Skip to content
Switch branches/tags
Go to file

Latest commit


Git stats


Failed to load latest commit information.
Latest commit message
Commit time


ExceptionContext : trying to improve business exceptions.


In "corporate" projects, exceptions are often categorized as TechnicalExceptions and BusinessExceptions.

As technical exceptions usually have a wide application range (eg. DataAccessExceptions), their stacktraces help finding out the exact technical context in which they occurred.

However, for business exceptions, which are often more specific (eg. AuthenticationException) and whose origin in the code is well known, stacktraces are usually useless - but still expensive to generate. What would be more interesting though, is the business context in which they occurred ; for example, knowing that the system failed while processing order #42's payment as part of the daily book-keeping batch is an important information.

To achieve this, a "business context stack" needs to be maintained, on which business-related informations can be pushed and popped as the business processes are respectively started and terminated. This is the role of the BusinessContext class, loosely inspired from Log4J's MDC and Spring Security's SecurityContextHolder.

public class BusinessContext {
    public static void reset();
    public static void push(Supplier<String> context);
    public static void pop();
    public static List<Supplier<String>> get();
    public static void set(List<Supplier<String>> newContext);

The BusinessContext can be managed in two ways :

  • By manually calling the push() and pop() methods at the beginning and end of each interesting method ;
  • Or by annotating those methods with @InBusinessContext and/or @ResetBusinessContext. Those annotations trigger the provided BusinessContextAspect, which, once woven by AspectJ, manages the BusinessContext accordingly for you.

This BusinessContext is then captured by BusinessContextExceptions (and its subclasses) as they are thrown, and displayed in a similar way as traditional stacktraces :

try {
} catch (BusinessException e) {
net.mokatech.exceptioncontext.BusinessException: Oh noes ! A business problem !
 while In method 1 with param hello
 while In method 2 with params hello and 42 
 while In a method throwing BusinessException

How to compile and run

This project is split in two :

  • "ExceptionContext" contains all the "core" classes and the aspect
  • "Test" shows how to use them, manually or via the aspect

Requirements :

  • Java 8
  • Maven 3

Getting started :

  1. Compile the library :
    In the "ExceptionContext" project root : mvn install
  2. Compile the tests :
    In the "Test" project root : mvn compile
  3. Run the tests :
    • Run the tests from your IDE (be sure to compile with Maven only, or enable AspectJ support in the IDE)
    • Run mvn exec:exec to run the TestAspect class

Technical details

This section details some technical points of interest and other glorious hacks.


For the sake of this experiment, BusinessContext maintains a simple List<Supplier<String>> stored in a static ThreadLocal.

In a real system, we would probably want to define more sophisticated data structures as stack elements, and to decouple the data from its storage system. A good inspiration would be Spring Security's SecurityContextHolder.


As stated above, stacktraces are expensive to generate.

As part of the process of instanciating any exception, Throwable's constructor ends up being called. One of its duties is to call the native fillInStackTrace() method, which is responsible for stopping the thread and walking all its stack to collect interesting data such as the class and method names, line numbers etc. As you can imagine, this puts quite a heavy burden on the JVM.

So in the rare event when stacktraces are not needed, an interesting option is to override the fillInStackTrace() method with a no-op one, or to use Throwable's constructor variant that accepts a boolean that controls the stacktrace generation.

By default, BusinessException's constructors use the latter to disable stacktrace generation, but the variants which accept a boolean parameter (withStackTrace) can be used to control that behaviour as needed.


The pointcut intercepting "all methods bearing the InBusinessContext annotation" is defined like this :

 @Around("execution(* *(..)) && @annotation(net.mokatech.exceptioncontext.annotation.InBusinessContext)")

The @annotation(...) part is obvious ; the execution(...) much less.

AspectJ sees a method call as two separate events : a call() in the caller method, and the execution() of the target method.
In both events, the target method would match a simple @annotation(...) pointcut definition, and the aspect would end up being woven twice. To prevent this, an additional call(...) or execution(...) predicate must be added.

Finally, it seems that AspectJ is still unaware of Java 8 features such as lambda expressions or method references. This is why the Supplier pushed onto the context stack is defined as an anonymous inline class, and not a method reference.


Please let me know what you think about this experiment !
Do you think technical and business exceptions should be differentiated and managed differently ? What about the idea of a "business context" (that could also be used elsewhere, eg. in the logs) ?

Do not hesitate to fork and/or comment !


An experiment to explore a new way to handle business exceptions in Java



No releases published


No packages published