Sentry doesn't capture aiohttp exception?

[lopatinay]

I followed an example https://docs.sentry.io/platforms/python/aiohttp/ and first thing I did

async def hello(request):
    print(1/0)

It's work fine and I got this Sentry Issue. But, when you try

async def hello(request):
    raise web_exceptions.HTTPInternalServerError  # any aiohttp exception

you won't get any Sentry Issue.
Is the only way to catch errors to use middleware?

async def errors_handler(request, handler):
    try:
        response = await handler(request)
    except web.HTTPException as e:
        capture_exception(e)

[xen]

I'm not sure if it helps at all. But maybe it gives some ideas for people who will work on this issue. It is possible to get an error if reraise exception again:

async def errors_handler(request, handler):
    try:
        response = await handler(request)
    except web.HTTPException as e:
        capture_exception(e)
        raise

But this is making impossible to capture error pages and publish custom error page.

[lopatinay]

But, if you use this tricks you will get in Sentry "[Can't show request body due to implementation details.]". I found this solution

try:
    response = await handler(request)
except web.HTTPException as e:
    with configure_scope() as scope:
        body = await request.json()
        scope.set_extra("body", body)
    capture_exception(e)
    raise e

[untitaker]

We explicitly ignore all the HTTPExceptions because otherwise you would get an event for random status codes you probably don't care about. Could one of you post an example of an actionable issue with HTTPException? We could add a config option for that

[lopatinay]

As example custom app error inherited from HTTPException. One of example:

from aiohttp.web_exceptions import HTTPServiceUnavailable

class WeatherProviderUnavailableError(HTTPServiceUnavailable):
    pass

[xen]

I assume that my view can end in error state. But I don't want to show user blank error page. So, I made middleware that do exactly this:

async def handle_error(request: Request, code: int = 404) -> Response:
    response = aiohttp_jinja2.render_template(f"misc/{code}.html", request, {})
    response.set_status(code)
    return response


@web.middleware
async def error_middleware(request: Request, handler: Callable) -> Union[FileResponse, Response]:
    error_codes = (403, 404, 405, 500)
    try:
        response = await handler(request)
        if response.status not in error_codes:
            # return response, everythink is ok
            return response
        # return error page because handler explicitly set different status code
        logging.info("Error: %s", response.status)
        return await handle_error(request, response.status)
    except web.HTTPException as ex:
        # this block for errors generated by aiohttp itself. 
        # I want to know that something invoke error
        # probably I will filter some error codes later
        logging.info("Error", exc_info=sys.exc_info())
        capture_exception(ex)
        if ex.status in errors:
            return await handle_error(request, code=ex.status)
        # we don't handle this error
        raise
    except Exception as ex:
        # This is place for all other Exceptions, for example ZeroDivisioinError 
        # This is most important part to capture all traceback
        logging.info("Error", exc_info=sys.exc_info())
        capture_exception(ex)
        return await handle_error(request, code=500)

What is preferred way to achieve correct error handling by Sentry?

[untitaker]

Closing this issue is a question. Generally there needs to be a strong case made as to why this behavior should change. Any of the above workarounds is fine as long as it works for you.