Why does the the Java VM not recover after "Too many open files" errors?
- by Michael
In certain well-understood circumstances, our application will open too many sockets (database connections) and reach the maximum open files that the OS allows. We understand this; we are fixing the issue and also bumping up the limit.
What we can't explain is why parts of our application don't recover even after the number of connections abates and we're well within the limit.
In this case, it's an application running under Tomcat.
When this happens, we first start seeing "Too many open files" errors:
SEVERE: Socket accept failed
java.net.SocketException: Too many open files
at java.net.PlainSocketImpl.socketAccept(Native Method)
at java.net.PlainSocketImpl.accept(PlainSocketImpl.java:390)
at java.net.ServerSocket.implAccept(ServerSocket.java:453)
at java.net.ServerSocket.accept(ServerSocket.java:421)
at org.apache.tomcat.util.net.DefaultServerSocketFactory.acceptSocket(DefaultServerSocketFactory.java:61)
at org.apache.tomcat.util.net.JIoEndpoint$Acceptor.run(JIoEndpoint.java:310)
at java.lang.Thread.run(Thread.java:619)
Eventually, we start seeing NoClassDefFoundErrors inside an application thread that's trying to open HTTP connections:
java.lang.NoClassDefFoundError: org/apache/commons/httpclient/protocol/ControllerThreadSocketFactory
at org.apache.commons.httpclient.protocol.DefaultProtocolSocketFactory.createSocket(DefaultProtocolSocketFactory.java:128)
at org.apache.commons.httpclient.HttpConnection.open(HttpConnection.java:707)
at org.apache.commons.httpclient.MultiThreadedHttpConnectionManager$HttpConnectionAdapter.open(MultiThreadedHttpConnectionManager.java:1349)
[...]
Caused by: java.lang.ClassNotFoundException: org.apache.commons.httpclient.protocol.ControllerThreadSocketFactory
at org.apache.catalina.loader.WebappClassLoader.loadClass(WebappClassLoader.java:1387)
at org.apache.catalina.loader.WebappClassLoader.loadClass(WebappClassLoader.java:1233)
at java.lang.ClassLoader.loadClassInternal(ClassLoader.java:320)
... 8 more
When the errant connections go away, the server starts accepting connections again, and everything seems ok, but we're left with the latter error constantly being spewed to stderr.
Although the application typically logs unloaded classes to stdout, I don't see any such logs just before, during or after the "Too many open files" errors.
My initial theory was that the Hotspot JVM would unload seemingly unused classes when it encounters "Too many open files," but if so, it doesn't log the fact. I'd also expect it to recover if that were the case.
Platform details:
Java(TM) SE Runtime Environment (build 1.6.0_14-b08)
Java HotSpot(TM) 64-Bit Server VM (build 14.0-b16, mixed mode)
Apache Tomcat Version 6.0.18
