Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 16 Mar 2004 23:01:41 -0300
From:      "Nick Gieczewski" <spam@msn2go.com>
To:        <freebsd-java@freebsd.org>
Subject:   java.net.SocketException: Too many open files
Message-ID:  <011a01c40bc3$cd534050$0200a8c0@rescue>

next in thread | raw e-mail | index | archive | help
Hello folks,

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



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?011a01c40bc3$cd534050$0200a8c0>