Identifying a bottleneck in a multithreaded MQTT publisher - java

I'm currently developing a MQTT client service with Eclipse Paho for a bigger software and I'm encountering performance issues. I'm getting a bunch of events that I want to publish to the broker and I'm using GSON for serialization of those events. I've multithreaded the serialization and publishing. According to a rudimentary benchmark serialization and publishing takes up to 1 ms.
I'm using an ExecutorService with a fixed threadpool of size 10 (for now).
My code is currently submitting about 50 Runnables per second to the ExecutorService, but my Broker reports only about 5-10 messages per second.
I've previously benchmarked my MQTT setup and managed to send about 9000+ MQTT messages per second in a non-multithreaded way.
Does the threadpool have that much of an overhead, that I'm only able to get this small amount of publishes out of it?
public class MqttService implements IMessagingService{
protected int PORT = 1883;
protected String HOST = "localhost";
protected final String SERVICENAME = "MQTT";
protected static final String COMMANDTOPIC = "commands";
protected static final String REMINDSPREFIX = "Reminds/";
protected static final String VIOLATIONTOPIC = "violations/";
protected static final String WILDCARDTOPIC = "Reminds/#";
protected static final String TCPPREFIX = "tcp://";
protected static final String SSLPREFIX = "ssl://";
private MqttClient client;
private MqttConnectOptions optionsPublisher = new MqttConnectOptions();
private ExecutorService pool = Executors.newFixedThreadPool(10);
public MqttService() {
this("localhost", 1883);
}
public MqttService(String host, int port) {
this.HOST = host;
this.PORT = port;
}
#Override
public void setPort(int port) {
this.PORT = port;
}
#Override
public void setHost(String host) {
this.HOST = host;
}
#Override
public void sendMessage(AbstractMessage message) {
pool.submit(new SerializeJob(client,message));
}
#Override
public void connect() {
try {
client = new MqttClient(TCPPREFIX + HOST + ":" + PORT, IDPublisher);
optionsPublisher.setMqttVersion(MqttConnectOptions.MQTT_VERSION_3_1_1);
client.connect(optionsPublisher);
client.setCallback(new MessageCallback());
client.subscribe(WILDCARDTOPIC, 0);
} catch (MqttException e1) {
e1.printStackTrace();
}
}
}
The following code is the Runnable executed by the ExecutorService. This should not be an issue by itself though, since it only takes up to 1-2 ms to finish.
class SerializeJob implements Runnable {
private AbstractMessage message;
private MqttClient client;
public SerializeJob(MqttClient client, AbstractMessage m) {
this.client = client;
this.message = m;
}
#Override
public void run() {
String serializedMessage = MessageSerializer.serializeMessage(message);
MqttMessage wireMessage = new MqttMessage();
wireMessage.setQos(message.getQos());
wireMessage.setPayload(serializedMessage.getBytes());
if (client.isConnected()) {
StringBuilder topic = new StringBuilder();
topic.append(MqttService.REMINDSPREFIX);
topic.append(MqttService.VIOLATIONTOPIC);
try {
client.publish(topic.toString(), wireMessage);
} catch (MqttPersistenceException e) {
e.printStackTrace();
} catch (MqttException e) {
e.printStackTrace();
}
}
}
}
I'm not quite sure what's holing me back. MQTT itself seems to allow for a lot of events, that also can have a big payload, and network cannot possibly an issue either, since I'm currently hosting the broker locally on the same machine as the client.
Edit with further testing:
I've synthetically benchmarked now my own setup which consisted of a locally hosted HiveMQ and Mosquitto broker that ran "natively" off the machine. Using the Paho libraries I have sent increasingly bigger messages in batches of 1000. For each batch I calculated the throughput in messages from first to last message. This scenario didn't use any multithreading. With this I've come up with the following performance chart:
The machine running both the client and the brokers is a desktop with an i7 6700 and 32 GB of RAM. The brokers had access to all cores and 8 GB of Memory for its VM.
For benchmarking I used the following code:
import java.util.Random;
import org.eclipse.paho.client.mqttv3.IMqttDeliveryToken;
import org.eclipse.paho.client.mqttv3.MqttCallback;
import org.eclipse.paho.client.mqttv3.MqttClient;
import org.eclipse.paho.client.mqttv3.MqttConnectOptions;
import org.eclipse.paho.client.mqttv3.MqttException;
import org.eclipse.paho.client.mqttv3.MqttMessage;
import org.eclipse.paho.client.mqttv3.MqttPersistenceException;
public class MqttBenchmarker {
protected static int PORT = 1883;
protected static String HOST = "localhost";
protected final String SERVICENAME = "MQTT";
protected static final String COMMANDTOPIC = "commands";
protected static final String REMINDSPREFIX = "Reminds/";
protected static final String VIOLATIONTOPIC = "violations/";
protected static final String WILDCARDTOPIC = "Reminds/#";
protected static final String TCPPREFIX = "tcp://";
protected static final String SSLPREFIX = "ssl://";
private static MqttClient client;
private static MqttConnectOptions optionsPublisher = new MqttConnectOptions();
private static String IDPublisher = MqttClient.generateClientId();
private static int messageReceived = 0;
private static long timesent = 0;
private static int count = 2;
private static StringBuilder out = new StringBuilder();
private static StringBuilder in = new StringBuilder();
private static final int runs = 1000;
private static boolean receivefinished = false;
public static void main(String[] args) {
connect();
Thread sendThread=new Thread(new Runnable(){
#Override
public void run() {
Random rd = new Random();
for (int i = 2; i < 1000000; i += i) {
byte[] arr = new byte[i];
// System.out.println("Starting test run for byte Array of size:
// "+arr.length);
long startt = System.currentTimeMillis();
System.out.println("Test for size: " + i + " started.");
for (int a = 0; a <= runs; a++) {
rd.nextBytes(arr);
try {
client.publish(REMINDSPREFIX, arr, 1, false);
} catch (MqttPersistenceException e) {
// TODO Auto-generated catch block
e.printStackTrace();
} catch (MqttException e) {
// TODO Auto-generated catch block
e.printStackTrace();
}
}
try {
while (!receivefinished) {
Thread.sleep(10);
}
receivefinished = false;
System.out.println("Test for size: " + i + " finished.");
out.append("Sending Payload size: " + arr.length + " achieved "
+ runs / ((System.currentTimeMillis() - startt) / 1000d) + " messages per second.\n");
} catch (InterruptedException e) {
// TODO Auto-generated catch block
e.printStackTrace();
}
}
System.out.println(out.toString());
System.out.println(in.toString());
}
});
sendThread.start();
}
private static class MessageCallback implements MqttCallback {
#Override
public void messageArrived(String arg0, MqttMessage arg1) throws Exception {
if (messageReceived == 0) {
timesent = System.currentTimeMillis();
}
messageReceived++;
if (messageReceived >= runs) {
receivefinished = true;
in.append("Receiving payload size " + count + " achieved "
+ runs / ((System.currentTimeMillis() - timesent) / 1000d) + " messages per second.\n");
count += count;
messageReceived = 0;
}
}
#Override
public void deliveryComplete(IMqttDeliveryToken arg0) {
// TODO Auto-generated method stub
}
#Override
public void connectionLost(Throwable arg0) {
// TODO Auto-generated method stub
}
}
public static void connect() {
try {
client = new MqttClient(TCPPREFIX + HOST + ":" + PORT, IDPublisher);
optionsPublisher.setMqttVersion(MqttConnectOptions.MQTT_VERSION_3_1_1);
optionsPublisher.setAutomaticReconnect(true);
optionsPublisher.setCleanSession(false);
optionsPublisher.setMaxInflight(65535);
client.connect(optionsPublisher);
while (!client.isConnected()) {
try {
Thread.sleep(50);
} catch (InterruptedException e) {
e.printStackTrace();
}
}
client.setCallback(new MessageCallback());
client.subscribe(WILDCARDTOPIC, 0);
} catch (MqttException e1) {
e1.printStackTrace();
}
}
}
What's weird is, that the serialized messages that I want to send from my application only use about 4000 bytes. So the theoretical throughput should be around 200 messages per second. Could this still be a problem caused by longer computations inside the callback function? I've already achieved far better results with the mosquitto broker, and I will further test, how far I can push performance with it.
Thanks for any suggestions!

