Fixing too many open files Exception(I am using try-catch-finally) - java

I have web service written on JAVA (version 1.8) which connects HSM and sends/receives data via socket. My application is deployed on Apache Tomcat/8.5.14 on linux.
Although I'm closing socket connection properly I have
java.net.SocketException: Too many open files
and here is myclass
public class myClass implements AutoCloseable {
Socket socket;
DataInputStream in;
DataOutputStream out;
public myClass(String ip, int port) throws Exception {
try {
socket = new Socket(ip, port);
in = new DataInputStream(new BufferedInputStream(socket.getInputStream()));
out = new DataOutputStream(new BufferedOutputStream(socket.getOutputStream()));
} catch (IOException e) {
throw new Exception("Connecting to HSM failed" + e);
}
}
public String sendCommandToHsm(String command) throws IOException {
out.writeUTF(command);
out.flush();
return in.readUTF();
}
#Override
public void close() {
if (socket != null && !socket.isClosed()) {
try {
socket.close();
} catch (IOException e) {
lgg.info("Closing of socket failed", e);
}
}
if (in != null) {
try {
in.close();
} catch (IOException e) {
lgg.info("Closing of inputStream failed", e);
}
}
if (out != null) {
try {
out.close();
} catch (IOException e) {
lgg.info("Closing of outputStream failed", e);
}
}
}
}
Here is using of my class
try (MyClass myClass = new MyClass(ip, port);) {
myClass.sendCommandToHsm("my command");
}
I increased maximum open files limit on server from default value(1024) to 8192 and few times later the same Exception occurred again.
I'm thinking about creating Socket Connection Pool, is it good idea?
Can you suggest any other solutions?

Although I'm closing socket connection properly ...
It looks like you are, but I think there are a couple of problems. (I don't know that these are the cause of your leak, but the first one is a plausible explanation.)
Problem 1.
public myClass(String ip, int port) throws Exception {
try {
socket = new Socket(ip, port);
in = new DataInputStream(new BufferedInputStream(socket.getInputStream()));
out = new DataOutputStream(new BufferedOutputStream(socket.getOutputStream()));
} catch (IOException e) {
throw new Exception("Connecting to HSM failed" + e);
}
}
If an exception is thrown during setup of the streams, then the socket will be leaked.
Problem 2.
public void close() {
if (socket != null && !socket.isClosed()) {
try {
socket.close();
} catch (IOException e) {
lgg.info("Closing of socket failed", e);
}
}
if (in != null) {
try {
in.close();
} catch (IOException e) {
lgg.info("Closing of inputStream failed", e);
}
}
if (out != null) {
try {
out.close();
} catch (IOException e) {
lgg.info("Closing of outputStream failed", e);
}
}
}
You are closing in the wrong order. You should1 close in and out before closing socket. In particular, if out has buffered data, then closing out will attempt to flush ... which will fail if you have closed socket already.
Also if socket.close() or in.close() fails for some other reason than an IOException, then the subsequent closes will be skipped. So you should use a finally here.
Also the isClosed() call is redundant. Calling close() on a resource that is already closed should do nothing. This is part of the contract of close().
Finally, calling close() on a socket should2 automatically close the low-level file descriptors beneath in and out. So it is arguably best to just do this:
public void close() {
if (socket != null) {
try {
socket.close();
} catch (IOException e) {
lgg.info("Closing of socket failed", e);
}
}
}
If this doesn't fix your leaks, I suggest that you use netstat and lsof to try to find out if the leakage is open files or open sockets.
I'm thinking about creating Socket Connection Pool, is it good idea?
Yes ... if you can find an existing (well designed and tested) library that meets your requirements. Implementing a reliable pool from scratch is not trivial.
But note:
An incorrectly implemented (or used) pool can leak file descriptors.
The server-side needs to be able to cope with a series of requests / replied on the same connection.
If you have too many simultaneous open connections to different places, then the pool needs a way to close some of them ...
1 - It is debatable whether you should close out before in. On the one hand, closing out flushes outstanding data to the server. On the other hand, by the time that myClass.close() has been called there won't be anything reading the server's response. And besides, the sendCommandToHsm method flushes ... so there shouldn't be any outstanding data.
2 - The javadoc for Socket.close() says: "Closing this socket will also close the socket's InputStream and OutputStream."

