Every RuntimeException Is a Bug
Properly written Java software uses checked exceptions to indicate environmental problems external to the program such as an I/O error and uses runtime exceptions to indicate problems inside the program itself, in other words bugs. Every time a run-time exception is thrown there’s a bug somewhere. Sometimes the bug is inside the method that throws the exception. Sometimes the bug is in the method that invokes the method that throws the exception. Sometimes the bug is in a third-party library. Sometimes the bug is that someone is trying to do something in a place where they’re not allowed to do it, for instance open files inside a compare method. But make no mistake: if a run-time exception is thrown, there’s a bug somewhere.
We have not yet learned how to write perfectly bug free software. I suspect we never will. Defense in depth suggests that it’s a good idea to have a try–catch block near the top of your application — for instance in the main method or in the run method — that will catch any runtime exceptions you weren’t expecting, log them, and perform any cleanup it reasonably can. However, this is not enough. It is also necessary that someone carefully read the logs and fix the bugs that caused the runtime exceptions in the first place. Otherwise, you might as well not be logging them in the first place.
A disturbing number of third-party libraries have started using runtime exceptions where they shouldn’t be. In particular, they’re using runtime exceptions to report errors communicating with external services such as databases and DNS servers. (Yes I’m talking to you Hibernate). If you’re using such a library — although I recommend you don’t — then you will find a lot of these exceptions in your logs. In this case, the logs are warning you that you’re missing a catch block deeper inside the code. It’s annoying that you find this out at run time rather than at compile time as you could have if the library had used the right kind of exception in the first place, but better late than never.
One thing you should not do is include checked exceptions in the same log as the runtime exceptions. They mean different things and they require different responses. Most checked exceptions indicate environmental problems that take care of themselves. Database connection errors are one frequent example. In fact, any network service is simply going to fail some of the time. That’s just how TCP/IP works. Unless it’s failing all the time or very frequently, it’s not worth worrying about. You don’t want to clutter up your logs with a lot of detail you’re not going to do anything about. Limit the error logs to the problems you actually need to resolve, and every runtime exception that is thrown is a real problem that you actually need to resolve.
Here’s another way of thinking about it: a problem you’re just going to catch and ignore should be a checked exception. No runtime exception should ever be ignored. Runtime exceptions should be subjected to postmortems and fixed in the code so they cannot reoccur. Architect your code so that runtime exceptions are logically impossible.
Of course, not all checked exceptions should be ignored. Many of them indicate serious problems that require handlers to deal with. But if an exception can sometimes be ignored, then that exception is a checked exception. Leaving a code path that is known to throw runtime exceptions in your program is leaving bugs in your program. Every uncaught runtime exception, every exception that bubbles up into your main loop unexpectedly is a bug, and it should be fixed.