One problem is in the test setup of the MQTT client.
You are using just one single MQTT client. What you are effectively testing is the size of the MQTT inflight window with this formula:
throughput <= inflight window-size / round-trip time
HiveMQ has by default a property enabled that is called <cluster-overload-protection> that limits the inflight window of a single client.
Additionally the paho client is not really suited for high throughput work in a multi threaded context. A better implementation for high performance scenarios would be the HiveMQ MQTT Client.
With 20 clients connected (10 Publishing and 10 receiving), I reach a sustained throughput of around 6000 qos=1 10kb messages per second.
Disclaimer: I work as a software developer for HiveMQ.

I'm still not exactly sure what the issue was, but switching my broker from HiveMQ to Mosquitto seems to have fixed the issue. Maybe Mosquitto has a different set of default setting compared to HiveMQ, or maybe the free trial version of HiveMQ is limited in some other way than just the number of connected clients.
In any way, Mosquitto worked much better for me, and handled all the messages I could throw at it from my application.

Related

Reducing jitter in Android's WiFi direct communication using UDP transport protocol

I'm working on a real-time application that requires transmitting about 512 bytes of data every 5ms over a WiFi direct connection from a slave phone/s to a master (group owner). Average transmission delay is about 15ms, which is not that good, but the biggest issue is that I noticed a quite consistent pattern of periods where delay is as much as 100ms about every 2~3 seconds, plus periods of about 9 seconds every minute where throughput greatly reduces. Please take a look at the graph scatter graph of delay over time. I also attached my socket programming code. Both master and slave internal clocks are synchronized within a 1ms error (don't worry about clock drift).
Please let me know if you too have experienced these results and whether there is something that can be done to improve on this issue. Interestingly enough there is typically 0 data loss, and when a packet gets delayed all packets before it too (easy to see from the graph) even though this is UDP. I read somewhere that packets need to get acknowledged at the MAC layer even if UDP is used, and this is what might be causing delays of about 100ms.
The following is the code at the slave (sender):
public class DataTransmissionServiceV3 {
public static final String TAG = "DataTransmissionService";
private long mClockOffset;
private DatagramChannel mDatagramChannel;
private boolean running;
public void start(InetAddress remoteAddress, long clockOffset) {
if (!running) {
running = true;
mClockOffset = clockOffset;
Thread t = new Thread(new TransmissionTask(remoteAddress));
t.setPriority(Thread.MAX_PRIORITY);
t.start();
}
}
public void close() {
running = false;
if (mDatagramChannel != null) {
mDatagramChannel.socket().close();
}
}
private class TransmissionTask implements Runnable {
private int mCount;
private final ByteBuffer mByteBuffer = ByteBuffer.allocateDirect(
Constants.DATA_PACKET_SIZE);
private final byte[] mBytes = new byte[Constants.DATA_PACKET_SIZE];
private final InetAddress mRemoteAddress;
public TransmissionTask(final InetAddress remoteAddress) {
mRemoteAddress = remoteAddress;
}
#Override
public void run() {
try {
mDatagramChannel = DatagramChannel.open();
mDatagramChannel.socket().connect(new InetSocketAddress(mRemoteAddress,
Constants.TRANS_MASTER_SERVER_PORT));
mDatagramChannel.configureBlocking(false);
} catch (IOException ioe) {
Log.e(TAG, "Exception while connecting", ioe);
}
while (running) {
TimeUtils.busyWait(5000000); // busy wait for 5ms
send();
}
}
private void send() {
mCount++;
DataPacket.setSequenceNumber(mBytes, mCount);
DataPacket.setTimestamp(mBytes, TimeUtils.getTimestamp() + mClockOffset); // synchronized timestamp to master device to measure delay
mByteBuffer.clear();
mByteBuffer.put(mBytes);
mByteBuffer.flip();
try {
mDatagramChannel.write(mByteBuffer);
} catch (Exception e) {
Log.e(TAG, "Exception while sending data packet", e);
running = false;
}
}
}
}
And the following is the code at the master (receiver):
public class DataTransmissionServerV3 {
public static final String TAG = "DataTransmissionServer";
private DatagramChannel mDatagramChannel;
private boolean mRunning;
private Handler mHandler;
public DataTransmissionServerV3(Handler handler) {
mHandler = handler;
}
public void start(InetAddress localAddress) {
if (!mRunning) {
mRunning = true;
Thread t = new Thread(new ReceiveDataTask(localAddress));
t.setPriority(Thread.MAX_PRIORITY);
t.start();
}
}
public void close() {
mRunning = false;
if (mDatagramChannel != null) {
mDatagramChannel.socket().close();
}
}
private class ReceiveDataTask implements Runnable {
private final InetAddress mLocalAddress;
private final ByteBuffer mByteBuffer = ByteBuffer.allocateDirect(
Constants.DATA_PACKET_SIZE);
private final byte[] mBytes = new byte[Constants.DATA_PACKET_SIZE];
private int mCount;
public ReceiveDataTask(final InetAddress localAddress) {
mLocalAddress = localAddress;
}
#Override
public void run() {
try {
bind();
while (mRunning) {
receive();
}
} catch (IOException ioe) {
Log.e(TAG, "Exception while binding datagram socket", ioe);
}
}
private void bind() throws IOException {
mDatagramChannel = DatagramChannel.open();
mDatagramChannel.socket().bind(new InetSocketAddress(mLocalAddress,
Constants.TRANS_MASTER_SERVER_PORT));
mDatagramChannel.configureBlocking(true);
}
private boolean receive() {
mByteBuffer.clear();
try {
SocketAddress isa = mDatagramChannel.receive(mByteBuffer);
long t2 = TimeUtils.getTimestamp();
if (isa != null) {
mByteBuffer.flip();
mByteBuffer.get(mBytes);
mCount++;
TransmissionStat stat = TransmissionStat.get(mBytes, mCount, t2);
handlePacket(stat); // a statistic that is saved to file for later analysis (ignore this)
return true;
}
} catch (IOException ioe) {
Log.e(TAG, "Exception while receiving data", ioe);
mRunning = false;
}
return false;
}
private void handlePacket(TransmissionStat stat) {
Message msg = mHandler.obtainMessage();
msg.what = Constants.TRANSMISSION_PACKET_RECEIVED_CODE;
msg.obj = stat;
msg.sendToTarget();
}
}

