I'm trying to better understand exception handling and logging in a j2ee environment to refactor some legacy code (we use log4j for our logging mechanism). Most of our current code does something like the code below on the business tier, however, I'd like to switch over to unchecked exceptions and just ignore them unless it makes sense to handle them somewhere:
try {
doSomething();
} catch (MyException e) {
log.error("Exception:", e);
throw e;
}
After the exception is thrown in the business tier, it is then propagated up to the presentation tier, which again catches the exception and usually wraps it in a PortletException or ServletException and throws it again. Then, it is handled by a Spring handler which shows a 'friendly' message to the user. I'd ultimately like to only handle exceptions for which we want to show a specific error message, and just ignore everything else.
Questions:
Is it necessary to log exceptions in the business tier? If not, do
I need to log exceptions at all (especially unchecked ones)?
What happens to uncaught exceptions that are not logged using log4j?
(If they're still printed in the console, what's the purpose of
log4j?)
I am confused as to how the process works...
Thanks.
EDIT: If an exception occurs in an outside library (Spring, Hibernate, etc), is it assumed that these exceptions will be printed out using whatever logging mechanism is being used? In that case, I guess that I would only need to log the exceptions that my code throws... or am I way off base here?
Before proceeding any further, please take a careful look at:
The following are some of the generally accepted principles of
exception handling:
If you can't handle an exception, don't catch it.
If you catch an exception, don't swallow it.
Catch an exception as close as possible to its source.
Log an exception where you catch it, unless you plan to rethrow it.
Structure your methods according to how fine-grained your exception handling must be.
Use as many typed exceptions as you need, particularly for application exceptions.
Point 1 is obviously in conflict with Point 3. The practical solution is a trade-off >?between how close to the source you catch an exception and how far you let it fall before you've completely lost the intent or content of the original exception.
IBM DeveloperWorks: EJB best practices
It is usually advised that you use checked exceptions for application exceptions at the business tier. I prefer to follow the business interface pattern to decouple the business tier from the user interface and web tier. This will allow me to think of your business tier as a service layer library and callers might want to handle different situations differently when calling this layer. That is one reason you might want to include checked exceptions, since you can react differently to different exceptions. Furthermore, including checked exceptions will usually help the caller code to be better aware of what different situations might arise from invoking some functionality. It could be worth it to take a look at the business delegate pattern and how it might help you with exception handling. In short, the business delegate pattern allows you to create a very thin layer between the business layer and the web layer where you can do things like exception handling.
No matter how you go about doing this, make sure that you understand the implication of adding an application exception to your Java EE application. You may need to investigate how it interacts with your transaction management logic, specifically when it comes to transaction rollbacks. In my line of work, I had to add an #ApplicationException(rollback=false) to forbid the transaction manager from rolling back my transaction when an exception is thrown and propagated upwards.
You may be able to tell I was working with EJB, but the concepts are probably very applicable to your design as well.
So back to your questions:
Is it necessary to log exceptions in the business tier?
It is not necessary if you plan to log it later on. You better devise a logging strategy at a high level and log all caught exceptions there.
If not, do I need to log exceptions at all (especially unchecked
ones)?
I think that you should log exceptions because that will help you debug any issues later on. The user is usually not savvy enough to capture any output that might be produced if the exception propagates and gets printed on his/her screen without you handling it.
What happens to uncaught exceptions that are not logged using log4j?
(If they're still printed in the console, what's the purpose of
log4j?)
I think it will eventually be caught by the web container and be printed out to the console. If an exception propagates upwards and reaches the web container exception handling safety nets, your exception is out of control. It is usually a sign of bad design. It is best if you keep your exceptions under control. Why wonder how a container will react to an uncaught exception? Also how beneficial will that exception be to the user? I think the information presented from uncaught exceptions are almost useless, as they are so far from the source of the error, that they become irrelevant and hard to work with when debugging.