Related

Sockets don't work as excpected with LAN or NET, but perfectly locally

I'm programming a network software with Java, but I have a real problem using my application through a "true" network.
Let a software be a host, and listening for client connexions.
Here is my Server loop :
public void run() {
while (mServerSocket != null) {
try {
Socket wClient = mServerSocket.accept();
System.out.println("Client connecté");
wClient.setSoTimeout(50);
wClient.setTcpNoDelay(false);
Client c = new Client(wClient);
synchronized(this) {
mWaitingClients.add(c);
c.start();
}
} catch(Exception ex) {
System.out.println("Server error : " + ex.getMessage());
}
}
}
When a client tried to connect to the server, I use this function :
public Client connect(InetAddress addr, int port) throws Exception {
Socket socket = new Socket(addr, port);
socket.setSoTimeout(50);
socket.setTcpNoDelay(false);
Client c = new Client(socket);
c.start();
return c;
}
And here is the client loop :
public void run() {
try {
ObjectOutputStream out = new ObjectOutputStream(mSocket.getOutputStream());
ObjectInputStream in = new ObjectInputStream(mSocket.getInputStream());
while(mSocket.isConnected() && !mSocket.isClosed()) {
for (int i = 0; i < mOutMessages.size(); i++) {
Message msg = mOutMessages.get(i);
out.writeObject(msg);
}
out.flush();
mOutMessages.clear();
Thread.sleep(50);
out.reset();
while(true) {
try {
Message m = (Message) in.readObject();
mInMessages.add(m);
} catch (Exception e) {
break;
}
}
Thread.sleep(50);
}
} catch(Exception ex) {
try {
mSocket.close();
} catch(Exception exx) {
exx.printStackTrace();
}
ex.printStackTrace();
}
}
Some other parts of the program do Message and put them in the Output list of the Client (mOutMessages).
Some other parts of the program read Message from the mInMessages of the Client.
But something is wrong with this. It works fine locally (server and client on the same computer), but fail or is hazardous (some messages are sent but never received) using two computers (with LAN or through the Internet).
Server ever detect connexions from the clients, send "handshake" messages to the client, but the client never receives them.
I'm more a C programmer than a Java one, and I never had this kind of problem using libc Sockets, so, why my way of doing is wrong ?
Thank you !
Edit :
My Server is created using this function :
public void open(int port) throws Exception {
mServerSocket = new ServerSocket(port);
start(); // Call the run mentionned above.
}
Edit :
Here is my solution, maybe it's not perfect but it works !
public void run() {
try {
BufferedOutputStream buf_out = new BufferedOutputStream(
mSocket.getOutputStream()
);
BufferedInputStream buf_in = new BufferedInputStream(
mSocket.getInputStream()
);
ObjectOutputStream out = new ObjectOutputStream(buf_out);
out.flush();
ObjectInputStream in = new ObjectInputStream(buf_in);
while(mSocket.isConnected() && !mSocket.isClosed()) {
for (int i = 0; i < mOutMessages.size(); i++) {
Message msg = mOutMessages.get(i);
out.writeObject(msg);
out.flush();
}
mOutMessages.clear();
out.reset();
while(true) {
try {
Message m = (Message) in.readObject();
mInMessages.add(m);
} catch (Exception e) {
break;
}
}
}
} catch(Exception ex) {
try {
mSocket.close();
} catch(Exception exx) {
exx.printStackTrace();
}
ex.printStackTrace();
}
If I understand right, both client and server use the run method. If both client and server happen to write sufficiently large messages (not fitting in involved buffers) at the same time then you get a deadlock because neither partner advances to reading (which would drain full buffers). Due to network delays, this might only happen in the non-local scenario, i.e. there may be enough time to pile up enough messages in the mOutMessages buffer.
Note that documentation of Socket.setSoTimeout (which you used) only says that it affects read()s. (For example, in my JDK, ObjectOutputStream seems to use a BlockDataOutputStream with a buffer size of 1024 bytes).
I recommend to either use a separate thread for reading/writing or (if you know the maximum messages size) use a sufficiently large buffer (by wrapping the SocketOutputStream in a BufferedOutputStream). If you opt for larger buffers, you may also want to write one message at a time (and try to read messages after each).