Getting ClassNotFound Exception in Flink SourceFunction

I'm using protocol buffer to send stream of data to Apache Flink.
I have two classes. one is Producer and one is Consumer.
Producer is a java thread class which reads the data from socket and Protobuf deserializes it and then I store it in my BlockingQueue
Consumer is a class which implements SourceFunction in Flink.
I tested this program with using:
DataStream<Event.MyEvent> stream = env.fromCollection(queue);
instead of custom source and it works fine.
But when I try to use a SourceFunction class it throws this exception:
Caused by: java.lang.RuntimeException: Unable to find proto buffer class
at com.google.protobuf.GeneratedMessageLite$SerializedForm.readResolve(GeneratedMessageLite.java:775)
...
Caused by: java.lang.ClassNotFoundException: event.Event$MyEvent
at java.net.URLClassLoader.findClass(URLClassLoader.java:381)
at java.lang.ClassLoader.loadClass(ClassLoader.java:424)
at sun.misc.Launcher$AppClassLoader.loadClass(Launcher.java:331)
...
And in another attempt I mixed both classed into one (the class which implements SourceFunction). I get data from socket and deserialize it with protobuf and store it in BlockingQueue and then I read from BlockingQueue right after that. My code works fine with this approach too.
But I want to use two separate classes (multi-threading) but it throws that exception.
I'm trying to solve it in last 2 days and also did lots of searching but no luck.
Any help would be apperciated.
Producer:
public class Producer implements Runnable {
Boolean running = true;
Socket socket = null, bufferSocket = null;
PrintStream ps = null;
BlockingQueue<Event.MyEvent> queue;
final int port;
public Producer(BlockingQueue<Event.MyEvent> queue, int port){
this.port = port;
this.queue = queue;
}
#Override
public void run() {
try {
socket = new Socket("127.0.0.1", port);
bufferSocket = new Socket(InetAddress.getLocalHost(), 6060);
ps = new PrintStream(bufferSocket.getOutputStream());
while (running) {
queue.put(Event.MyEvent.parseDelimitedFrom(socket.getInputStream()));
ps.println("Items in Queue: " + queue.size());
}
}catch (Exception e){
e.printStackTrace();
}
}
}
Consumer:
public class Consumer implements SourceFunction<Event.MyEvent> {
Boolean running = true;
BlockingQueue<Event.MyEvent> queue;
Event.MyEvent event;
public Consumer(BlockingQueue<Event.MyEvent> queue){
this.queue = queue;
}
#Override
public void run(SourceContext<Event.MyEvent> sourceContext) {
try {
while (running) {
event = queue.take();
sourceContext.collect(event);
}
}catch (Exception e){
e.printStackTrace();
}
}
#Override
public void cancel() {
running = false;
}
}
Event.MyEvent is my protobuf class. I'm using version 2.6.1 and I compiled classes with v2.6.1 . I double checked the versions to be sure it's not the problem.
The Producer class is working fine.
I tested this with both Flink v1.1.3 and v1.1.4.
I'm running it in local mode.
EDIT: Answer was included in question, posted it separately and removed it here.
UPDATE 12/28/2016
...
But I'm still curious. What is causing this error? Is it a bug in Flink or am I doing something wrong?
...
The asker already found a way to make this working. I have extracted the relevant part from the question. Note that the reason why it happened remains unexplained.
I did not use quote syntax as it is a lot of text, but the below was shared by the asker:
So finally I got it to work. I created my BlockingQueue object inside SourceFunction (Consumer), and called Producer class from inside the SourceFunction class (Consumer) instead of making BlockingQueue and calling Producer class in main method of the program. and it now works!
Here's my full working code in Flink:
public class Main {
public static void main(String[] args) throws Exception {
final int port, buffer;
//final String ip;
try {
final ParameterTool params = ParameterTool.fromArgs(args);
port = params.getInt("p");
buffer = params.getInt("b");
} catch (Exception e) {
System.err.println("No port number and/or buffer size specified.");
return;
}
final StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
DataStream<Event.MyEvent> stream = env.addSource(new Consumer(port, buffer));
//DataStream<Event.MyEvent> stream = env.fromCollection(queue);
Pattern<Event.MyEvent, ?> crashedPattern = Pattern.<Event.MyEvent>begin("start")
.where(new FilterFunction<Event.MyEvent>() {
#Override
public boolean filter(Event.MyEvent myEvent) throws Exception {
return (myEvent.getItems().getValue() >= 120);
}
})
.<Event.MyEvent>followedBy("next").where(new FilterFunction<Event.MyEvent>() {
#Override
public boolean filter(Event.MyEvent myEvent) throws Exception {
return (myEvent.getItems().getValue() <= 10);
}
})
.within(Time.seconds(3));
PatternStream<Event.MyEvent> crashed = CEP.pattern(stream.keyBy(new KeySelector<Event.MyEvent, String>() {
#Override
public String getKey(Event.MyEvent myEvent) throws Exception {
return myEvent.getEventType();
}
}), crashedPattern);
DataStream<String> alarm = crashed.select(new PatternSelectFunction<Event.MyEvent, String>() {
#Override
public String select(Map<String, Event.MyEvent> pattern) throws Exception {
Event.MyEvent start = pattern.get("start");
Event.MyEvent next = pattern.get("next");
return start.getEventType() + " | Speed from " + start.getItems().getValue() + " to " + next.getItems().getValue() + " in 3 seconds\n";
}
});
DataStream<String> rate = alarm.windowAll(TumblingProcessingTimeWindows.of(Time.seconds(1)))
.apply(new AllWindowFunction<String, String, TimeWindow>() {
#Override
public void apply(TimeWindow timeWindow, Iterable<String> iterable, Collector<String> collector) throws Exception {
int sum = 0;
for (String s: iterable) {
sum ++;
}
collector.collect ("CEP Output Rate: " + sum + "\n");
}
});
rate.writeToSocket(InetAddress.getLocalHost().getHostName(), 7070, new SimpleStringSchema());
env.execute("Flink Taxi Crash Streaming");
}
private static class Producer implements Runnable {
Boolean running = true;
Socket socket = null, bufferSocket = null;
PrintStream ps = null;
BlockingQueue<Event.MyEvent> queue;
final int port;
Producer(BlockingQueue<Event.MyEvent> queue, int port){
this.port = port;
this.queue = queue;
}
#Override
public void run() {
try {
socket = new Socket("127.0.0.1", port);
bufferSocket = new Socket(InetAddress.getLocalHost(), 6060);
ps = new PrintStream(bufferSocket.getOutputStream());
while (running) {
queue.put(Event.MyEvent.parseDelimitedFrom(socket.getInputStream()));
ps.println("Items in Queue: " + queue.size());
}
}catch (Exception e){
e.printStackTrace();
}
}
}
private static class Consumer implements SourceFunction<Event.MyEvent> {
Boolean running = true;
final int port;
BlockingQueue<Event.MyEvent> queue;
Consumer(int port, int buffer){
queue = new ArrayBlockingQueue<>(buffer);
this.port = port;
}
#Override
public void run(SourceContext<Event.MyEvent> sourceContext) {
try {
new Thread(new Producer(queue, port)).start();
while (running) {
sourceContext.collect(queue.take());
}
}catch (Exception e){
e.printStackTrace();
}
}
#Override
public void cancel() {
running = false;
}
}