You could create your own exception hierarchy to wrap them to quickly identify from where in your application architecture it is originating. Also one can go a step further to provide codes and reasons that covers majoority of the use cases.
What logging helps you is to identify the sequence of events, when potentially multiple clients are slamming the same use case or hitting a bottle neck. The logging give you a sequence to trace back because requests increasesed there was congestion in this query causing it to timeout and thus other users where seeing another exception.
While handling the application and showing it to the user is another issue.
Cheers!
We definitely do not have to handle exceptions just to log them. I believe we should catch exception if we then throw other exception that contains the source exception as its cause or if we implement some logic that must be implemented in current layer when exception is thrown.
Yes, it is a little bit verbose to declare all methods as throws MyException. This is the reason that Spring (that you use) prefer working with unchecked exceptions. BTW this is the new feature in Java 7: you can ignore exceptions without declaring that method throws it.
I believe that we still need checked exceptions for development of libraries that expose API to 3rd party applications. Application layer exceptions should be mostly runtime and be caught in one central place.
Is it necessary to log exceptions in the business tier?
No. But more logs = better understanding what is going on. In other way more logs = lower performance.
What happens to uncaught exceptions that are not logged using log4j?
You lost them. Logger pretty things that you can save logs in the place where you need and use filters to get only actual logs for now.
You don't need to catch the business exception (as long as it's unchecked). You don't need to handle them or to log them. You can just swallow them. The problem is - what is such exception for?
Exception signals some inappropriate behaviour in the workflow of your application. If it's checked you, typically, can do something about it - try do some operation again, do some workaround, try different action, etc.
If it's unchecked, it is typically an exception you don't know how or can't handle.
It's considered a code smell if all you do is catch an exception and log it. It's not as bad as swallowing it, but still - it's not good.
Some of the containers (i.e. EJB) is required to log occurred exceptions. Moreover, in EJB 3.x if you're in a JTA managed transaction, and you won't catch an unchecked exception which is not marked as #ApplicationException(rollback=false) than the transaction will be automatically rolled back. This might be the reason why you can see some try...catch blocks with just logging code inside.
If you want to get rid of exception handling code in your business logic, you might introduce an interceptor which will react appropriately upon particular exceptions.
HTH!
Firstly, you can and must log the exception stack trace of all exceptions. In fact, IMO if you only log that an exception happened you might as well not log it at all. However, very often what this leads to is a relaxed view of exceptions. What you should strive for is to have 2 log files or one specific category that if exceptions are in that category they mean something critical happened and must be addressed. Even if that means logging the same exception many times. Rather too many than not at all.
Secondly, it's fine to change all exceptions to checked exceptions - the majority of exceptions are not "recoverable". What I have done that worked well is simply wrap all transactions in an exception handling wrapper which logged the exception, then I can guarantee that all exceptions are logged. Furthermore, create a bunch of exception classes that extend runtime exception - this is much better than rethrowing exceptions as runtime exceptions as very often the inner exception stack trace is not logged in full when you wrap exceptions.
And thirdly, it is important to create a mechanism to map exceptions that do filter through to the front end with the back end cause. This is challenging but it's quite important. Errors that the user sees are far easy to track down if you can map them back to an exception stack trace in the log file.
Related
I have a generic question regarding to how to design error handling. I want to use some third-party service in my code. Normally I wrap the service within a client class. Then the rest of my code only deals with my client class and is blind to the real service under the hook. My client class has some mechanism to log errors. But it doesn't want to catch and deal any exception from the service. Ideally it should just ignore the exception handling and let the exception propagates to outside. However, if I want to log the exception, I have to do something like this:
try{
.... // call 3rd party service;
}catch(Exception e){ // e is triggered from the service;
Log.error("Oops, an error: " + e); // shall I log the exception??
throw e; // don't swallow the exception;
}
On one hand, I don't want to do this. I can ignore the handling and logging of the exception. Let the caller of my client class handles exceptions or logs errors. The question is, when should I log exceptions and when should I not? I'd like to hear some common practice and principles. Thank you.
