWebServer.startWebServer();
int webPort = mWebServer.getLocalPort();
// Get the worker id
mWorkerNetAddress =
new NetAddress(
NetworkAddressUtils.getConnectHost(ServiceType.WORKER_RPC, mTachyonConf),
mPort,
mDataServer.getPort(),
webPort);
WorkerIdRegistry.registerWithBlockMaster(mBlockMasterClient, mWorkerNetAddress);
mBlockMasterSync =
new BlockMasterSync(mBlockDataManager, mWorkerNetAddress, mBlockMasterClient);
// Setup PinListSyncer
mPinListSync = new PinListSync(mBlockDataManager, mFileSystemMasterClient);
// Setup session cleaner
mSessionCleanerThread = new SessionCleaner(mBlockDataManager);
// Setup space reserver
if (mTachyonConf.getBoolean(Constants.WORKER_TIERED_STORE_RESERVER_ENABLED)) {
mSpaceReserver = new SpaceReserver(mBlockDataManager);
}
}
@Override
public List<String> getFileLocations(String path) throws IOException {
List<String> ret = new ArrayList<String>();
ret.add(NetworkAddressUtils.getConnectHost(ServiceType.WORKER_RPC, mTachyonConf));
return ret;
}
/**
* Reads from a remote byte buffer.
*
* @param tachyonFS a TachyonFS
* @param blockInfo the block information
* @param offset offset to start the read at
* @param len number of bytes to read
* @param conf Tachyon configuration
* @return <code>ByteBuffer</code> containing the bytes read
*/
public ByteBuffer readRemoteByteBuffer(
TachyonFS tachyonFS, ClientBlockInfo blockInfo, long offset, long len, TachyonConf conf) {
ByteBuffer buf = null;
try {
List<NetAddress> blockLocations = blockInfo.getLocations();
LOG.info("Block locations:" + blockLocations);
String localhost = NetworkAddressUtils.getConnectHost(ServiceType.WORKER_RPC, conf);
for (NetAddress blockLocation : blockLocations) {
String host = blockLocation.mHost;
int port = blockLocation.mSecondaryPort;
// The data is not in remote machine's memory if primary port == -1. We check primary port
// because if the data is in the under storage, the secondary port (data transfer port)
// will be set.
if (blockLocation.mPort == -1) {
continue;
}
if (host.equals(InetAddress.getLocalHost().getHostName())
|| host.equals(InetAddress.getLocalHost().getHostAddress())
|| host.equals(localhost)) {
LOG.warn(
"Master thinks the local machine has data, but not!"
+ "(or local read is disabled) blockId:{}",
blockInfo.blockId);
}
LOG.info(
host
+ ":"
+ port
+ " current host is "
+ localhost
+ " "
+ NetworkAddressUtils.getLocalIpAddress(conf));
try {
buf =
retrieveByteBufferFromRemoteMachine(
new InetSocketAddress(host, port), blockInfo.blockId, offset, len, conf);
if (buf != null) {
break;
}
} catch (IOException e) {
LOG.error(
"Fail to retrieve byte buffer for block "
+ blockInfo.blockId
+ " from remote "
+ host
+ ":"
+ port
+ " with offset "
+ offset
+ " and length "
+ len,
e);
buf = null;
}
}
} catch (IOException e) {
LOG.error("Failed to get read data from remote ", e);
buf = null;
}
return buf;
}