Device discovery over Wifi for defined period of time

I am writing a code to send a UDP Multicast over Wifi from my mobile device. There is a server code running on other devices in the network. The servers will listen to the multicast and respond with their IP Address and Type of the system (Type: Computer, Mobile Device, Raspberry Pi, Flyports etc..)
On the mobile device which has sent the UDP Multicast, I need to get the list of the devices responding to the UDP Multicast.
For this I have created a class which will work as the structure of the device details.
DeviceDetails.class
public class DeviceDetails
{
String DeviceType;
String IPAddr;
public DeviceDetails(String type, String IP)
{
this.DeviceType=type;
this.IPAddr=IP;
}
}
I am sending the UDP Multicast packet at the group address of 225.4.5.6 and Port Number 5432.
I have made a class which will call a thread which will send the UDP Packets. And on the other hand I have made a receiver thread which implements Callable Interface to return the list of the devices responding.
Here is the code:
MulticastReceiver.java
public class MulticastReceiver implements Callable<DeviceDetails>
{
DatagramSocket socket = null;
DatagramPacket inPacket = null;
boolean check = true;
public MulticastReceiver()
{
try
{
socket = new DatagramSocket(5500);
}
catch(Exception ioe)
{
System.out.println(ioe);
}
}
#Override
public DeviceDetails call() throws Exception
{
// TODO Auto-generated method stub
try
{
byte[] inBuf = new byte[WifiConstants.DGRAM_LEN];
//System.out.println("Listening");
inPacket = new DatagramPacket(inBuf, inBuf.length);
if(check)
{
socket.receive(inPacket);
}
String msg = new String(inBuf, 0, inPacket.getLength());
Log.v("Received: ","From :" + inPacket.getAddress() + " Msg : " + msg);
DeviceDetails device = getDeviceFromString(msg);
Thread.sleep(100);
return device;
}
catch(Exception e)
{
Log.v("Receiving Error: ",e.toString());
return null;
}
}
public DeviceDetails getDeviceFromString(String str)
{
String type;
String IP;
type=str.substring(0,str.indexOf('`'));
str = str.substring(str.indexOf('`')+1);
IP=str;
DeviceDetails device = new DeviceDetails(type,IP);
return device;
}
}
The following code is of the activity which calls the Receiver Thread:
public class DeviceManagerWindow extends Activity
{
public void searchDevice(View view)
{
sendMulticast = new Thread(new MultiCastThread());
sendMulticast.start();
ExecutorService executorService = Executors.newFixedThreadPool(1);
List<Future<DeviceDetails>> deviceList = new ArrayList<Future<DeviceDetails>>();
Callable<DeviceDetails> device = new MulticastReceiver();
Future<DeviceDetails> submit = executorService.submit(device);
deviceList.add(submit);
DeviceDetails[] devices = new DeviceDetails[deviceList.size()];
int i=0;
for(Future<DeviceDetails> future :deviceList)
{
try
{
devices[i] = future.get();
}
catch(Exception e)
{
Log.v("future Exception: ",e.toString());
}
}
}
}
Now the standard way of receiving the packet says to call the receive method under an infinite loop. But I want to receive the incoming connections only for first 30seconds and then stop looking for connections.
This is similar to that of a bluetooth searching. It stops after 1 minute of search.
Now the problem lies is, I could use a counter but the problem is thread.stop is now depricated. And not just this, if I put the receive method under infinite loop it will never return the value.
What should I do.? I want to search for say 30 seconds and then stop the search and want to return the list of the devices responding.
Instead of calling stop(), you should call interrupt(). This causes a InterruptedException to be thrown at interruptable spots at your code, e.g. when calling Thread.sleep() or when blocked by an I/O operation. Unfortunately, DatagramSocket does not implement InterruptibleChannel, so the call to receive cannot be interrupted.
So you either use DatagramChannel instead of the DatagramSocket, such that receive() will throw a ClosedByInterruptException if Thread.interrupt() is called. Or you need to set a timeout by calling DatagramSocket.setSoTimeout() causing receive() to throw a SocketTimeoutException after the specified interval - in that case, you won't need to interrupt the thread.
Simple approach
The easiest way would be to simply set a socket timeout:
public MulticastReceiver() {
try {
socket = new DatagramSocket(5500);
socket.setSoTimeout(30 * 1000);
} catch (Exception ioe) {
throw new RuntimeException(ioe);
}
}
This will cause socket.receive(inPacket); to throw a SocketTimeoutException after 30 seconds. As you already catch Exception, that's all you need to do.