Java Sockets passing between methods

I am just starting out with Java Socket Programming, and I have been reading literature on sockets here. The below code is a sample from a textbook I've taken which asks me to find the bug. Comparing with the literature though I am not seeing any bugs. The creation of the socket, bufferedreader, and printwriter seem correct, and they are surrounded in a try-catch block as well. The are properly "close()"ed in a try-catch block as well. Is there an error when passing these to process()? Any help would be appreciated.
import java.net.*;
import java.io.*;
class main{
public void process(PrintWriter out, BufferedReader in, Socket echoSocket){
//....
}
public void processData() {
Socket echoSocket;
PrintWriter out;
BufferedReader in;
try{
echoSocket = new Socket("server.company.com", 8081);
out = new PrintWriter(echoSocket.getOutputStream(), true);
in = new BufferedReader(new InputStreamReader(echoSocket.getInputStream()));
}
catch (Exception e) {
System.err.println("Exception has occured");
return;
}
process(out, in, echoSocket);
try {
out.close();
in.close();
echoSocket.close();
}
catch(IOException e) {
System.err.println("IOException has occurred.");
}
}
}
Although, typos notwithstanding, one can only guess what the actual "bug" is, this code has an issue with error handling. Specifically, in the disposal of resources.
Discussion about resources
What are resources ?
Basically : any Java Object that relies on underlying OS level resources. Mostly : IO resources (input and output streams, channels), Sockets. But more importantly : if the "thing" you're using has a close, dispsose, shutdown or any of the like, it surely holds on to resources internally.
There are some exceptions (notably ByteArrayInputStream holds no resource but memory), but these are implementation details : if you stick to their interface (and you should, this is a "contract"), every stream should be closed.
Since Java 7, most of these objects in the Java API implement the AutoCloseable interface, but many 3rd parties have not necessarily ported this to their code (and maybe some can't for other reasons).
As one of the code reviewers at my company : I stop reading and I reject any code as soon as I do not see a secure call to the close method of a resource. By secure I mean inside a finally clause, that is guaranteed to be executed.
Rule of thumb about resources
Any resource obtained by your program should be freed in a finally clause (some even add : of its own).
What is the typical lifecycle of a resource
Well:
You obtain it
You use it
You release it
In your code, that is
ResourceObject myObject = null;
try {
myObject = getResource();
processResource(myObject);
} finally {
if(myObject != null) {
try {
myObject.close();
} catch (Exception e) {
// Usually there is nothing one can do but log
}
}
}
Since Java 7, if the resource object implements AutoCloseableyou have a new way of writing that, it's called the "try with resources".
try(ResourceObject myObject = getResource()) {
process(myObject);
}
You do not see the finally, but it's there, the compiler writes the finally clause for you in that case.
What about multiple resources ?
Well : multiple resources, multiple finallys. The idea is to separate the causes of failures in different finally clauses.
Say you want to copy a file...
public void myCopy() throws IOException {
InputStream source = null;
try {
source = new FileInputStream("yourInputFile");
// If anything bad happens, I have a finally clause that protects this now
OutputStream destination = null;
try {
destination = new FileOutputStream("yourOurputFile"); // If fails, my Input will be closed thanks to its own finally
performCopy(source, destination); // If this fail, my destination will also be closed thanks to its own finally
} finally {
if(destination!=null) { try { destination.close(); } catch (Exception e) {/* log*/ }}
}
} finally {
if(source!=null) { try { source.close(); } catch (Exception e) {/* log*/ }}
}
}
Or, with Java 7 syntax, we have the shorter (disclaimer : I have no Java7 right now, so can't really check if this compiles) :
try(
InputStream input = new FileInputStream("in");
OutputStream output = new FileOutputStream("out")) {
performCopy(input, output);
} catch(IOException e) {
// You still have to deal with it of course.
}
This is SO MUCH BOILERPLATE !
Yes it is. That's why we have libraries. One could argue you should not write such code. Use standard, well behaved libraries like commons IO, or use one of their utility methods. Or newer JDK methods like the Files API, and see how this works.
Commons IO has a handy IOUtils.closeQuietly() suite of methods for closing streams.
Try with resources Gotchas
There are ramifications in the "try with resources" call that go a bit deeper than that. These include: What if I want to do something with the exceptions that occur in the finally clause ? How do I differentiate that from an exception that would have occured during performCopy?
Another case is : what happens here :
try(Reader reader = new InputStreamReader(new FileInputStream("in"), "an encoding that is not supported")) {
// Whatever
}
It happens that an UnsupportedEncodingException is thrown but after the FileInputStream is instanciated. But as the FileInputStream is not the subject of the try clause, it will NOT be closed. An you have a File descriptor leak. Try that a thousand times, and your JVM will not be able to open files anymore, you OS will tell you "max number of open files exceeded" (ulimit generally does that in UNIX)
Back to your sockets
So what are your resources ?
Well, first, we can notice that you have only ONE true resource, your Socket instance, because the Socket javadoc says (javadoc):
* <p> Closing this socket will also close the socket's
* {#link java.io.InputStream InputStream} and
* {#link java.io.OutputStream OutputStream}.
So your Input and Output streams are tied to your Socket, and this is enough.
What's wrong with your code
Adding comments one your original code:
try{
echoSocket = new Socket("server.company.com", 8081);
out = new PrintWriter(echoSocket.getOutputStream(), true); // This can throw IOException
in = new BufferedReader(new InputStreamReader(echoSocket.getInputStream())); // Ditto
}
catch (Exception e) {
// If an exception was thrown getting any of the streams, we get there
System.err.println("Exception has occured");
// And you return without closing the socket. It's bad !
return;
}
// Let's assume everything worked, no exception.
process(out, in, echoSocket); // This may throw an exception (timeout, socket closed by peer, ...)
// that is uncaught (no catch clause). Your socket will be left unclosed as a result.
try {
out.close(); // This can fail
in.close(); // This too
echoSocket.close(); // And this too - although nothing you can do about it
}
catch(IOException e) {
// if out.close fails, we get here, and in.close and socket.close
// never got a chance to be called. You may be leaking resources
System.err.println("IOException has occurred.");
}
A safe implementation
Socket echoSocket = null;
try {
// open socket,
echoSocket = new Socket("server.company.com", 8081); // protected by finally
out = new PrintWriter(echoSocket.getOutputStream(), true); // protected
in = new BufferedReader(new InputStreamReader(echoSocket.getInputStream())); // protected
process(out, in, echoSocket); // Still protected
} catch (Exception e) {
// Your current error handling
} finally {
// Anyway, this close will be called if needs be.
if(echoSocket != null) {
try { echoSocket.close(); } catch (Exception e) { /* log */}
// See javadoc, this has closed the in & out streams too.
}
}
public void process(){PrintWriter out, BufferedReader in, Socket echoSocket){
should be
public void process(PrintWriter out, BufferedReader in, Socket echoSocket){
otherwise everything seems fine to me
Try this I think you missed one semicolon
public void processData() {
Socket echoSocket;
PrintWriter out;
BufferedReader in;
try{
echoSocket = new Socket("localhost", 8080);
out = new PrintWriter(echoSocket.getOutputStream(), true);
in = new BufferedReader(new InputStreamReader(echoSocket.getInputStream()));
}
catch (IOException e) {
System.err.println("Exception has occured");
return;
}
process(out, in, echoSocket);
try {
out.close();
in.close();
echoSocket.close();
}
catch(IOException e) {
System.err.println("IOException has occurred.");
}
}
public void process (PrintWriter out, BufferedReader in, Socket echoSocket)
{
}

