/ September 2012 ~ Java EE Support Patterns

9.28.2012

Log4j Thread Deadlock – A Case Study

This case study describes the complete root cause analysis and resolution of an Apache Log4j thread race problem affecting a Weblogic Portal 10.0 production environment. It will also demonstrate the importance of proper Java classloader knowledge when developing and supporting Java EE applications.

This article is also another opportunity for you to improve your thread dump analysis skills and understand thread race conditions.

Environment specifications

  • Java EE server: Oracle Weblogic Portal 10.0
  • OS: Solaris 10
  • JDK: Oracle/Sun HotSpot JVM 1.5
  • Logging API: Apache Log4j 1.2.15
  • RDBMS: Oracle 10g
  • Platform type: Web Portal

Troubleshooting tools

  • Quest Foglight for Java (monitoring and alerting)
  • Java VM Thread Dump (thread race analysis)

Problem overview

Major performance degradation was observed from one of our Weblogic Portal production environments. Alerts were also sent from the Foglight agents indicating a significant surge in Weblogic threads utilization up to the upper default limit of 400.

Gathering and validation of facts

As usual, a Java EE problem investigation requires gathering of technical and non technical facts so we can either derived other facts and/or conclude on the root cause. Before applying a corrective measure, the facts below were verified in order to conclude on the root cause:

  • What is the client impact? HIGH
  • Recent change of the affected platform? Yes, a recent deployment was performed involving minor content changes and some Java libraries changes & refactoring
  • Any recent traffic increase to the affected platform? No
  • Since how long this problem has been observed?  New problem observed following the deployment
  • Did a restart of the Weblogic server resolve the problem? No, any restart attempt did result in an immediate surge of threads
  • Did a rollback of the deployment changes resolve the problem? Yes

Conclusion #1: The problem appears to be related to the recent changes. However, the team was initially unable to pinpoint the root cause. This is now what we will discuss for the rest of the article.

Weblogic hogging thread report

The initial thread surge problem was reported by Foglight. As you can see below, the threads utilization was significant (up to 400) leading to a high volume of pending client requests and ultimately major performance degradation.






As usual, thread problems require proper thread dump analysis in order to pinpoint the source of threads contention. Lack of this critical analysis skill will prevent you to go any further in the root cause analysis.

For our case study, a few thread dump snapshots were generated from our Weblogic servers using the simple Solaris OS command kill -3 <Java PID>. Thread Dump data was then extracted from the Weblogic standard output log files.

Thread Dump analysis

The first step of the analysis was to perform a fast scan of all stuck threads and pinpoint a problem “pattern”. We found 250 threads stuck in the following execution path:

"[ACTIVE] ExecuteThread: '20' for queue: 'weblogic.kernel.Default (self-tuning)'" daemon prio=10 tid=0x03c4fc38 nid=0xe6 waiting for monitor entry [0x3f99e000..0x3f99f970]
       at org.apache.log4j.Category.callAppenders(Category.java:186)
       - waiting to lock <0x8b3c4c68> (a org.apache.log4j.spi.RootCategory)
       at org.apache.log4j.Category.forcedLog(Category.java:372)
       at org.apache.log4j.Category.log(Category.java:864)
       at org.apache.commons.logging.impl.Log4JLogger.debug(Log4JLogger.java:110)
       at org.apache.beehive.netui.util.logging.Logger.debug(Logger.java:119)
       at org.apache.beehive.netui.pageflow.DefaultPageFlowEventReporter.beginPageRequest(DefaultPageFlowEventReporter.java:164)
       at com.bea.wlw.netui.pageflow.internal.WeblogicPageFlowEventReporter.beginPageRequest(WeblogicPageFlowEventReporter.java:248)
       at org.apache.beehive.netui.pageflow.PageFlowPageFilter.doFilter(PageFlowPageFilter.java:154)
       at weblogic.servlet.internal.FilterChainImpl.doFilter(FilterChainImpl.java:42)
       at com.bea.p13n.servlets.PortalServletFilter.doFilter(PortalServletFilter.java:336)
       at weblogic.servlet.internal.FilterChainImpl.doFilter(FilterChainImpl.java:42)
       at weblogic.servlet.internal.RequestDispatcherImpl.invokeServlet(RequestDispatcherImpl.java:526)
       at weblogic.servlet.internal.RequestDispatcherImpl.forward(RequestDispatcherImpl.java:261)
       at <App>.AppRedirectFilter.doFilter(RedirectFilter.java:83)
       at weblogic.servlet.internal.FilterChainImpl.doFilter(FilterChainImpl.java:42)
       at <App>.AppServletFilter.doFilter(PortalServletFilter.java:336)
       at weblogic.servlet.internal.FilterChainImpl.doFilter(FilterChainImpl.java:42)
       at weblogic.servlet.internal.WebAppServletContext$ServletInvocationAction.run(WebAppServletContext.java:3393)
       at weblogic.security.acl.internal.AuthenticatedSubject.doAs(AuthenticatedSubject.java:321)
       at weblogic.security.service.SecurityManager.runAs(Unknown Source)
       at weblogic.servlet.internal.WebAppServletContext.securedExecute(WebAppServletContext.java:2140)
       at weblogic.servlet.internal.WebAppServletContext.execute(WebAppServletContext.java:2046)
       at weblogic.servlet.internal.ServletRequestImpl.run(Unknown Source)
       at weblogic.work.ExecuteThread.execute(ExecuteThread.java:200)
       at weblogic.work.ExecuteThread.run(ExecuteThread.java:172)


