A walk in GraphQL — Errors — Day 7

8 articles to learn GraphQL incrementally, keeping the implementation agnostic spirit of the SDL

Original Photo by Matt Whitacre on Unsplash

Errors

If you’re new to GraphQL you might find this section very surprising — but believe us — you’ll need to re-think “what an error is” to comfortably navigate GraphQL’s waters.

TL;DR

1. One Graph to rule them all — The Spec.

As GraphQL is — technically — something “in between”, a flaw may come from:

  • the client,
  • within the GraphQL server
  • the services/persistence providers;

GraphQL will treat them all in a consistent way — as defined by the spec. — but before digging deeper into the spec. let’s organize them.

By Severity (roughly)

  • Fatal
    — The service is unresponsive
  • Non Fatal
    — A fault occurred but an “operable” response can still be provided.

By Origin

  • Client/Request
    — Whoever is consuming it
  • Provider
    — Whatever layer is providing a service to GraphQL — e.g. persistence layer
  • Server
    — Runtime

By GraphQL Phase

And in the spec. the rules are quiet consistent:

It’ll try — at any cost — to propagate the error and continue operating whenever possible.

If you want to see all the cases, one of the most educational ways is to browse the spec. and search for “errors”, you’ll obtain about 170 different places related to the type definitions, the validation, the execution, the non-nullability cases and particularly how the Response should be provided, formatted and even serialized!

2. One Place to find them — 200 OK & errors

Unless something goes really bad with the server, you probably won’t see other than a 200 OK HTTP status code.

Do you remember, “It’s Graphs All the Way Down” sentence?
If you have a reference error inside a resolver’s code, your server will run perfectly until that resolver is executed, then the reference error will be thrown, but the server won’t fail, it won’t stop executing, it’ll simply propagate the error up and up until the response is being prepared, and there, an errors node will be added to the JSON response.

Shocking? (◉ω◉)

The errors node is a list of error objects containing the following structure:

Now let’s break down the errors entry and the properties of each error in the list.
In the GraphQL spec June 2018 - 7.1.2 Errors section everything is described in great detail, check below some highlights.

errors entry

  • is a non-empty list of errors
  • shouldn’t be present in the response if no errors were found
  • if data entry is absent, then errors MUST be present
  • data and errors entries can coexist in the response

Each error in the list is a map containing the following properties:

  • message
    mandatory
    — a descriptive string of the error
    — directed to developers
  • locations
    mandatory
    — a list of locations
    — each location is a { "line": Number, "column": Number } where Number is a 1‐indexed positive integer
  • path
    mandatory
    — path segments from root to the identified field which experienced the error
    — if a segment represents a field, it should be a string containing the field name
    — if a segment represents an index of a list, it should be a 0‐indexed integer
  • extensions
    optional
    — implementation dependent
    — unrestricted arbitrary content
    — since adding other entries to the errors map are highly discouraged — but not prohibited — this might be the RIGHT PLACE for engineers to add extra info whenever required (see example below)

GraphQL services may provide an additional entry to errors with key extensions. This entry, if set, must have a map as its value. This entry is reserved for implementors to add additional information to errors however they see fit, and there are no additional restrictions on its contents.

GraphQL services should not provide any additional entries to the error format since they could conflict with additional entries that may be added in future versions of this specification.

Source: GraphQL spec June 2018 — Example nº187

3. One Object to bring them all — data & errors are siblings

You may have both a data and an errors properties in the Response body, containing expected data as well as 1 or more errors.

Going back to our characters list example, let’s say the name property is nullable for Istary but non-nullable for Hobbit and one of the records has a null value; this is what we get as a response:

TypeDef

Query

Response

4. and in your mind bind them — “errors” as “unrequested results”

Let’s stay put for a moment and analyse what just happened above.

So:

… and many other questions can derive from this behavior!

So, what do we do about this?
There’s no “the right answer” for that.

In terms of community trends related to best practices there are different opinions:

— Intentionally throw errors!
— Do Not! ୧༼ಠ益ಠ༽୨

— Format them!
— Whatever! (◔_◔)

— Make Errors part of your schema!
— Do Not! ୧༼ಠ益ಠ༽୨

— Leverage your GraphQL server app error features!
— Do Not! ୧༼ಠ益ಠ༽୨

— Be careful with extension field!
— Hell yeah! ᕦ( ͡° ͜ʖ ͡°)ᕤ

— Disable stacktrace for production! (should be the default behaviour of your server app in production mode, but still...)
— Hell yeah! ᕦ( ͡° ͜ʖ ͡°)ᕤ

Unless the only thing you have is a fatal error, you’ll need to start thinking of “errors” as results — which is true in the GraphQL mindset — even though they might not be the results you’re expecting for.

You can think of them as requested results — the data node— and unrequested results —the errors node— and make sure all the organization is aligned on how to organize and treat them.

5. In the land of the runtime where the implementation lies — The runtime handles the rest

Everything that exceeds what GraphQL defines in the spec. will be handled by the runtime. It delegates other implementation details (custom handling) to the server applications, the language and the execution environment; and it’ll determine the level of interaction you will have with the error handling system.

Exercise

Until now, we didn’t care about several aspects outside of GraphQL’s plate — like data consistency. An example?

If you run the following mutation:

with the following variables:

and there’s no record with ID 8000, you may have different outcomes.
Here some of them:

A. Before/during insertion

B.During query — after insertion —

If we take A.3 and B.2 together with the current schema

you’ll obtain this response:

The invalid parent id was persisted — or not, GraphQL doesn’t care — and the parent’s record couldn’t be retrieved during the subsequent query, but neither an error was thrown nor an invalid type was provided — since parent is nullable— and therefore the data node came along with no errors sibling.

  • Is this an error?
  • who should take the responsibility for the inconsistency?
  • when, where, how should we handle this situation?

Clearly the first one is an easy one to answer, OF COURSE IT IS! but, what about the rest? Well, that will depend on many factors but they’re all related to engineering practices and not to GraphQL strictly speaking.

Exercise requirements

All the mutations provided on the previous days intentionally LACK or DIVERGE on this kind of verification depending on their specific stack.

Since this practice hasn’t “A RIGHT SOLUTION”, we propose you to review the current code and — given the specific technology you’re working with, and your criteria — explore, propose and implement 3 solutions for the createSkill mutation —you can work on other mutations too if you want, that'll depend on your time—

Then take notes regarding the pros and cons of each approach and share it with your team mates.

IMPORTANT: On the learning resources section there are several incredibly useful videos and links; take a look at them before putting your hands on the exercise.

Adding new types for this exercise is allowed but changing the current types is not.

Technologies

Select the exercise on your preferred technology:

Learning resources

GraphQL Spec (June 2018)

Apollo Blog

Apollo GraphQL

YouTube

HowToGraphQL

GraphQL .NET

Other articles

A walk in GraphQL Series at Medium

Github Pages

A Walk in GraphQL

Collaborators

  • Ezequiel Alvarez — @ealvarezk
    Reviewer
  • Ezequiel Tejerina — @quequitejerina
    Python exercises contributor and reviewer
  • Franco Gribaudo — @fgriba
    Java exercises contributor and reviewer
  • Cristian Buffa — @cristianbuffa & José Ignacio Aguilera — @jiaguilera
    NetCore exercises contributors and reviewers
  • Javier Valderrama — @Jaxolotl
    Author, JavaScript exercise contributor and reviewer