/**
* Sends message to Hadoop process.
*
* @param desc
* @param msg
* @throws IgniteCheckedException
*/
public void sendMessage(HadoopProcessDescriptor desc, HadoopMessage msg)
throws IgniteCheckedException {
assert desc != null;
assert msg != null;
if (log.isTraceEnabled())
log.trace("Sending message to Hadoop process [desc=" + desc + ", msg=" + msg + ']');
HadoopCommunicationClient client = null;
boolean closeOnRelease = true;
try {
client = reserveClient(desc);
client.sendMessage(desc, msg);
closeOnRelease = false;
} finally {
if (client != null) {
if (closeOnRelease) {
client.forceClose();
clients.remove(desc.processId(), client);
} else client.release();
}
}
}
/**
* Returns existing or just created client to node.
*
* @param desc Node to which client should be open.
* @return The existing or just created client.
* @throws IgniteCheckedException Thrown if any exception occurs.
*/
private HadoopCommunicationClient reserveClient(HadoopProcessDescriptor desc)
throws IgniteCheckedException {
assert desc != null;
UUID procId = desc.processId();
while (true) {
HadoopCommunicationClient client = clients.get(procId);
if (client == null) {
if (log.isDebugEnabled())
log.debug(
"Did not find client for remote process [locProcDesc="
+ locProcDesc
+ ", desc="
+ desc
+ ']');
// Do not allow concurrent connects.
Object sync = locks.lock(procId);
try {
client = clients.get(procId);
if (client == null) {
HadoopCommunicationClient old = clients.put(procId, client = createNioClient(desc));
assert old == null;
}
} finally {
locks.unlock(procId, sync);
}
assert client != null;
}
if (client.reserve()) return client;
else
// Client has just been closed by idle worker. Help it and try again.
clients.remove(procId, client);
}
}
/** {@inheritDoc} */
@Override
public void onDisconnected(GridNioSession ses, @Nullable Exception e) {
if (log.isDebugEnabled()) log.debug("Closed connection for session: " + ses);
if (e != null) U.error(log, "Session disconnected due to exception: " + ses, e);
HadoopProcessDescriptor desc = ses.meta(PROCESS_META);
if (desc != null) {
HadoopCommunicationClient rmv = clients.remove(desc.processId());
if (rmv != null) rmv.forceClose();
}
HadoopMessageListener lsnr0 = lsnr;
if (lsnr0 != null)
// Notify listener about connection close.
lsnr0.onConnectionLost(desc);
}
/**
* Creates new shared memory communication server.
*
* @return Server.
* @throws IgniteCheckedException If failed.
*/
@Nullable
private IpcSharedMemoryServerEndpoint resetShmemServer() throws IgniteCheckedException {
if (boundTcpShmemPort >= 0)
throw new IgniteCheckedException(
"Shared memory server was already created on port " + boundTcpShmemPort);
if (shmemPort == -1 || U.isWindows()) return null;
IgniteCheckedException lastEx = null;
// If configured TCP port is busy, find first available in range.
for (int port = shmemPort; port < shmemPort + locPortRange; port++) {
try {
IpcSharedMemoryServerEndpoint srv =
new IpcSharedMemoryServerEndpoint(
log.getLogger(IpcSharedMemoryServerEndpoint.class),
locProcDesc.processId(),
gridName);
srv.setPort(port);
srv.omitOutOfResourcesWarning(true);
srv.start();
boundTcpShmemPort = port;
// Ack Port the TCP server was bound to.
if (log.isInfoEnabled())
log.info(
"Successfully bound shared memory communication to TCP port [port="
+ boundTcpShmemPort
+ ", locHost="
+ locHost
+ ']');
return srv;
} catch (IgniteCheckedException e) {
lastEx = e;
if (log.isDebugEnabled())
log.debug(
"Failed to bind to local port (will try next port within range) [port="
+ port
+ ", locHost="
+ locHost
+ ']');
}
}
// If free port wasn't found.
throw new IgniteCheckedException(
"Failed to bind shared memory communication to any port within range [startPort="
+ locPort
+ ", portRange="
+ locPortRange
+ ", locHost="
+ locHost
+ ']',
lastEx);
}
/** {@inheritDoc} */
@Override
public void onMessageReceived(GridNioSession ses, Object msg) throws IgniteCheckedException {
HadoopProcessDescriptor desc = ses.meta(PROCESS_META);
UUID rmtProcId = desc == null ? null : desc.processId();
if (rmtProcId == null) {
if (!(msg instanceof ProcessHandshakeMessage)) {
log.warning(
"Invalid handshake message received, will close connection [ses="
+ ses
+ ", msg="
+ msg
+ ']');
ses.close();
return;
}
ProcessHandshakeMessage nId = (ProcessHandshakeMessage) msg;
if (log.isDebugEnabled())
log.debug("Received handshake message [ses=" + ses + ", msg=" + msg + ']');
ses.addMeta(PROCESS_META, nId.processDescriptor());
if (!ses.accepted())
// Send handshake reply.
ses.send(locIdMsg);
else {
//
rmtProcId = nId.processDescriptor().processId();
if (log.isDebugEnabled()) log.debug("Finished handshake with remote client: " + ses);
Object sync = locks.tryLock(rmtProcId);
if (sync != null) {
try {
if (clients.get(rmtProcId) == null) {
if (log.isDebugEnabled())
log.debug("Will reuse session for descriptor: " + rmtProcId);
// Handshake finished flag is true.
clients.put(rmtProcId, new HadoopTcpNioCommunicationClient(ses));
} else {
if (log.isDebugEnabled())
log.debug(
"Will not reuse client as another already exists [locProcDesc="
+ locProcDesc
+ ", desc="
+ desc
+ ']');
}
} finally {
locks.unlock(rmtProcId, sync);
}
} else {
if (log.isDebugEnabled())
log.debug(
"Concurrent connection is being established, will not reuse client session ["
+ "locProcDesc="
+ locProcDesc
+ ", desc="
+ desc
+ ']');
}
}
if (log.isDebugEnabled())
log.debug(
"Handshake is finished for session [ses="
+ ses
+ ", locProcDesc="
+ locProcDesc
+ ']');
HandshakeFinish to = ses.meta(HANDSHAKE_FINISH_META);
if (to != null) to.finish();
// Notify session opened (both parties).
proceedSessionOpened(ses);
} else {
if (msgQueueLimit > 0) {
GridNioMessageTracker tracker = ses.meta(TRACKER_META);
if (tracker == null) {
GridNioMessageTracker old =
ses.addMeta(TRACKER_META, tracker = new GridNioMessageTracker(ses, msgQueueLimit));
assert old == null;
}
tracker.onMessageReceived();
}
proceedMessageReceived(ses, msg);
}
}