Kobe – an example of exception handling done wrong

time to read 3 min | 426 words

In Kobe, pretty much all exception handling is this:

catch (Exception ex)
    bool rethrow = ExceptionPolicy.HandleException(ex, "Service Policy");
    if (rethrow && WebOperationContext.Current == null)
    return null;

These lines appear in just about any method in the application.

They are wrong. In fact, they are wrong on more than one level. Violating DRY is a big offence, but even ignoring that, they are still wrong.

Exception handling is not something that you should take lightly, and there is a strong tie between the way that you write the code and the way errors are handled. The whole purpose of this exception handling is to avoid throwing an exception by setting some configuration parameter.

The problem with this is that this purpose is wrong. This is a case of writing the code without actually trying to understand what the implications are. You cannot do this sort of thing, because the calling code never expects a null, it expect a valid result or an exception.

This approach bring us back to the bad old days of error codes. Not to mention that even if we actually checked for a null return value, what the hell is the error that actually caused this to happen.  Oh, it was put in the log, so now we need to go and look at the log file and try to match timestamps (if we even can).

Exception handling should appear in exactly two places:

  • When an error is expected (making a web request call, for example) and there is some meaningful behavior to be done in the case of a failure (such as retrying after some delay)
  • On a system boundary, in which case you need to make a decision about how you are going to expose the error to the outside world.

It is important to note that I explicitly used the term system boundary, instead of service boundary, which is more commonly used. Exception handling within a system should not try to hide errors. Error should propagate all the way to the system boundary, because otherwise we lose a lot of important information.

System boundary is defined as whatever is visible to the actual user or other systems.

Oh, and exception handling in the system boundary is a system infrastructure problem, it is not something that you have to write code for all over the place.