How to capture exceptions?

[MrTin]

We're trying to forward exceptions to Sentry but it seems the graphene-django seems to be swallowing everything. Is there any way to catch the exception and send it to Sentry?

Thank you!

Regards,
Martin

[syrusakbary]

Hi @MrTin,

There are few ways for catching the exceptions that GraphQL is swallowing (intentionally), the way I recommend is via Middleware:

class SentryMiddleware(object):
    def on_error(self, error):
        # report error to sentry
    def resolve(self, next, root, args, context, info):
       return next(root, args, context, info).catch(self.on_error)

Here are the docs for Middleware in graphene: http://docs.graphene-python.org/en/latest/execution/middleware/

Hope this helps!

[MrTin]

Hi @syrusakbary

Sorry for my late reply – thank you very much for taking the time to reply.

If I understand this solution correct it requires me to add this middleware manually to each query? Or is there a way to set middlewares for all my queries in one place?

[sdcooke]

@syrusakbary I gave this a go but catch was never actually called (I was just raising an exception in a resolve_ method) - it looked like the exceptions were just collected on the execution context. An alternative solution I found was to check execution_result.errors before returning the response which works for getting a list of exceptions (error.original_error), but doesn't work particularly well for Sentry because Sentry wants to use sys.exc_info(). So I have two questions:

1. Do you think it would be possible to get sys.exc_info() into the errors returned on the execution result?
2. Under what scenario do you think this middleware with catch would work?

Thanks!

[danielholmes]

I'm seeing the same issue as @sdcooke . Looking at this line: https://github.com/graphql-python/graphql-core/blob/master/graphql/execution/middleware.py#L57

def make_it_promise(next, *a, **b):
  return Promise.resolve(next(*a, **b))

It runs the execution, then places the result in a promise. So if next(*a, **b) returns a successful result it's wrapped in a promise but if there's an exception it bubbles instead.

Changing it to the following allows the middleware to catch errors:

def make_it_promise(next, *a, **b):
  return Promise.promisify(lambda: next(*a, **b))()

However that breaks down execution in other places.

i.e. Some parts of the code and @syrusakbary above seem to assume make_it_promise converts exceptions to rejected promises as in my version, but most of the code seems to assume the current behaviour of raising exceptions but wrapping success values.

[gamugamu]

I'm really struggling in catching exception on graphQL. The middelWare seems unhelpful as he is not triggered when there is an exception and the propagation stop before the middleware. (So it swallows all the client / server communication EVEN with a middleware)
That look like an anti-design pattern to not be able to get catch simply error from graphene. I know this is more related to graphene instead of graphene-django. But is there true helpful debugging example for graphene? This one is not working for my case (debugging 400 code error request)

[MrTin]

Unfortunately, I never got around trying this out and ended up with my original solution using a custom GraphQLView.

You can grab the full solution and read why in my article on medium.

Here's also the code if you just want the solution:

# views.py
from graphene_django.views import GraphQLView
from raven.contrib.django.raven_compat.models import client as sentry_client


class SentryGraphQLView(GraphQLView):
    def execute_graphql_request(self, *args, **kwargs):
        """Extract any exceptions and send them to Sentry"""
        result = super().execute_graphql_request(*args, **kwargs)
        if result.errors:
            for error in result.errors:
                try:
                    raise error.original_error
                except Exception:
                    sentry_client.captureException()
        return result


# urls.py
from .views import SentryGraphQLView


urlpatterns = [
    url(r'^graphql', SentryGraphQLView.as_view(schema=schema)),
]

[picturedots]

Has this been resolved in graphene? I am seeing graphene exceptions in Sentry and I don't recall implementing any workarounds (with graphene==2.0.1, graphene-django==2.0.0, raven==6.6.0). Or is this a side-effect of using graphene_django.debug.DjangoDebugMiddleware setting for GRAPHENE["MIDDLEWARE"] ?

[wilk]

The @syrusakbary 's solution works well, actually, with the latest version of graphql-core: https://github.com/graphql-python/graphql-core/blob/master/graphql/execution/middleware.py#L74
As you can see, the middleware execution is wrapped inside a well-constructed promise.

As a work-around, for previous versions of graphql-core, try this:

import sys
import logging

class SentryMiddleware(object):
  def resolve(self, next, root, info, **args):
    try:
      return next(root, info, **args)
    except:
      err = sys.exc_info()
      logging.error(err)
      return err[1]

[zbyte64]

Is it a good idea to add logging statements to views.py so that this can be done through configuring django LOGGING?

[shabbyrobe]

The middleware solution does not catch exceptions in dataloaders; next returns a promise.

[vojto]

Here's @MrTin's solution updated to Sentry unified Python SDK:

from sentry_sdk import capture_exception

class SentryGraphQLView(GraphQLView):
    def execute_graphql_request(self, *args, **kwargs):
        """Extract any exceptions and send them to Sentry"""
        result = super().execute_graphql_request(*args, **kwargs)
        if result.errors:
            for error in result.errors:
                try:
                    raise error.original_error
                except Exception as e:
                    capture_exception(e)
        return result

# In routes:

path('graphql', csrf_exempt(SentryGraphQLView.as_view(graphiql=True))),

[picturedots]

After I updated to sentry_sdk, I started receiving the correct data exceptions in Sentry, without any addition changes to views or middleware. The errors were double-reported as GraphQLLocatedError errors, so I added a filter to remove them:

from graphql.error import GraphQLLocatedError

def before_send(event, hint):
    """
    Filters GraphQLLocatedError events to prevent double reporting at Sentry.
    """
    if 'exc_info' in hint:
        _exc_type, exc_value, _tb = hint['exc_info']
        if isinstance(exc_value, GraphQLLocatedError):
            return None
    return event

sentry_sdk.init(before_send=before, ...)

[mvanlonden]

Closing since this issue is not specific to graphene-django

[sandwichsudo]

In case this helps someone - the class based middleware didn't work for me, but this function based middleware did:

def log_error(error):
    logger.error(error) # we've set this up to send issues to sentry
    raise graphql.GraphQLError("An internal error occurred")


def uncaught_exception_middleware(next, root, info, **args):
    return next(root, info, **args).catch(log_error)

then in our settings:

GRAPHENE = {
    "MIDDLEWARE": [
        "octoenergy.interfaces.apisite.graphql.middleware.uncaught_exception_middleware",
    ],
}