Making MulticastReceiver interruptible
This is a more radical refactoring.
public class MulticastReceiver implements Callable<DeviceDetails> {
private DatagramChannel channel;
public MulticastReceiver() {
try {
channel = DatagramChannel.open();
channel.socket().bind(new InetSocketAddress(5500));
} catch (IOException ioe) {
throw new RuntimeException(ioe);
}
}
public DeviceDetails call() throws Exception {
ByteBuffer inBuf = ByteBuffer.allocate(WifiConstants.DGRAM_LEN);
SocketAddress socketAddress = channel.receive(inBuf);
String msg = new String(inBuf.array(), 0, inBuf.capacity());
Log.v("Received: ","From :" + socketAddress + " Msg : " + msg);
return getDeviceFromString(msg);;
}
}
The DeviceManagerWindow looks a bit different; I'm not sure what you intend to do there, as you juggle around with lists and arrays, but you only have one future... So I assume you want to listen for 30 secs and fetch as many devices as possible.
ExecutorService executorService = Executors.newFixedThreadPool(1);
MulticastReceiver receiver = new MulticastReceiver();
List<DeviceDetails> devices = new ArrayList<DeviceDetails>();
long runUntil = System.currentTimeMillis() + 30 * 1000;
while (System.currentTimeMillis() < runUntil) {
Future<Object> future = executorService.submit(receiver);
try {
// wait no longer than the original 30s for a result
long timeout = runUntil - System.currentTimeMillis();
devices.add(future.get(timeout, TimeUnit.MILLISECONDS));
} catch (Exception e) {
Log.v("future Exception: ",e.toString());
}
}
// shutdown the executor service, interrupting the executed tasks
executorService.shutdownNow();
That's about it. No matter which solution you choose, don't forget to close the socket/channel.
I have solved it.. you can run your code in following fashion:
DeviceManagerWindow.java
public class DeviceManagerWindow extends Activity
{
public static Context con;
public static int rowCounter=0;
Thread sendMulticast;
#Override
protected void onCreate(Bundle savedInstanceState) {
super.onCreate(savedInstanceState);
setContentView(R.layout.activity_device_manager_window);
WifiManager wifi = (WifiManager)getSystemService( Context.WIFI_SERVICE );
if(wifi != null)
{
WifiManager.MulticastLock lock = wifi.createMulticastLock("WifiDevices");
lock.acquire();
}
TableLayout tb = (TableLayout) findViewById(R.id.DeviceList);
tb.removeAllViews();
con = getApplicationContext();
}
public void searchDevice(View view) throws IOException, InterruptedException
{
try
{
sendMulticast = new Thread(new MultiCastThread());
sendMulticast.start();
sendMulticast.join();
}
catch(Exception e)
{
Log.v("Exception in Sending:",e.toString());
}
here is the time bound search.... and you can quit your thread using thread.join
//Device Will only search for 1 minute
for(long stop=System.nanoTime()+TimeUnit.SECONDS.toNanos(1); stop>System.nanoTime();)
{
Thread recv = new Thread(new MulticastReceiver());
recv.start();
recv.join();
}
}
public static synchronized void addDevice(DeviceDetails device) throws InterruptedException
{
....
Prepare your desired list here.
....
}
}
Dont add any loop on the listening side. simply use socket.receive
MulticastReceiver.java
public class MulticastReceiver implements Runnable
{
DatagramSocket socket = null;
DatagramPacket inPacket = null;
public MulticastReceiver()
{
try
{
socket = new DatagramSocket(WifiConstants.PORT_NO_RECV);
}
catch(Exception ioe)
{
System.out.println(ioe);
}
}
#Override
public void run()
{
byte[] inBuf = new byte[WifiConstants.DGRAM_LEN];
//System.out.println("Listening");
inPacket = new DatagramPacket(inBuf, inBuf.length);
try
{
socket.setSoTimeout(3000)
socket.receive(inPacket);
String msg = new String(inBuf, 0, inPacket.getLength());
Log.v("Received: ","From :" + inPacket.getAddress() + " Msg : " + msg);
DeviceDetails device = getDeviceFromString(msg);
DeviceManagerWindow.addDevice(device);
socket.setSoTimeout(3000)will set the listening time for the socket only for 3 seconds. If the packet dont arrive it will go further.DeviceManagerWindow.addDevice(device);this line will call the addDevice method in the calling class. where you can prepare your list
}
catch(Exception e)
{
Log.v("Receiving Error: ",e.toString());
}
finally
{
socket.close();
}
}
public DeviceDetails getDeviceFromString(String str)
{
String type;
String IP;
type=str.substring(0,str.indexOf('`'));
str = str.substring(str.indexOf('`')+1);
IP=str;
DeviceDetails device = new DeviceDetails(type,IP);
return device;
}
}
Hope that works.. Well it will work.
All the best. Let me know if any problem.

