Open kapilt opened 2 years ago
Thank you, this works fine for now even if we need to subclass.
for you perhaps, but as i mentioned the underlying issue is handling exceptions from third party libraries which makes the current interface more than unwieldy, and effectively requires reimplementing the middleware mechanisms on chalice for developer experience (app exceptions bubble). the native lambda interface handles that well.. chalice not so much atm unless every app reimplements redundant top level handlers to rethrow everything as exception subclasses
the underlying issue is handling exceptions from third party libraries Yep, exactly. We're using pydantic and we would've liked to use the middleware to manage the exceptions
I am also looking for a solution for this issue.
Here is my current objective. I will be submitting a PR shortly if all seems right. The approach proposed is similar to the solution that already exists for other framework (ie Flask) and is trying to stay as close to the existing design of Chalice.
While the ChaliceUnhandledError
works fine for custom errors, third party libraries can raise Exceptions that can't be subclassed. Having a try/except
in a middleware won't successfully catch the Exception as it is already a Response
at that point.
InternalServerError
response is built.Response
Response
or an exception raised in the handler should still return the standard InternalServerError
and have the stack trace replaced by the expected json object when not in debug modeAs existing middlewares can already successfully intercept all exceptions for other events, and ChaliceUnhandledError
only exists in the scope of http
event, so will the focus of this handler.
A new @app.errorhandler
is to be added. The decorated function will take one arg, the exception.
Similar to the way middlewares can be registered, we will be able to register the error handler in app or in a blueprint with either a decorator or app.register_error
.
# Decorator
@app.errorhandler(MyCustomError)
def my_handler(e: MyCustomError):
return Response(body="", status_code:400)
# Register
app.register_error_handler(MyCustomError, my_handler)
# Blueprint
@my_blueprint.errorhandler(MyCustomError)
def my_handler(e: MyCustomError):
return Response(body="", status_code:400)
In the examples above, any http endpoint raising MyCustomError
will then return a 400
with no body.
If, however, the return value isn't a Response
or an exception was raised during execution of the handler, the normal flow of the RestAPIEventHandler
will be maintained.
@app.errorhandler(MyCustomError)
def my_handler(e: MyCustomError):
return "Not a Response object"
Will return
{
"Code": "InternalServerError",
"Message": "An internal server error occurred."
}
Hey there any movement on this?
Kind of a hacky way around around this, but you have to add a decorator to all routes.
def error_wrapper(fn):
def wrapper():
try:
return fn()
except Exception as ex:
print('Error wrapper')
print(str(ex))
return Response(
{'result': 'error', 'error': 'server_error'},
status_code=500,
)
return wrapper
@app.route('/api/something', methods=['GET'])
@error_wrapper
def do_something():
raise ValueError('a fake error.')
Or instead of decorating each route, if you like to live dangerously:
def wrap_routes(app):
for path, views in app.routes.items():
for view in list(views.values()):
print(f'Wrapping route {view.method} {path}, fn: {view.view_function.__name__}')
view.view_function = error_wrapper(view.view_function)
wrap_routes(app)
While the ChaliceUnhandledError works fine for custom errors, third party libraries can raise Exceptions that can't be subclassed. Having a try/except in a middleware won't successfully catch the Exception as it is already a Response at that point.
I am having this precise problem. Was there any solution given to this?
as part of #1541 and #1549 middleware support for catching exceptions was added, but the interface to it is unwieldy as it requires all exceptions to subclass a specific base exception .. given use of any library code or larger applications this is rather difficult to accomodate in practice. Given we have a default global error handler that does the exact same thing as the view response, i think uncaught exceptions (minus chalice view error) should propagate by default so middleware has a chance to handle it, and the global error acts as a fallback.