How to handle db constraint violations in the user

问题:

We implement the majority of our business rules in the database, using stored procs.

I can never decide how best to pass data constraint violation errors from the database back to the user interface. The constraints I'm talking about are tied more to business rules than data integrity.

For example, a db error such as "Cannot insert duplicate key row" is the same as the business rule "you can't have more than one Foo with the same name". But we've "implemented" it at the most common sense location: as a unique constraint that throws an exception when the rule is violated.

Other rules such as "You're only allowed 100 Foos per day" do not cause errors per-say, since they're gracefully handled by custom code such as return empty dataset that the application code checks for and passes back to the ui layer.

And therein lies the rub. Our ui code looks like this (this is AJAX.NET webservices code, but any ajax framework will do):

WebService.AddFoo("foo", onComplete, onError); // ajax call to web service

function onComplete(newFooId) {
    if(!newFooId) {
        alert('You reached your max number of Foos for the day')
        return
    }
    // update ui as normal here
}

function onError(e) {
    if(e.get_message().indexOf('duplicate key')) {
        alert('A Foo with that name already exists');
        return;
    }
    // REAL error handling code here
}

(As a side note: I notice this is what stackoverflow does when you submit comments too quickly: the server generates a HTTP 500 response and the ui catches it.)

So you see, we are handling business rule violations in two places here, one of which (ie the unique constaint error) is being handled as a special case to the code that is supposed to handle real errors (not business rule violations), since .NET propagates Exceptions all the way up to the onError() handler.

This feels wrong. My options I think are:

1. catch the 'duplicate key violation' exception at the app server level and convert it to whatever it is the ui expects as the "business rule violated" flag,
2. preempt the error (say, with a "select name from Foo where name = @Name") and return whatever it is the app server expects as the "business rule violated" flag,
3. in the same ballpark as 2): leverage the unique constraint built into the db layer and blindly insert into Foo, catching any exceptions and convert it to whatever it is the app server expects as the "business rule violated" flag
4. blindly insert into Foo (like 3) and let that Exception propagate to the ui, plus have the app server raise business rule violations as real Exceptions (as opposed to 1). This way ALL errors are handled in the ui layer's onError() (or similar) code.

What I like about 2) and 3) is that the business rule violations are "thrown" where they are implemented: in the stored proc. What I don't like about 1) and 3) is I think they involve stupid checks like "if error.IndexOf('duplicate key')", just like what is in the ui layer currently.

Edit: I like 4), but most people say to use Exceptions only in exceptional circumstances.

So, how do you people handle propagating business rule violations up to the ui elegantly?

回答1:

A stored procedure may use the RAISERROR statement to return error information to the caller. This can be used in a way that permits the user interface to decide how the error will appear, while permitting the stored procedure to provide the details of the error.

RAISERROR can be called with a msg_id, severity and state, and with a set of error arguments. When used this way, a message with the given msg_id must have been entered into the database using the sp_addmessage system stored procedure. This msg_id can be retrieved as the ErrorNumber property in the SqlException that will be raised in the .NET code calling the stored procedure. The user interface can then decide on what sort of message or other indication to display.

The error arguments are substituted into the resulting error message similarly to how the printf statement works in C. However, if you want to just pass the arguments back to the UI so that the UI can decide how to use them, simply make the error messages have no text, just placeholders for the arguments. One message might be '"%s"|%d' to pass back a string argument (in quotes), and a numeric argument. The .NET code could split these apart and use them in the user interface however you like.

RAISERROR can also be used in a TRY CATCH block in the stored procedure. That would allow you to catch the duplicate key error, and replace it with your own error number that means "duplicate key on insert" to your code, and it can include the actual key value(s). Your UI could use this to display "Order number already exists", where "x" was the key value supplied.

回答2:

We don't perform our business logic in the database but we do have all of our validation server-side, with low-level DB CRUD operations separated from higher level business logic and controller code.

What we try to do internally is pass around a validation object with functions like Validation.addError(message,[fieldname]). The various application layers append their validation results on this object and then we call Validation.toJson() to produce a result that looks like this:

{
    success:false,
    general_message:"You have reached your max number of Foos for the day",
    errors:{
        last_name:"This field is required",
        mrn:"Either SSN or MRN must be entered",
        zipcode:"996852 is not in Bernalillo county. Only Bernalillo residents are eligible"
    }
}

This can easily be processed client side to display messages related to individual fields as well as general messages.

Regarding constraint violations we use #2, i.e. we check for potential violations before insert/update and append the error to the validation object.

回答3:

The problem is really one of a limitation in the architecture of your system. By pushing all logic into the database, you need to handle it in two places (as opposed to building a layer of business logic that links the UI with the database. Then again, the minute you have a layer of business logic you lose all the benefits of having logic in stored procs. Not advocating one or the other. The both suck about equally. Or don't suck. Depending on how you look at it.

Where was I? Right.

I think a combination of 2 and 3 is probably the way to go.

By pre-empting the error you can create a set of procedures that can be called from the UI-facing code to provide detailed implementation-specific feedback to the user. You don't necessarily need to do this with ajax on a field-by-field basis, but you could.

The unique constraints and other rules that are in the database then become the final sanity-check for all data, and can assume that data is good before being sent, and throw Exceptions as a matter of course (the premise being that these procedures should always be called with valid data and therefor invalid data is an Exceptional circumstance).

回答4:

In defense of #4, SQL Server has a pretty orderly hierarchy of error severity levels predefined. Since as you point out it's well to handle errors where the logic is, I'd be inclined to handle this by convention between the SP and the UI abstraction, rather than adding a bunch of extra coupling. Especially since you can raise errors with both a value and a string.

回答5:

I've seen lots of Ajax based applications doing a real-time check on fields such as username (to see if it already exists) as soon as the user leaves the edit box. It seems to me a better approach than leaving to the database to raise an exception based on a db constraint - it is more proactive since you have a real process: get the value, check to see if it is valid, show error if not, allow to continue if no error. So it seems option 2 is a good one.

回答6:

This is how I do things, though it may not be best for you:

I generally go for the pre-emptive model, though it depends a lot on your application architecture.

For me (in my environment) it makes sense to check for most errors in the middle (business objects) tier. This is where all the other business-specific logic takes place, so I try to keep as much of the rest of my logic here too. I think of the database as somewhere to persist my objects.

When it comes to validation, the easiest errors can be trapped in javascript (formatting, field lengths, etc.), though of course you never assume that those error checks took place. Those errors also get checked in the safer, more controlled world of server-side code.

Business rules (such as "you can only have so many foos per day") get checked in the server-side code, in the business object layer.

Only data rules get checked in the database (referential integrity, unique field constraints, etc.). We pre-empt checking all of these in the middle tier too, to avoid hitting the database unnecessarily.

Thus my database only protects itself against the simple, data-centric rules that it's well equipped to handle; the more variable, business-oriented rules live in the land of objects, rather than the land of records.