Netty echo server can't handle 100+ channels

I want to create chat-server which would handle 100-500 users in different rooms.
I decided to use Netty framework, because of event-based arcitecture (which is very familiar to me). I started with small server which respond "NYA" for everything it recieve.
main:
public class ChatServer{
public static void main(String[] args) throws Exception {
ExecutorService bossExec = new OrderedMemoryAwareThreadPoolExecutor(1, 400000000, 2000000000, 60, TimeUnit.SECONDS);
ExecutorService ioExec = new OrderedMemoryAwareThreadPoolExecutor(4 /* число рабочих потоков */, 400000000, 2000000000, 60, TimeUnit.SECONDS);
ChannelFactory factory = new NioServerSocketChannelFactory(bossExec,ioExec);
ServerBootstrap bootstrap = new ServerBootstrap(factory);
bootstrap.setPipelineFactory(new ChannelPipelineFactory(){
public ChannelPipeline getPipeline(){
return Channels.pipeline(new ChatChannelHandler());
}
});
bootstrap.setOption("child.tcpNoDelay", true);
bootstrap.setOption("child.keepAlive", true);
bootstrap.setOption("reuseAddress", true);
bootstrap.bind(new InetSocketAddress(5555));
System.out.println("ChatServer started at last...");
}
}
handler:
public class ChatChannelHandler extends SimpleChannelUpstreamHandler {
private final ChannelBuffer messageBuffer = ChannelBuffers.dynamicBuffer();
private boolean processingMessage = false;
private short messageLength;
#Override
public void channelConnected(ChannelHandlerContext ctx, ChannelStateEvent e) throws Exception {
System.out.println("Connection from: " + e.getChannel().getRemoteAddress().toString());
}
#Override
public void messageReceived(ChannelHandlerContext ctx, MessageEvent e) {
e.getChannel().write(ChannelBuffers.copiedBuffer("NYA!", "UTF-8");
}
#Override
public void exceptionCaught(ChannelHandlerContext ctx, ExceptionEvent e) {
System.out.println("Exception: " + e.getCause());
Channel ch = e.getChannel();
ch.close();
}
}
I use simulation program which creates N connections and then randomly write stuff to server with 1-5 second pause. Then i connect with my Flash-based client, to see how it is working. So problem is that under 10-90 connections my flash client immediately get responds from the server, but when number of simulated connections exceed 100 my flash client remain silent. I simply don't understand why.
I figured out, that all messages from my 100+ client gets into buffer, but messageReceive event simply not fired for them. Looks like event queue reachs some limit of registered events, or something like that.
It's really sadden me, because i read about even simpler servers handles 1k+ request per second.
I work under Windows 7, if it is necessary. Also my server never use more then 2-3% of CPU.
My simulation generator:
public class ClientLoadEmulation implements Runnable {
private String host;
private int port;
private static final int minStringLength = 5;
private static final int maxStringLength = 40;
private static final int minPause = (int) (1 * 1000);
private static final int maxPause = (int) (5 * 1000);
Random rand = new Random();
public ClientLoadEmulation(String host, int port, int numThreads) {
this.host = "192.168.0.100";
this.port = 5555;
for (int i = 0; i < numThreads; ++i) {
new Thread(this).start();
try {
Thread.sleep(100);
} catch (Exception ex) {
}
}
}
#Override
public void run() {
byte buffer[] = new byte[255];
try {
Socket s = new Socket(host, port);
InputStream in = s.getInputStream();
OutputStream out = s.getOutputStream();
while (true) {
String govno = "";
...STRING GENERATION STUFF CUTTED...
ByteBuffer LENGTH_BUFFER = ByteBuffer.allocate(2);
LENGTH_BUFFER.putShort((short) govno.length());
LENGTH_BUFFER.flip();
out.write(LENGTH_BUFFER.array());
out.write(govno.getBytes("UTF-8"));
//System.out.println(Thread.currentThread() + " wrote " + govno);
int pause = minPause
+ (int) (rand.nextDouble() * (maxPause - minPause));
try {
Thread.sleep(pause);
} catch (InterruptedException ie) {
}
}
} catch (IOException ie) {
System.out.println("ERROR: " + ie.getMessage());
}
}
static public void main(String args[]) throws Exception {
new ClientLoadEmulation("127.0.0.1", 5555, 100);
}
}
(Sorry for my English skills)
I'm not sure about your sample, but when I tried to run ChatServer you provided, it was not waiting for the client conneciton, but rather quit.
Moreover I don't see there port 5555 setup at all (the one expected from the client).
Please post the working sample to check.

