java.net.SocketException: Too many open files
Nick Gieczewski
My FreeBSD 4.8-STABLE server is running a Java server application that =
has recently had an increase in concurrent users. The JDK is =
jdk-1.4.2p5. Since the user surge, the main thread has been dying =
repeatedly during peak hours at ServerSocket.accept() with the following =
message:
Exception in thread "main" java.net.SocketException: Too many open files
at java.net.PlainSocketImpl.socketAccept(Native Method)
Now, here comes the weird stuff: My system is _not_ running out of file =
descriptors when this happens (kern.maxfiles =3D 16384, kern.openfiles =
=3D ~630), _nor_ is the per-process file descriptor limit being reached =
(it's set to unlimited). The heap is not full, either; its maximum size =
is set to 128 MB and it's under 60 MB when the thread dies. And there's =
plenty of both physical and virtual memory left.
I have written a small program that attempts to open 2,048 concurrent =
connections to the server in oder to reproduce this. On my FreeBSD box, =
the thread dies with the above exception at 500-600 open connections. =
When running the server application on Win32, on the other hand, I'm =
able to open all the 2,048 connections without a hitch.
In a nutshell, I'm getting a "Too many open files" error message but the =
message appears to be lying. The system-wide file descriptor limit is =
far from being reached, the per-process limit can never be reached, the =
heap is far from being full, and there's a reasonable amount of free =
physical and virtual memory. And the problem does not occur when the =
server application is run on the Win32 version of the JDK.
Has anyone run into this problem before, or does anyone know of a fix or =
whether it's been fixed in jdk-1.4.2p6? I could just go ahead and =
compile p6, but it would take many hours and I can't stress the old =
server that much while running this service.
Any input will be greatly appreciated.
Thanks,
Nick
_______________________________________________
freebs...@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-java
To unsubscribe, send any mail to "freebsd-java...@freebsd.org"
Aditya
> On Tue, 16 Mar 2004 23:01:41 -0300, "Nick Gieczewski" <sp...@msn2go.com> said:
> My FreeBSD 4.8-STABLE server is running a Java server application
> that has recently had an increase in concurrent users. The JDK is
> jdk-1.4.2p5. Since the user surge, the main thread has been dying
> repeatedly during peak hours at ServerSocket.accept() with the
> following message:
> Exception in thread "main" java.net.SocketException: Too many open
> files at java.net.PlainSocketImpl.socketAccept(Native Method)
> Now, here comes the weird stuff: My system is _not_ running out of
> file descriptors when this happens (kern.maxfiles = 16384,
> kern.openfiles = ~630), _nor_ is the per-process file descriptor
> limit being reached (it's set to unlimited). The heap is not full,
> either; its maximum size is set to 128 MB and it's under 60 MB when
> the thread dies. And there's plenty of both physical and virtual
> memory left.
I complained about something similar in the diablo 1.3 JDK package in
a message about a week ago and got no replies:
http://lists.freebsd.org/pipermail/freebsd-java/2004-March/001810.html
the crux of the problem seems to be that there is a max file
descriptor setting within the JVM that is set to be very low in the
FreeBSD native JVMs but interestingly enough, not in the linux ones
which I can run under emulation in FreeBSD.
This might not be the particular bug in question, but the test program
does give similar results:
http://developer.java.sun.com/developer/bugParade/bugs/4189011.html
I'd be interested in finding out a solution to this before 1.5 as that
bug suggests.
Thanks,
Adi
Panagiotis Astithas
> I have written a small program that attempts to open 2,048 concurrent
> connections to the server in oder to reproduce this. On my FreeBSD
> box, the thread dies with the above exception at 500-600 open
> connections. When running the server application on Win32, on the
> other hand, I'm able to open all the 2,048 connections without a
> hitch.
Hello,
would you mind posting your small test program? It would help shed some
light I think.
Cheers,
--
Panagiotis Astithas
Electrical & Computer Engineer, PhD
Network Management Center
National Technical University of Athens, Greece