As you can see, it appears that all the threads are waiting to acquire a lock on an Apache Log4j object monitor (org.apache.log4j.spi.RootCategory) when attempting to log debug information to the configured appender and log file. How did we figure that out from this thread stack trace? Let’s dissect this thread stack trace in order for you to better understand this thread race condition e.g. 250 threads attempting to acquire the same object monitor concurrently.


At this point the main question is why are we seeing this problem suddenly? An increase of the logging level or load was also ruled out at this point after proper verification. The fact that the rollback of the previous changes did fix the problem did naturally lead us to perform a deeper review of the promoted changes. Before we go to the final root cause section, we will perform a code review of the affected Log4j code e.g. exposed to thread race conditions.

Apache Log4j 1.2.15 code review


## org.apache.log4j.Category
/**
        * Call the appenders in the hierrachy starting at <code>this</code>. If no
        * appenders could be found, emit a warning.
        *
        * <p>
        * This method calls all the appenders inherited from the hierarchy
        * circumventing any evaluation of whether to log or not to log the
        * particular log request.
        *
        * @param event
        *            the event to log.
        */
       public void callAppenders(LoggingEvent event) {
             int writes = 0;

             for (Category c = this; c != null; c = c.parent) {
                    // Protected against simultaneous call to addAppender,
                    // removeAppender,...
                    synchronized (c) {
                           if (c.aai != null) {
                                 writes += c.aai.appendLoopOnAppenders(event);
                           }
                           if (!c.additive) {
                                 break;
                           }
                    }
             }

             if (writes == 0) {
                    repository.emitNoAppenderWarning(this);
             }
            
As you can see, the Catelogry.callAppenders() is using a synchronized block at the Category level which can lead to a severe thread race condition under heavy concurrent load. In this scenario, the usage of a re-entrant read write lock would have been more appropriate (e.g. such lock strategy allows concurrent “read” but single “write”). You can find reference to this known Apache Log4j limitation below along with some possible solutions.


Does the above Log4j behaviour is the actual root cause of our problem? Not so fast…
Let’s remember that this problem got exposed only following a recent deployment. The real question is what application change triggered this problem & side effect from the Apache Log4j logging API?

Root cause: a perfect storm!

Deep dive analysis of the recent changes deployed did reveal that some Log4j libraries at the child classloader level were removed along with the associated “child first” policy. This refactoring exercise ended-up moving the delegation of both Commons logging and Log4j at the parent classloader level. What is the problem?

Before this change, the logging events were split between Weblogic Beehive Log4j calls at the parent classloader and web application logging events at the child class loader. Since each classloader had its own copy of the Log4j objects, the thread race condition problem was split in half and not exposed (masked) under the current load conditions. Following the refactoring, all Log4j calls were moved to the parent classloader (Java EE app); adding significant concurrency level to the Log4j components such as Category. This increase concurrency level along with this known Category.java thread race / deadlock behaviour was a perfect storm for our production environment.

In other to mitigate this problem, 2 immediate solutions were applied to the environment:

  • Rollback the refactoring and split Log4j calls back between parent and child classloader
  • Reduce logging level for some appenders from DEBUG to WARNING
This problem case again re-enforce the importance of performing proper testing and impact assessment when applying changes such as library and class loader related changes. Such changes can appear simple at the "surface" but can trigger some deep execution pattern changes, exposing your application(s) to known thread race conditions.

A future upgrade to Apache Log4j 2 (or other logging API’s) will also be explored as it is expected to bring some performance enhancements which may address some of these thread race & scalability concerns.

Please provide any comment or share your experience on thread race related problems with logging API's.