Don't log the exception at every level. Log it only at the "top" level. Log it at the point where not logging it would cause the exception to be missed.
That depends on you. My recommendations are to log the exceptions when they indicate something truly unexpected. Granted, the name "exception" would seem to indicate that this is always the case, but not necessarily.
Sometimes, code returns exceptions for scenarios that are unexeceptional.
Sometimes, code calls other code, knowing it will invoke exceptions in certain cases, but the set of exceptions is "OK".
Sometimes, an exception indicates something unexpected really did happen, and that's usually a bad thing. That's when you should log it, or better yet, make sure your code complains loudly.
If there is a fairly low chance of triggering the exception, I see no real reason not to log it. The only reason to not log an exception to help with debugging is if it is continually firing. If you have an exception that has this characteristic, then I suggest you write your code so that you don't have exceptions continually being triggered.
tl;dr: Exceptions are (hopefully) rare, so log them.
When would you want to propagate an exception to another Class/Method versus catching the Exception in the same Class/Method?
You catch the exception where you have to handle it.
As a rule of thumb, you should let your exceptions bubble up, but if you don't want your subroutine to crash due to a (possibly expected) error, than you handle the exception, which normally involves logging an error and/or displaying an error message to the user.
The good practice is "throw early and catch late". That allows you to better understand the cause of the exception.
This topic is fairly broad; fortunately there are good resources already in place:
Guidelines on Exception propagation (in Java)
http://www.javacodegeeks.com/2012/04/exception-handling-guidelines-best.html
Best practices for exception management in Java or C#
The great majority of exceptions occuring in real-life code are not recoverable in the sense there's any meaningful code that will retry the operation or try to do it differently. The only recovery happening is aborting the current unit of work in an orderly manner—logging the exception, releasing any resources and similar.
This means that, as a first rule, you'll always want to propagate the exception towards that well-defined exception barrier that demarcates your unit of work.
If your code demands anything different than this, it is probably going to be obvious enough, so you don't need to think about it in the general.
I've got a decently complex little game going on in Java (solitaire, essentially like the Windows version), but I have yet to do very much error handling.
Almost all of the methods across my classes will end up either getting called by an initial constructor (eventually main()), a paintComponent() method, or a mouse event. So, my question is, is it bad practice to just use "throws Exception" on all of my lower-level methods, and only do a try/catch at my top-level methods to catch ALL the errors at once? (e.g. 3 try/catches - one for the painting, one for mouse events, one for the main method).
I realize this prevents me from easily dealing with errors on-the-spot, but I don't really plan on doing that anyways. My error handling is going to consist of writing to a log, telling the user, and killing the program. Keeping this in mind, is there anything bad with doing my error handling this way?
It depends on how you want to approach the situation.
If you just want to catch any possible exception and you don't mind about the handler code, you could simply use "throws exception", and there's nothing BAD with it either. It's like a try-catch block that covers all the function.
If you want to write specific code for specific exceptions, you should use try-catch blocks to write appropriate code for each handler.
Based on what you're saying, any caught exception would just notify the user and exit the application. Well, in this case you could just use the first approach. It's not necessarily the BEST approach, and neither is killing the application, however, if that's your strategy you could just use "throws" for each function.
Hope that helps!
If that's all you wan't to do in a case of an error, then it makes perfect sense to do it that way. This eliminates code duplication and scattering of related code. However, if you're thinking of changing how things work in the future (if there's a possibility of this happening), I would suggest to try and push the catch down as far as possible (maybe even eliminating the need for exceptions at all and just logging and exiting right away).
If you use the exception's inner fields (specifically message, which you can set in construction time), you can even eliminate the need for 3 different catch blocks and just use one (depending on your actual actions in case of an error, of course).
I wouldn't - the big reason being that it breaks encapsulation. The reason why this is important in this case is that your error handling code has one of two futures:
Becoming enormous to handle in an informative way every error the program can throw.
Be tiny but not helpful at all: "Some error occurred somewhere".