Should I call reset() before close() on an ObjectOutputSream?

I'm taking over an existing JAVA project which containing the following code:
class ConnectionHandler extends Thread {
private Socket socket;
public ConnectionHandler(Socket s) {
this.socket = s;
}
private void doSthForRequest(ObjectInputStream in, ObjectOutputStream out) throws Exception {
// Do something and write output to out:
// out.writeObject(someOutput);
}
public void run() {
ObjectOutputStream out = null;
ObjectInputStream in = null;
try {
in = new ObjectInputStream(socket.getInputStream());
out = new ObjectOutputStream(socket.getOutputStream());
while (true) {
out.reset();
doSthForRequest(in, out);
}
} catch (Exception ex) {
if (out != null && !socket.isOutputShutdown()) {
try {
out.writeObject(ex);
out.flush();
} catch (Exception ex2) {}
}
} finally {
if (out != null) {
try {
out.reset(); // any reason for this?
} catch (Exception ee) {}
}
if (out != null) {
try {
out.close();
} catch (Exception ee) {}
}
try {
socket.close();
} catch (Exception e) {}
}
socket = null;
}
}
There are ConnectionHandler threads which serving request and producing output on a socket. And my question is:
Does the reset() call still make any sense if there is a close() call immediately after it?
The original author just leaves one line comment // clear outputstream cache which makes me confused...
Appreciate your help!
No. reset() sends a tag over the wire that tells the peer to clear its handle table. As you're about to close the stream anyway, the reset operation has no meaning, and it's an extra network operation to go wrong. Just close it.
As for other problems:
Construct the ObjectOutputStream before the ObjectInputStream. Otherwise a deadlock can occur.
Use the try-with-resources syntax here. It will simplify the code a lot.