9.21.2012

OutOfMemoryError: unable to create new native thread – Problem Demystified

As you may have seen from my previous tutorials and case studies, Java Heap Space OutOfMemoryError problems can be complex to pinpoint and resolve. One of the common problems I have observed from Java EE production systems is OutOfMemoryError: unable to create new native thread; error thrown when the HotSpot JVM is unable to further create a new Java thread. 

This article will revisit this HotSpot VM error and provide you with recommendations and resolution strategies. 

If you are not familiar with the HotSpot JVM, I first recommend that you look at a high level view of its internal HotSpot JVM memory spaces. This knowledge is important in order for you to understand OutOfMemoryError problems related to the native (C-Heap) memory space.


OutOfMemoryError: unable to create new native thread – what is it?

Let’s start with a basic explanation. This HotSpot JVM error is thrown when the internal JVM native code is unable to create a new Java thread. More precisely, it means that the JVM native code was unable to create a new “native” thread from the OS (Solaris, Linux, MAC, Windows...).

We can clearly see this logic from the OpenJDK 1.6 and 1.7 implementations as per below:



Unfortunately at this point you won’t get more detail than this error, with no indication of why the JVM is unable to create a new thread from the OS…


HotSpot JVM: 32-bit or 64-bit?

Before you go any further in the analysis, one fundamental fact that you must determine from your Java or Java EE environment is which version of HotSpot VM you are using e.g. 32-bit or 64-bit.

Why is it so important? What you will learn shortly is that this JVM problem is very often related to native memory depletion; either at the JVM process or OS level. For now please keep in mind that:

  • A 32-bit JVM process is in theory allowed to grow up to 4 GB (even much lower on some older 32-bit Windows versions).
  • For a 32-bit JVM process, the C-Heap is in a race with the Java Heap and PermGen space e.g. C-Heap capacity = 2-4 GBJava Heap size (-Xms, -Xmx) – PermGen size (-XX:MaxPermSize)
  • A 64-bit JVM process is in theory allowed to use most of the OS virtual memory available or up to 16 EB (16 million TB)
 As you can see, if you allocate a large Java Heap (2 GB+) for a 32-bit JVM process, the native memory space capacity will be reduced automatically, opening the door for JVM native memory allocation failures.

For a 64-bit JVM process, your main concern, from a JVM C-Heap perspective, is the capacity and availability of the OS physical, virtual and swap memory.


OK great but how does native memory affect Java threads creation?

Now back to our primary problem. Another fundamental JVM aspect to understand is that Java threads created from the JVM requires native memory from the OS. You should now start to understand the source of your problem…

The high level thread creation process is as per below:

  • A new Java thread is requested from the Java program & JDK
  • The JVM native code then attempt to create a new native thread from the OS
  • The OS then attempts to create a new native thread as per attributes which include the thread stack size. Native memory is then allocated (reserved) from the OS to the Java process native memory space; assuming the process has enough address space (e.g. 32-bit process) to honour the request
  • The OS will refuse any further native thread & memory allocation if the 32-bit Java process size has depleted its memory address space e.g. 2 GB, 3 GB or 4 GB process size limit
  • The OS will also refuse any further Thread & native memory allocation if the virtual memory of the OS is depleted (including Solaris swap space depletion since thread access to the stack can generate a SIGBUS error, crashing the JVM * http://bugs.sun.com/view_bug.do?bug_id=6302804

In summary:

  • Java threads creation require native memory available from the OS; for both 32-bit & 64-bit JVM processes
  • For a 32-bit JVM, Java thread creation also requires memory available from the C-Heap or process address space

Problem diagnostic

Now that you understand native memory and JVM thread creation a little better, is it now time to look at your problem. As a starting point, I suggest that your follow the analysis approach below:

  1. Determine if you are using HotSpot 32-bit or 64-bit JVM
  2. When problem is observed, take a JVM Thread Dump and determine how many Threads are active
  3. Monitor closely the Java process size utilization before and during the OOM problem replication
  4. Monitor closely the OS virtual memory utilization before and during the OOM problem replication; including the swap memory space utilization if using Solaris OS

Proper data gathering as per above will allow you to collect the proper data points, allowing you to perform the first level of investigation. The next step will be to look at the possible problem patterns and determine which one is applicable for your problem case.

Problem pattern #1 – C-Heap depletion (32-bit JVM)

From my experience, OutOfMemoryError: unable to create new native thread is quite common for 32-bit JVM processes. This problem is often observed when too many threads are created vs. C-Heap capacity.
JVM Thread Dump analysis and Java process size monitoring will allow you to determine if this is the cause.

Problem pattern #2 – OS virtual memory depletion (64-bit JVM)

In this scenario, the OS virtual memory is fully depleted. This could be due to a few 64-bit JVM processes taking lot memory e.g. 10 GB+ and / or other high memory footprint rogue processes. Again, Java process size & OS virtual memory monitoring will allow you to determine if this is the cause.