Java: Multithreading & UDP Socket Programming

I am new to multithreading & socket programming in Java. I would like to know what is the best way to implement 2 threads - one for receiving a socket and one for sending a socket. If what I am trying to do sounds absurd, pls let me know why! The code is largely inspired from Sun's tutorials online.I want to use Multicast sockets so that I can work with a multicast group.
class Server extends Thread
{
static protected MulticastSocket socket = null;
protected BufferedReader in = null;
public InetAddress group;
private static class Receive implements Runnable
{
public void run()
{
try
{
byte[] buf = new byte[256];
DatagramPacket pkt = new DatagramPacket(buf,buf.length);
socket.receive(pkt);
String received = new String(pkt.getData(),0,pkt.getLength());
System.out.println("From server#" + received);
Thread.sleep(1000);
}
catch (IOException e)
{
System.out.println("Error:"+e);
}
catch (InterruptedException e)
{
System.out.println("Error:"+e);
}
}
}
public Server() throws IOException
{
super("server");
socket = new MulticastSocket(4446);
group = InetAddress.getByName("239.231.12.3");
socket.joinGroup(group);
}
public void run()
{
while(1>0)
{
try
{
byte[] buf = new byte[256];
DatagramPacket pkt = new DatagramPacket(buf,buf.length);
//String msg = reader.readLine();
String pid = ManagementFactory.getRuntimeMXBean().getName();
buf = pid.getBytes();
pkt = new DatagramPacket(buf,buf.length,group,4446);
socket.send(pkt);
Thread t = new Thread(new Receive());
t.start();
while(t.isAlive())
{
t.join(1000);
}
sleep(1);
}
catch (IOException e)
{
System.out.println("Error:"+e);
}
catch (InterruptedException e)
{
System.out.println("Error:"+e);
}
}
//socket.close();
}
public static void main(String[] args) throws IOException
{
new Server().start();
//System.out.println("Hello");
}
}
First thing is first: your classes should start with a capital letter per the Java Naming Conventions:
Class names should be nouns, in mixed case with the first letter of
each internal word capitalized. Try to
keep your class names simple and
descriptive. Use whole words-avoid
acronyms and abbreviations (unless the
abbreviation is much more widely used
than the long form, such as URL or
HTML).
Second:
Try to break down the code into coherent sections and organize them around some common feature that you're dealing with... perhaps around the functionality or the model you're programming.
The (basic) model for the server is that the only thing it does is receive socket connections... the server relies on a handler to handle those connections and that's it. If you try to build that model it would look something like this:
class Server{
private final ServerSocket serverSocket;
private final ExecutorService pool;
public Server(int port, int poolSize) throws IOException {
serverSocket = new ServerSocket(port);
pool = Executors.newFixedThreadPool(poolSize);
}
public void serve() {
try {
while(true) {
pool.execute(new Handler(serverSocket.accept()));
}
} catch (IOException ex) {
pool.shutdown();
}
}
}
class Handler implements Runnable {
private final Socket socket;
Handler(Socket socket) { this.socket = socket; }
public void run() {
// receive the datagram packets
}
}
Third: I would recommend that you look at some existing examples.
Multi-threaded Client/Server Applications:
http://www.ase.md/~aursu/ClientServerThreads.html
Doug Lea:
http://www.developer.com/java/ent/article.php/3645111/Java-5s-BlockingQueue.htm (thanks to John)
http://gee.cs.oswego.edu/dl/cpj/index.html (still can't find the exact example, but it's there somewhere... if you feel brave look over his allcode.java file).
Concurrency in Practice examples:
http://www.javaconcurrencyinpractice.com/listings.html
Java Concurrency Tutorials:
http://java.sun.com/docs/books/tutorial/essential/concurrency/
Updated per comments:
OK Ravi, there are some big issues with your code and some minor issues with it:
I assume that the Receive class is your client... you should pull that out as a separate program (with its own main class) and run your server and multiple clients at the same time. Spawning a new "client thread" from your server for every new UDP package you send is a disturbing idea (big issue).
When you make your client application, you should make it run the receiving code in its own while loop (minor issue), e.g.:
public class Client extends Thread
{
public Client(/*..*/)
{
// initialize your client
}
public void run()
{
while(true)
{
// receive UDP packets
// process the UDP packets
}
}
public static void main(String[] args) throws IOException
{
// start your client