To my mind, the best structure is to catch the error, log it, alert the user, and exit as far down as possible. There's nothing that says your mouse handling code can't exit, right?
In fact, I would create an error handler class that you can call from just about anywhere, and it handles the notification and logging. Then your exception handlers can just populate it with the message to display/log, and all other code is shared. It will cost you less typing to delegate to this class than adding throws Exception at the end of every function everywhere.
If you must have a top level handler, it should just catch any unexpected runtime errors, so that you can log it and show the user that the program is really quitting for an error, and not just dump to the desktop. All other errors - even if you just want to bail out - should be caught as close to "where the exception has meaning" as possible.
I do the same thing as you are describing most of the time. What you're basically doing is working around the stupid checked exceptions that Java has. It shouldn't even be necessary to add 'throws Exception' everywhere.
If I'm working on an API that other users will use I may create specific exceptions to show what is going on in case they want to handle different exceptions in different ways.
If an error is severe enough to always exit the program, you may be better of throwing a RuntimeException instead. They are used to indicate unrecoverable errors and will avoid problems with putting "throws Exception" everywhere. You should have a handler for RuntimeExceptions anyway to provide a user-friendly error report in case they happen.
If you throw checked exceptions, they should be as specific as possible. Throwing and catching Exception can hide other exceptions that are thrown (including RuntimeExceptions) that you didn't intend and could have been handled differently. If you want to throw a general exception, you can always create your own exception class and throw that instead.
The exception handling can depend on the context so there's not one way to handle everything. If the user clicks a button to open a file and there's an error reading it, then it would be ok to throw an IOException up to the UI layer and display an error message there. On the other hand, an IOException while creating a temporary file could be handled lower down by retrying in another directory.
Say I have 3 tier app- frontend domain and data access. I have read that it is a good idea to catch exceptions high in the call stack...so if I get a data-access exception, the domain layer merely does a finally, like so
try{
}finally{
//cleans up
}
and lets the data-access exception percolate to the frontend layer. Does this not break layering by making the front-end layer deal with the innards ? I think that each layer should either handler or wrap and throw exception that it cannot handle to its calling layer...
any thoughts ?
Lots of good feedback so far, I'll give you my take.
Rule #1. ONLY catch exceptions you are going to actually handle. By handle, I mean handle in such a way that the client's request can continue. You may catch things long enough to log information (don't abuse this, usually the stack is enough information) or to convert to a different error that propagates easier (ala Runtime based). But, if you can't handle it, don't bother catching it. That's just extra code that is useless and confusing. Even when you log or convert, you end up rethrowing.
Realize that most of the time, you can NOT handle an exception. Truly. Many fail to grasp this. But the reality is, if you get an IOException reading or writing to the disk, game over. That request cannot be completed for the user. If your network is flaky and you can't talk to the database, same thing.
Rule #2. When you do get an exception that you cannot handle, the only thing you can do is try to fail in such a way that it is helpful to the user. This means, log it for later analysis (including original stack/cause), and then report something as helpful as possible to the user. Clean up whatever you must, so that the system remains in a consistent state.
Given that this communication with the end user happens at a very high level, that means you usually have to catch at that level. Most of the time, I find that there is very little value in any exception handling between it's inception point and the top level where you catch it for logging and reporting to the user. I often convert to a form of RuntimeException, but that's only done to ease propagation through the layers.
The biggest and most important thing is to realize that you usually can't handle exceptions, so any code you write for them should be as simple as possible.
I don't think layering is such a pure idea that this breaks it.
Wrapping and rethrowing doesn't add much value either.
What's wrong with having the service layer handle exceptions? That ought to be the end of the line, the last line of defense. This design lets the service log the exception - once and for all - and send a user friendly message to the UI for display.
You generally want to catch exceptions higher in the call stack, but only to the point that is makes sense. If the data level can handle and log the exception and just pass a message back to the front-end then that will keep things simple and more flexible.