Also, please verify if you are not hitting OS related threshold such as ulimit -u or NPROC (max user processes). Default limits are usually low and will prevent you to create let's say more than 1024 threads per Java process.

Problem pattern #3 – OS virtual memory depletion (32-bit JVM)

The third scenario is less frequent but can still be observed. The diagnostic can be a bit more complex but the key analysis point will be to determine which processes are causing a full OS virtual memory depletion. Your 32-bit JVM processes could be either the source or the victim such as rogue processes using most of the OS virtual memory and preventing your 32-bit JVM processes to reserve more native memory for its thread creation process.

Please note that this problem can also manifest itself as a full JVM crash (as per below sample) when running out of OS virtual memory or swap space on Solaris.

#
# A fatal error has been detected by the Java Runtime Environment:
#
# java.lang.OutOfMemoryError: requested 32756 bytes for ChunkPool::allocate. Out of swap space?
#
#  Internal Error (allocation.cpp:166), pid=2290, tid=27
#  Error: ChunkPool::allocate
#
# JRE version: 6.0_24-b07
# Java VM: Java HotSpot(TM) Server VM (19.1-b02 mixed mode solaris-sparc )
# If you would like to submit a bug report, please visit:
#   http://java.sun.com/webapps/bugreport/crash.jsp
#

---------------  T H R E A D  ---------------

Current thread (0x003fa800):  JavaThread "CompilerThread1" daemon [_thread_in_native, id=27, stack(0x65380000,0x65400000)]

Stack: [0x65380000,0x65400000],  sp=0x653fd758,  free space=501k
Native frames: (J=compiled Java code, j=interpreted, Vv=VM code, C=native code)
………………


Native memory depletion: symptom or root cause?

You now understand your problem and know which problem pattern you are dealing with. You are now ready to provide recommendations to address the problem…are you?

Your work is not done yet, please keep in mind that this JVM OOM event is often just a “symptom” of the actual root cause of the problem. The root cause is typically much deeper so before providing recommendations to your client I recommend that you really perform deeper analysis. The last thing you want to do is to simply address and mask the symptoms. Solutions such as increasing OS physical / virtual memory or upgrading all your JVM processes to 64-bit should only be considered once you have a good view on the root cause and production environment capacity requirements.

The next fundamental question to answer is how many threads were active at the time of the OutOfMemoryError? In my experience with Java EE production systems, the most common root cause is actually the application and / or Java EE container attempting to create too many threads at a given time when facing non happy paths such as thread stuck in a remote IO call, thread race conditions etc. In this scenario, the Java EE container can start creating too many threads when attempting to honour incoming client requests, leading to increase pressure point on the C-Heap and native memory allocation. Bottom line, before blaming the JVM, please perform your due diligence and determine if you are dealing with an application or Java EE container thread tuning problem as the root cause.

Once you understand and address the root cause (source of thread creations), you can then work on tuning your JVM and OS memory capacity in order to make it more fault tolerant and better “survive” these sudden thread surge scenarios.

Recommendations:

  • First, quickly rule out any obvious OS memory (physical & virtual memory) & process capacity (e.g. ulimit -u / NPROC) problem.
  • Perform a JVM Thread Dump analysis and determine the source of all the active threads vs. an established baseline. Determine what is causing your Java application or Java EE container to create so many threads at the time of the failure
  • Please ensure that your monitoring tools closely monitor both your Java VM processes size & OS virtual memory. This crucial data will be required in order to perform a full root cause analysis. Please remember that a 32-bit Java process size is limited between 2 GB - 4 GB depending of your OS
  • Look at all running processes and determine if your JVM processes are actually the source of the problem or victim of other processes consuming all the virtual memory
  • Revisit your Java EE container thread configuration & JVM thread stack size. Determine if the Java EE container is allowed to create more threads than your JVM process and / or OS can handle
  • Determine if the Java Heap size of your 32-bit JVM is too large, preventing the JVM to create enough threads to fulfill your client requests. In this scenario, you will have to consider reducing your Java Heap size (if possible), vertical scaling or upgrade to a 64-bit JVM
Capacity planning analysis to the rescue

As you may have seen from my past article on the Top 10 Causes of Java EE Enterprise Performance Problems, lack of capacity planning analysis is often the source of the problem. Any comprehensive load and performance testing exercise should also properly determine the Java EE container threads, JVM & OS native memory requirement for your production environment; including impact measurements of "non-happy" paths. This approach will allow your production environment to stay away from this type of problem and lead to better system scalability and stability in the long run.

Please provide any comment and share your experience with JVM native thread troubleshooting.