How to identify if ObjectInputStream has available object?

Getting error NullPointerException while trying to identify if serialized object is available and receive it using socket. How to identify if ObjectInputStream has available object?
Firs off I try to read a text then try to read from the same socket Lot object ()which may not be there.
public class ThreadIn extends Thread{
BufferedReader in;
PrintStream outConsole;
Socket socket;
ObjectInputStream ois;
String str;
Lot lot;
ThreadIn(BufferedReader input, PrintStream inOutput, Socket s){
str = "";
in= input;
socket= s;
try {
ois = new ObjectInputStream(socket.getInputStream());
} catch (IOException e) {
e.printStackTrace();
}
outConsole = inOutput;
}
public void run() {
while(!EXIT_THREAD){
try {
if(in.ready()){
try {
str= in.readLine();
Thread.sleep(100);
} catch (IOException e) {
EXIT_THREAD= true;
break;
} catch (InterruptedException e) {
e.printStackTrace();
}
outConsole.println("Received:"+str);
}
} catch (IOException e) {
e.printStackTrace();
}
try {
if((Lot)ois.readObject() != null){
lot = (Lot)ois.readObject();
if (lot!=null){outConsole.println(lot.toString());}
outConsole.println((String)ois.readObject());
}
} catch (Exception e) {
e.printStackTrace();
}
}
if((Lot)ois.readObject() != null)
this part itself reads the object from Socket., So you are reading 3 times the object from Socket in your code. If you have only one Object coming in the socket, or more, you can read the object and catch the exception!.
Just like below
//..loop start
try {
lot = (Lot)ois.readObject();
}
catch (Exception e)
{
// do some handling, skip the object! put a continue: or something
}
//do what ever you want to do with `lot`
//..loop end
and now, as per your code, you have not initialized your ObjectInputStream Object.
Do a ObjectInputStream ois = new ObjectInputStream(socket.getInputStream());
If you omitted the code here, well my mistake, else pls do initialize the socket also!
As per the other answers, including the deleted ones, you are calling readObject() twice and throwing the first result away. You should reorganize your code so it can block in readObject().
You have other problems. You are testing the result of readObject() for null?, but it only returns null if you wrote a null at the sender. I suspect you are using this as an EOS test, but it is invalid. readObject() throws EOFException at EOS. You should reorganize your code so it can block in readObject().

How to close ServerSocket connection when catching IOException?

Sorry for question, but I'm totally noob in Java. What is the best practice to execute ServerSocket.close() when caught IOException from ServerSocket? According to docs, ServerSocket.close() throws IOException and compiler asks us to catch it. What is the proper way to close connection on IOException?
try {
server = new ServerSocket(this.getServerPort());
while(true) {
socket = server.accept();
new Handler( socket );
}
} catch (IOException e) {
if (server != null && !server.isClosed()) {
server.close(); //compiler do not allow me to do because I should catch IOExceoption from this method also...
}
}
Thank you!
That's ugly in Java. I hate it, but this is the way you should do it: Wrapping it into another try-catch:
try {
server = new ServerSocket(this.getServerPort());
while(true) {
socket = server.accept();
new Handler( socket );
}
} catch (IOException e) {
if (server != null && !server.isClosed()) {
try {
server.close();
} catch (IOException e)
{
e.printStackTrace(System.err);
}
}
}
If you are going to close the ServerSocket outside of the try{}catch{} anyways, you may as well put it in a finally{}
try {
server = new ServerSocket(this.getServerPort());
while(true) {
socket = server.accept();
new Handler( socket );
}
} catch (IOException e) {
// Do whatever you need to do here, like maybe deal with "socket"?
}
finally {
try {
server.close();
} catch(Exception e) {
// If you really want to know why you can't close the ServerSocket, like whether it's null or not
}
}
In Java SE 7 or later you can use try-with-resources statement, ServerSocket implements java.io.Closeable, so you don't need to explicitly #close() the socket when used in this way.
try (ServerSocket server = new ServerSocket(this.getServerPort())) {
while(true) {
socket = server.accept();
new Handler( socket );
}
} catch (IOException e) {
// It's already closed, just print the exception
System.out.println(e);
}
You can close the resources in the finally block,
http://download.oracle.com/javase/tutorial/essential/exceptions/finally.html
} finally {
try {
socket.close();
} catch(IOException e) {
e.printStackTrace();
}
}

Categories

Resources