WebLogic Server

A "StackOverFlow" message is usually indicative of an error in the application code of the user, an error in the Java Virtual Machine, or in BEA WebLogic Server itself.

This message is usually seen right before a Java Virtual Machine core dump or the WebLogic Server process just "goes away." It is because of either an unintentional recursive call in user/application code or a scenario where arrays of arrays of Objects can cause the stack to overflow (there are bug reports on http://java.sun.com about these types of issues). This is unfortunate, as it may require programmers to think about the implementation details of the Java Virtual Machine on which they are running. In order to investigate a "StackOverFlow" Error more thoroughly to determine the exact cause, first go through the items discussed here.

Look at any recent application code changes and see if anything could possibly be called recursively. If there is no stack trace produced for the "StackOverflow", try to add debug statements where there is suspect, code. If some application code is suspect, then you can make modifications and add the following:

catch ( StackOverflowError e ) {
System.err.println("Exception: " + e );
// Here is the important thing to do
// when catching StackOverflowError's:
e.printStackTrace();
// do some cleanup, destroy the thread or unravel if possible.
}

The next thing you can look at involves JSPs (if you're using them). A few issues involving recursive problems have been resolved by using the following information for JSPs:

• Check within your jsp_error page to see if you have the tag <%@ page errorPage="jsp_error"%> as this would cause an infinite recursion. Please remove this and simply print the stack trace of any exceptions that could occur. This way you can find the problem in the error page. Once this is resolved you can look at the original error that gets sent to this page.
• Feedback provided to BEA Customer Support about the JSP tag was that another problem was due to bad jsp/servlet login/auth/error reporting code. Fixing this code fixed the recursion and therefore the crash.
• If you are using BEA WebLogic JSP Form Validation Tags, make sure that for the <wl:form> you don't set the action attribute to the same page containing the <wl:form> tag because this will create an infinite loop resulting in a "StackOverFlow" exception. For an example, see http://e-docs.bea.com/wls/docs61/jsp/validation_tags.html#67370.

A strange "StackOverFlow" was caused by the following code snippet within some application code. Please check to see if you are doing this in your code since a known Sun Issue (#4906193) addresses it. Instead of doing this in the application code:

Properties p = new Properties(System.getProperties());

do the following:

Properties p = new Properties();
p = System.getProperties();

in order to avoid the recursive call stack trace that was observed.

A suggested "possible" workaround for some "StackOverFlow" messages is to increase the size of the thread stacks with the -Xss argument to the JVM. However, if a recursive call truly has caused this, then this option will not really help at all and will only delay the inevitable. Some background on this argument to the JVM: each Java thread has two stacks, one for Java code and one for C code. This option sets the maximum stack size that can be used by C code in a thread to the value specified. For a complete definition of the "-Xss" flag see "Non-Standard Options" at http://java.sun.com/j2se/1.3/docs/tooldocs/solaris/java.html.

If the first couple of suggestions don't help pinpoint the problem, then try periodically collecting thread dumps of the JVM as it is running when you think the problem may occur (if it occurs at a specific time or a specific sequence of events causes the problem), usually about 5-10 seconds apart. Using this information, you may be able to find the recursive code and correct it or this may enable BEA Customer Support to have a better idea of what could be causing the problem. To collect thread dumps you need to do the following on the Java Process ID (PID):

• On all "Unix-like" platforms you can do "kill -3" before expecting the crash. "kill -3 <jvm-pid>" dumps java threads.
• On all "Windows" platforms you can do a "kill -3" on the JVM PID. You can also do a "<CTRL> <BREAK>" in the window where the JVM is running.

The same applies to the BEA JRockit JVM as well as all platforms. You need to make sure that you do this on the root java process. To get a tree structure of the processes on Linux, use the "--forest" option. For example, to find the processes started by user "wlsuser", execute this: "ps -lU wlsuser -forest". For a specific example of getting thread dumps with the BEA JRockit JVM on Linux, see http://e-docs.bea.com/wls/docs70/cluster/trouble.html#602852.

If you cannot "time" the thread dumps to get a thread dump right before the "StackOverFlow" happens, you can set the following flags to allow a thread dump to be taken of the server right before a core happens to get the state of the threads at that moment. The option is "-XX:+ShowMessageBoxOnError" option on the Sun JVM (which is not officially documented on Sun's Web site). When the JVM crashes, the program will prompt: "Do you want to debug the problem?" You can then take a thread dump of the JVM. This option will be available on the 8.1 SP2 version of the BEA JRockit JVM when it is released. However, in that version the corresponding option will be "-Djrockit.waitonerror".

If a binary core file is produced from a "StackOverFlow", then you can run a debugger on the resulting core file to get a stack trace. This may help in pointing out the offending code to you. If you are unsure, then contact BEA Customer Support with this information so they can investigate the stack trace more thoroughly. If you are on a Windows platform, then a "Dr. Watson" file may be produced so please send this file to BEA Customer Support when opening a case. Otherwise, check the following "Unix" operating system values to make sure that they have already been properly set in order to generate a core file:

1. Check the "ulimit -c" (configured size of the core file) at a system and user level to make sure that it is set and that the value is not set too low to produce a meaningful core file.
2. Check the available disk space for the user. For example: Is there a disk quota?
3. Check the following parameter, which on Solaris is usually in /etc/system file and can be used to disable core files:

   set sys:coredumpsize=0

4. On Linux, the coredump is turned off by default on all systems. In RedHat Advanced Server 2.1 it should be under "/etc/security". There should be a self-explanatory file called limits.conf; within that file look for the word "core". If it is set to "0", then coredump files are disabled.
5. On HP-UX check the setting called "kernel parm maxdsiz" (max_per_proc_data_size, which increases the User Process Data Segment Size) from the old value of, say, 64m, to something higher like 134M.

Please get a stack trace (or back trace) from your debugger. For example, here are the commands needed when using "dbx" or "gdb".

dbx

• $ java -version: Need to use right version of JDK
• $ ls /opt/bin/dbx: Need to know dbx location or "which dbx"
• $ export DEBUG_PROG=/opt/bin/dbx: Wherever "dbx" is located
• $ <path to java command>/java corefile

Now you are in the debugger. Execute the following commands:

• (dbx) where: Shows a summary of the stack)
• (dbx) threads: Shows the state of the existing threads
• (dbx) quit

gdb

• $ java -version: Need to use right version of JDK
• $ ls /usr/local/bin/gdb: Need to know gdb location or "which gdb"
• $ export DEBUG_PROG=/usr/local/bin/gdb: Wherever "gdb" is located
• $ <path to java command>/java corefile

Now you're in the debugger. Execute the following commands:

• (gdb) where: Shows a summary of the stack
• (gdb) thr: Switch among threads or show the current thread
• (gdb) info thr: Inquire about existing threads
• (gdb) thread apply 1 bt: Apply a command to a list of threads, specifically the back trace to thread #1
• (gdb) quit

Further Information
For additional information you can also go to http://support.bea.com and find some published solutions on "StackOverflows". In the "Question" field type "S-19795" or "S-19361" to display the information from those solutions.

If none of these hints direct you towards a solution or an identifier in your application, then contact BEA Customer Support for further diagnosis. You can open a case with a valid support contract by logging in at http://support.bea.com/login.jsp

Add Your Feedback

In order to post a comment you need to be registered and logged in.

Register | Log in

Please wait while we process your request...