new Client().start();
}
}
You should only need just one thread per client and one thread per server (you technically don't even a separate thread in there since main has its own thread), so you might not find the ExecutorService that useful.
Otherwise your approach is correct... but I would still recommend that you check out some of examples.
Wanting to create threads in an application is not absurd! You won't need exactly 2 threads, but I think you're talking about 2 classes that implement the Runnable interface.
The threading API has gotten better since Java 1.5 and you don't need to mess with java.lang.Thread anymore. You can simply create a java.util.concurrent.Executor and submit Runnable instances to it.
The book Java Concurrency in Practice uses that exact problem - creating a threaded socket server - and walks through several iterations of the code to show the best way to do it. Check out the free sample chapter, which is great. I won't copy/paste the code here, but look specifically at listing 6.8.
It's a good thing Eclipse's history works even for a day back :) Thanks to that, I am able to give both Ravi a working example and Lirik his answer on leakage.
Let me first start of by stating that I have no clue what is causing this leak, but if I leave it long enough, it will fail on a OutOfMemoryError.
Second, I left the working code commented out for Ravi for a working basic example of my UDP server. The timeout was there to test how long my firewall would kill the receivers end (30 seconds). Just remove anything with the pool, and you're good to go.
So here is, a working but leaking version of my example threaded UDP server.
public class TestServer {
private static Integer TIMEOUT = 30;
private final static int MAX_BUFFER_SIZE = 8192;
private final static int MAX_LISTENER_THREADS = 5;
private final static SimpleDateFormat DateFormat = new SimpleDateFormat("yyyy-dd-MM HH:mm:ss.SSSZ");
private int mPort;
private DatagramSocket mSocket;
// You can remove this for a working version
private ExecutorService mPool;
public TestServer(int port) {
mPort = port;
try {
mSocket = new DatagramSocket(mPort);
mSocket.setReceiveBufferSize(MAX_BUFFER_SIZE);
mSocket.setSendBufferSize(MAX_BUFFER_SIZE);
mSocket.setSoTimeout(0);
// You can uncomment this for a working version
//for (int i = 0; i < MAX_LISTENER_THREADS; i++) {
// new Thread(new Listener(mSocket)).start();
//}
// You can remove this for a working version
mPool = Executors.newFixedThreadPool(MAX_LISTENER_THREADS);
} catch (IOException e) {
e.printStackTrace();
}
}
// You can remove this for a working version
public void start() {
try {
try {
while (true) {
mPool.execute(new Listener(mSocket));
}
} catch (Exception e) {
e.printStackTrace();
}
} finally {
mPool.shutdown();
}
}
private class Listener implements Runnable {
private final DatagramSocket socket;
public Listener(DatagramSocket serverSocket) {
socket = serverSocket;
}
private String readLn(DatagramPacket packet) throws IOException {
socket.receive(packet);
return new BufferedReader(new InputStreamReader(new ByteArrayInputStream(packet.getData())), MAX_BUFFER_SIZE).readLine();
}
private void writeLn(DatagramPacket packet, String string) throws IOException {
packet.setData(string.concat("\r\n").getBytes());
socket.send(packet);
}
#Override
public void run() {
DatagramPacket packet = new DatagramPacket(new byte[MAX_BUFFER_SIZE], MAX_BUFFER_SIZE);
String s;
while (true) {
try {
packet = new DatagramPacket(new byte[MAX_BUFFER_SIZE], MAX_BUFFER_SIZE);
s = readLn(packet);
System.out.println(DateFormat.format(new Date()) + " Received: " + s);
Thread.sleep(TIMEOUT * 1000);
writeLn(packet, s);
System.out.println(DateFormat.format(new Date()) + " Sent: " + s);
} catch (IOException e) {
e.printStackTrace();
} catch (InterruptedException e) {
e.printStackTrace();
}
}
}
}
public static void main(String[] args) {
if (args.length == 1) {
try {
TIMEOUT = Integer.parseInt(args[0]);
} catch (Exception e) {
TIMEOUT = 30;
}
}
System.out.println(DateFormat.format(new Date()) + " Timeout: " + TIMEOUT);
//new TestServer(4444);
new TestServer(4444).start();
}
}
btw. #Lirik, I witnessed this behavior first in Eclipse, after which I tested it from the command line. And again, I have NO clue what is causing it ;) sorry...
2 threads is fine. One reader another writer. Remember that with UDP you should not spawn new handler threads (unless what you're doing takes a long time), I recommend throwing the incoming messages into a processing Queue. The same for the send, have a send thread that blocks on an incoming Queue for UDP send.

Categories

Resources