Enterprise Java

Filter irrelevant stack trace lines in logs

I love stack traces. Not because I love errors, but the moment they occur, stack trace is priceless source of information. For instance in web application the stack trace shows you the complete request processing path, from HTTP socket, through filters, servlets, controllers, services, DAOs, etc. – up to the place, where an error occurred. You can read them as a good book, where every event has cause and effect. I even implemented some enhancements in the way Logback prints exceptions, see Logging exceptions root cause first.

But one thing’s been bothering me for a while. The infamous “stack trace from hell” symptom – stack traces containing hundreds of irrelevant, cryptic, often auto-generated methods. AOP frameworks and over-engineered libraries tend to produce insanely long execution traces. Let me show a real-life example. In a sample application I am using the following technology stack:

Colours are important. According to framework/layer colour I painted a sample stack trace, caused by exception thrown somewhere deep while trying to fetch data from the database:

No longer that pleasant, don’t you think? Placing Spring between application and Hibernate in the first diagram was a huge oversimplification. Spring framework is a glue code that wires up and intercepts your business logic with surrounding layers. That is why application code is scattered and interleaved by dozens of lines of technical invocations (see green lines). I put as much stuff as I could into the application (Spring AOP, method-level @Secured annotations, custom aspects and interceptors, etc.) to emphasize the problem – but it is not Spring specific. EJB servers generate equally terrible stack traces (…from hell) between EJB calls. Should I care? Think about it, when you innocently call BookService.listBooks() from BookController.listBooks() do you expect to see this?

at com.blogspot.nurkiewicz.BookService.listBooks()
at com.blogspot.nurkiewicz.BookService$$FastClassByCGLIB$$e7645040.invoke()
at net.sf.cglib.proxy.MethodProxy.invoke()
at org.springframework.aop.framework.Cglib2AopProxy$CglibMethodInvocation.invokeJoinpoint()
at org.springframework.aop.framework.ReflectiveMethodInvocation.proceed()
at org.springframework.aop.aspectj.MethodInvocationProceedingJoinPoint.proceed()
at com.blogspot.nurkiewicz.LoggingAspect.logging()
at sun.reflect.NativeMethodAccessorImpl.invoke0()
at sun.reflect.NativeMethodAccessorImpl.invoke()
at sun.reflect.DelegatingMethodAccessorImpl.invoke()
at java.lang.reflect.Method.invoke()
at org.springframework.aop.aspectj.AbstractAspectJAdvice.invokeAdviceMethodWithGivenArgs()
at org.springframework.aop.aspectj.AbstractAspectJAdvice.invokeAdviceMethod()
at org.springframework.aop.aspectj.AspectJAroundAdvice.invoke()
at org.springframework.aop.framework.ReflectiveMethodInvocation.proceed()
at org.springframework.aop.interceptor.AbstractTraceInterceptor.invoke()
at org.springframework.aop.framework.ReflectiveMethodInvocation.proceed()
at org.springframework.transaction.interceptor.TransactionInterceptor.invoke()
at org.springframework.aop.framework.ReflectiveMethodInvocation.proceed()
at org.springframework.aop.interceptor.ExposeInvocationInterceptor.invoke()
at org.springframework.aop.framework.ReflectiveMethodInvocation.proceed()
at org.springframework.aop.framework.Cglib2AopProxy$DynamicAdvisedInterceptor.intercept()
at com.blogspot.nurkiewicz.BookService$$EnhancerByCGLIB$$7cb147e4.listBooks()
at com.blogspot.nurkiewicz.web.BookController.listBooks()

And have you even noticed there is custom aspect in between? That’s the thing, there is so much noise in the stack traces nowadays that following the actual business logic is virtually impossible. One of the best troubleshooting tools we have is bloated with irrelevant framework-related stuff we don’t need in 99% of the cases.

Tools and IDEs are doing a good job of reducing the noise. Eclipse has stack trace filter patterns for Junit, IntelliJ IDEA supports console folding customization. See also: Cleaning noise out of Java stack traces, which inspired me to write this article. So why not having such possibility at the very root – in the logging framework such as Logback?

I implemented a very simple enhancement in Logback. Basically you can define a set of stack trace frame patterns that are suppose to be excluded from stack traces. Typically you will use package or class names that you are not interested in seeing. This is a sample logback.xml excerpt with the new feature enabled:

<root level="ALL">
  <appender name="STDOUT" class="ch.qos.logback.core.ConsoleAppender">
    <encoder>
      <pattern>%d{HH:mm:ss.SSS} | %-5level | %thread | %logger{1} | %m%n%rEx{full,
          java.lang.reflect.Method,
          org.apache.catalina,
          org.springframework.aop,
          org.springframework.security,
          org.springframework.transaction,
          org.springframework.web,
          sun.reflect,
          net.sf.cglib,
          ByCGLIB
        }
        </pattern>
    </encoder>
  </appender>
</root>

I am a bit extreme in filtering almost whole Spring framework + Java reflection and CGLIB classes. But it is just to give you an impression how much can you get. The very same error after applying my enhancement to Logback:

Just as a reminder, green is our application. Finally in one place, finally you can really see what was your code doing when an error occurred:

at com.blogspot.nurkiewicz.DefaultBookHelper.findBooks()
at com.blogspot.nurkiewicz.BookService.listBooks()
at com.blogspot.nurkiewicz.LoggingAspect.logging()
at com.blogspot.nurkiewicz.web.BookController.listBooks()

Simpler? If you like this feature, I opened a ticket LBCLASSIC-325: Filtering out selected stack trace frames. Vote and discuss. This is only a proof-of-concept, but if you like to have a look at the implementation (improvements are welcome!), it is available under my fork of Logback (around 20 lines of code).

Reference: Filtering irrelevant stack trace lines in logs from our JCG partner Tomasz Nurkiewicz at the Java and neighbourhood blog.

Tomasz Nurkiewicz

Java EE developer, Scala enthusiast. Enjoying data analysis and visualization. Strongly believes in the power of testing and automation.
Subscribe
Notify of
guest

This site uses Akismet to reduce spam. Learn how your comment data is processed.

2 Comments
Oldest
Newest Most Voted
Inline Feedbacks
View all comments
Michael Gantman
Michael Gantman
10 years ago

I agree. I wrote a java lib that filters the stacktrace and it could be integrated with log4j. Only my lib takes the package prefix that you would be interested in and it cuts the rest. but the idea is the same

Michael Gantman
Michael Gantman
8 years ago

I actually wrote a library (https://github.com/michaelgantman/Mgnt/releases/tag/1.01) that contains several utilities. One of them is a general purpose stacktrace filter that I used extensively and found it very useful. The class is called TextUtils and it has method getStacktrace() with several overridden signatures. It takes a Throwable instance and allows to set a package prefix of the packages that are relevant. Let’s say your company’s code always resides in packages that start with “com.plain.*” So you set such a prefix and do this logger.info(TextUtils.getStacktrace(e, true, “com.plain.”)); this will filter out very smartly all the useless parts of the trace leaving you… Read more »

Back to top button