/**
* Starts communication.
*
* @throws IgniteCheckedException If failed.
*/
public void start() throws IgniteCheckedException {
try {
locHost = U.getLocalHost();
} catch (IOException e) {
throw new IgniteCheckedException("Failed to initialize local address.", e);
}
try {
shmemSrv = resetShmemServer();
} catch (IgniteCheckedException e) {
U.warn(log, "Failed to start shared memory communication server.", e);
}
try {
// This method potentially resets local port to the value
// local node was bound to.
nioSrvr = resetNioServer();
} catch (IgniteCheckedException e) {
throw new IgniteCheckedException("Failed to initialize TCP server: " + locHost, e);
}
locProcDesc.address(locHost.getHostAddress());
locProcDesc.sharedMemoryPort(boundTcpShmemPort);
locProcDesc.tcpPort(boundTcpPort);
locIdMsg = new ProcessHandshakeMessage(locProcDesc);
if (shmemSrv != null) {
shmemAcceptWorker = new ShmemAcceptWorker(shmemSrv);
new IgniteThread(shmemAcceptWorker).start();
}
nioSrvr.start();
}
/**
* Establish TCP connection to remote hadoop process and returns client.
*
* @param desc Process descriptor.
* @return Client.
* @throws IgniteCheckedException If failed.
*/
protected HadoopCommunicationClient createTcpClient(HadoopProcessDescriptor desc)
throws IgniteCheckedException {
String addr = desc.address();
int port = desc.tcpPort();
if (log.isDebugEnabled())
log.debug(
"Trying to connect to remote process [locProcDesc="
+ locProcDesc
+ ", desc="
+ desc
+ ']');
boolean conn = false;
HadoopTcpNioCommunicationClient client = null;
IgniteCheckedException errs = null;
int connectAttempts = 1;
long connTimeout0 = connTimeout;
int attempt = 1;
while (!conn) { // Reconnection on handshake timeout.
try {
SocketChannel ch = SocketChannel.open();
ch.configureBlocking(true);
ch.socket().setTcpNoDelay(tcpNoDelay);
ch.socket().setKeepAlive(true);
if (sockRcvBuf > 0) ch.socket().setReceiveBufferSize(sockRcvBuf);
if (sockSndBuf > 0) ch.socket().setSendBufferSize(sockSndBuf);
ch.socket().connect(new InetSocketAddress(addr, port), (int) connTimeout);
HandshakeFinish fin = new HandshakeFinish();
GridNioSession ses = nioSrvr.createSession(ch, F.asMap(HANDSHAKE_FINISH_META, fin)).get();
client = new HadoopTcpNioCommunicationClient(ses);
if (log.isDebugEnabled()) log.debug("Waiting for handshake finish for client: " + client);
fin.await(connTimeout0);
conn = true;
} catch (HadoopHandshakeTimeoutException e) {
if (client != null) {
client.forceClose();
client = null;
}
if (log.isDebugEnabled())
log.debug(
"Handshake timedout (will retry with increased timeout) [timeout="
+ connTimeout0
+ ", desc="
+ desc
+ ", port="
+ port
+ ", err="
+ e
+ ']');
if (attempt == reconCnt || connTimeout0 > maxConnTimeout) {
if (log.isDebugEnabled())
log.debug(
"Handshake timed out (will stop attempts to perform the handshake) "
+ "[timeout="
+ connTimeout0
+ ", maxConnTimeout="
+ maxConnTimeout
+ ", attempt="
+ attempt
+ ", reconCnt="
+ reconCnt
+ ", err="
+ e.getMessage()
+ ", addr="
+ addr
+ ']');
if (errs == null)
errs =
new IgniteCheckedException(
"Failed to connect to remote Hadoop process "
+ "(is process still running?) [desc="
+ desc
+ ", addrs="
+ addr
+ ']');
errs.addSuppressed(e);
break;
} else {
attempt++;
connTimeout0 *= 2;
// Continue loop.
}
} catch (Exception e) {
if (client != null) {
client.forceClose();
client = null;
}
if (log.isDebugEnabled())
log.debug("Client creation failed [addr=" + addr + ", port=" + port + ", err=" + e + ']');
if (X.hasCause(e, SocketTimeoutException.class))
LT.warn(
log,
null,
"Connect timed out (consider increasing 'connTimeout' "
+ "configuration property) [addr="
+ addr
+ ", port="
+ port
+ ']');
if (errs == null)
errs =
new IgniteCheckedException(
"Failed to connect to remote Hadoop process (is process still running?) "
+ "[desc="
+ desc
+ ", addrs="
+ addr
+ ']');
errs.addSuppressed(e);
// Reconnect for the second time, if connection is not established.
if (connectAttempts < 2
&& (e instanceof ConnectException || X.hasCause(e, ConnectException.class))) {
connectAttempts++;
continue;
}
break;
}
}
if (client == null) {
assert errs != null;
if (X.hasCause(errs, ConnectException.class))
LT.warn(
log,
null,
"Failed to connect to a remote Hadoop process (is process still running?). "
+ "Make sure operating system firewall is disabled on local and remote host) "
+ "[addrs="
+ addr
+ ", port="
+ port
+ ']');
throw errs;
}
if (log.isDebugEnabled()) log.debug("Created client: " + client);
return client;
}