Personally, if I need to have a try and a finally then I would like to also catch and do something about the situation there rather than pass it up to the caller. Just keep in mind there are always exceptions to good design rules (normally another rule like KISS).
There are three interlocking problems here.
First, constantly re-wrapping exceptions can be done but what value is it providing? You are just creating more layers around the original exception. I only wrap an exception when I can provide additional information about the exception or when the first exception causes another.
Second, the idea of an exception is to respond that a function can not be completed normally. You should catch the exception at the place where it makes the most sense to deal with the problem. If the code has "another alternative" the exception should be trapped at that point. Otherwise log it for the user or developer to work out.
Third, the try/finally block. These are useful when an exception would cause resources to hang out in a open or allocated state. I always use try/finally to clean up resources that might be left open (my favorite is the Statement/ResultSet from java.sql, a huge memory hog). A really good programmer has a lot of this in their code as a way to recover gracefully without creating huge memory leaks or resource constraints.
I am considering a design where all fatal exceptions will be handled using a custom UncaughtExceptionHandler in a Swing application. This will include unanticipated RuntimeExceptions but also custom exceptions which are thrown when critical resources are unavailable or otherwise fail (e.g. a settings file not found, or a server communication error). The UncaughtExceptionHandler will do different things depending on the specific custom exception (and one thing for all the unanticipated), but in all cases the application will show the user an error message and exit. The alternative would be to keep the UncaughtExceptionHandler for all unanticipated exceptions, but handle all other fatal scenarios close to their origin.
Is the design I'm considering sound, or should I use the alternative? What is the typical approach used for handling fatal exceptions?
Usually, it is hard to find a good exception handling strategy. Every approach has its drawbacks. In particular, Yours is good in some sense (a centralized location for handling failures) but suffers from this flaw:
The exception handler you're describing will have special handling for each possible exception. Over time it will become a focal point of your application: every time you add new functionality you will also need to add exception-processing logic to your handler. This means that:
The handler is highly dependents on other parts, a change in impl. of some functionality is likely to trigger corresponding changes in the handler. You will need to be careful in keeping these two in sync.
The handler has poor coherency (has many reasons to change) - it contains the intersection of all features of your app.
Another problem is error recovery. After an exception is thrown (and a some notification is presented to the user), the user want to continue using the application. This means that if your code started modifying the internal data structures and then stopped due to exception, you will need to undo these modifications (or at least get the data structure back to a workable condition) before you allow additional interaction of the user. Achieving this requires a new thinking about the way your data is organized. One possible solution is to DB transactions. On the other hand, this kind of representation is more complicated than plain-old data structures so you need to weigh it against the needs of your app (is it a toy/prototype?)
I have successfully used a mix of local handling and a centralized handling in a large Swing application. The centralized handler only handled I think two or three specific types, and all uncaught exceptions. It's been a while so I don't remember all the details, but we ended up with a centralized handler that took care of two or three specific types of exceptions and all uncaught exceptions.
We used local handling where ever possible, but also defined something like ErrorMessageException that could be thrown also from background threads that did not interact with the UI. This exception had a specific handling in the central handler. We didn't think this is the most beautiful solution available, but it was simple and easy and worked well.
Any uncaught exceptions were handled as "general error" or something similar. Worked well, and of course we tried to release code that did not cause any unhandled exceptions. However, this was very helpful feature in testing.
With this approach the centralized handler was easy to maintain and did not grow unexpectedly. Also we did not see any symptoms of tight coupling, quite the opposite actually.
It would be more straight-forward to wrap the logic of the main method in an try catch block; e.g.
public static void main(String[] args) {
try {
// everything happens here
System.exit(0);
} catch (SpecificException ex) {
...
} catch (AnotherException ex) {
...
} catch (Throwable ex) {
// deal with anything else.
...
}
System.exit(1); // tell the world that we failed.
}
If you have a multithreaded application (like most Swing apps) you might want to consider sending the exceptions to a central exception-handling thread through some async queue.