I'm running an apollo-server-express as a gateway application. Setting up a few underlying GraphQL Applications with makeRemoteExecutableSchema and an apollo-link-http.
Usually every call just works. If an error is part of the response and data is null it also works. But if data contains just the data and errors contains an error. Data will be passed though but errors is empty
const headerSet = setContext((request, previousContext) => {
return setHeaders(previousContext);
});
const errorLink = onError(({ response, forward, operation, graphQLErrors, networkError }) => {
if (graphQLErrors) {
graphQLErrors.map((err) => {
Object.setPrototypeOf(err, Error.prototype);
});
}
if (networkError) {
logger.error(networkError, 'A wild network error appeared');
}
});
const httpLink = new HttpLink({
uri: remoteURL,
fetch
});
const link = headerSet.concat(errorLink).concat(httpLink);
Example A "Working Example":
Query
{
checkName(name: "namethatistoolooooooong")
}
Query Response
{
"errors": [
{
"message": "name is too long, the max length is 20 characters",
"path": [
"checkName"
],
"extensions": {
"code": "INPUT_VALIDATION_ERROR"
}
}
],
"data": null
}
Example B "Errors hidden":
Query
mutation inviteByEmail {
invite(email: "invalid!!!~~~test!--#example.com") {
status
}
}
Response from remote service (httpLink)
response.errors and graphQLErrors in onError method also contains the error
{
"errors": [
{
"message": "Email not valid",
"path": [
"invite"
],
"extensions": {
"code": "INPUT_VALIDATION_ERROR"
}
}
],
"data": {
"invite": {
"status": null
}
}
}
Response
{
"data": {
"invite": {
"status": null
}
}
}
According to graphql spec I would have expected the errors object to not be hidden if it is part of the response
https://graphql.github.io/graphql-spec/June2018/#sec-Errors
If the data entry in the response is present (including if it is the value null), the errors entry in the response may contain any errors that occurred during execution. If errors occurred during execution, it should contain those errors.
Related
due to well-known N+1 problem we decided to move away from #ResolveField() feature of NestJS and use our implementation of DataLoader instead. By doing so, we must handle errors of resolvers manually because the resolution of graphql data is not driven by NestJS (or apollo) anymore.
This put us into a problem when we want to return a GraphQL error response (e.g. "Book not found") from graphql query in a standard manner like this:
{
"data": {
user: {
id: 1
book: null
}
}
"errors": [
{
"message": "Book not found",
"statusCode": 400
}
]
}
But since we are not using #ResolveField() anymore we resolve nested data (book) manually
we receive this response:
{
"data": null
"errors": [
{
"message": "Book not found",
"statusCode": 400
}
]
}
Is there any way to populate GraphQL error response manually?
#Query(() => User)
async user(#Args('id') id: number): Promise<User> {
const user = await this.userService.findOne(id);
try{
const book = await this.bookService.findOne(user.bookId);
user.book = book;
} catch (e) {
// How to populate GraphQL error response manually?
user.book = null;
}
return user;
}
Thanks for your help and have a nice day!
In an express-graphql app, I have a userLogin resolver like so:
const userLogin = async ({ id, password }), context, info) => {
if (!id) {
throw new Error('No id provided.')
}
if (!password) {
throw new Error('No password provided.')
}
// actual resolver logic here
// …
}
If the user doesn't provide an id AND a password, it will throw only one error.
{
"errors": [
{
"message": "No id provided.",
"locations": [
{
"line": 2,
"column": 3
}
],
"path": [
"userLogin"
]
}
],
"data": {
"userLogin": null
}
}
How is it possible to throw multiple errors in the errors response array?
There is no way to throw an array of errors in JavaScript or otherwise have a single resolver reject with more than one error. A GraphQL response includes an errors array and not just a single error object because the total response can include multiple errors when those errors originate from different fields. Consider this schema and resolvers:
type Query {
a: String
b: String
c: String
}
const resolvers = {
Query: {
a: () => { throw new Error('A rejected') },
b: () => { throw new Error('B rejected') },
c: () => 'Still works!',
},
}
If you query all three fields...
query {
a
b
c
}
Your data will look something like this:
{
"errors": [
{
"message": "A rejected",
"locations": [
{
"line": 2,
"column": 3
}
],
"path": [
"a"
]
},
{
"message": "B rejected",
"locations": [
{
"line": 3,
"column": 3
}
],
"path": [
"b"
]
}
],
"data": {
"a": null,
"b": null,
"c": "Still works!"
}
}
This is because GraphQL supports partial responses. However, keep in mind that this works because the fields are nullable. If they were non-null, those errors would bubble up to the closest nullable parent field.
Here are some alternative approaches:
You can utilize formatError to change how the errors returned by GraphQL are displayed to the client. That means you can include any sort of extra information with your errors, like an error code or multiple error messages. A simple example:
// The middleware
app.use('/graphql', graphqlExpress({
schema: schema,
formatError: (error) => ({
message: error.message,
path: error.path,
locations: error.locations,
errors: error.originalError.details
})
}))
// The error class
class CustomError extends Error {
constructor(detailsArray) {
this.message = String(details)
this.details = details
}
}
// The resolver
const userLogin = async ({ id, password }), context, info) => {
const errorDetails = []
if (!id) errorDetails.push('No id provided.')
if (!password) errorDetails.push('No password provided.')
if (errorDetails.length) throw new CustomError(errorDetails)
// actual resolver logic here
}
Your response then looks more like this:
{
"errors": [
{
"message": "[No id provided.,No password provided.]",
"locations": [
{
"line": 2,
"column": 3
}
],
"path": [
"userLogin"
]
"errors" [
"No id provided.",
"No password provided."
]
}
],
"data": {
"userLogin": null
}
}
That said, there's something a bit unsavory about returning user-facing error messages alongside GraphQL validation errors. Another approach that some APIs have taken is to include an errors field alongside the actual mutation response. For example:
type Mutation {
userLogin: UserLoginResponse
}
type UserLoginResponse {
response: User
errors: [String!]
}
You can also use unions to achieve a similar effect:
type Mutation {
userLogin: UserLoginResponse
}
type Errors {
errors: [String!]!
}
union UserLoginResponse = User | Errors
I am using the Strapi v3.0.0-beta.18.7 UI and the error in the UI are shown partly so it is impossible to read the full text of the error message.
I suggest use a custom middleware to manage your needs.
Here is the documentation to create a middleware - https://strapi.io/documentation/3.0.0-beta.x/concepts/middlewares.html
Step 1: Create the middleware
Path — middlewares/log/index.js
module.exports = strapi => {
return {
initialize() {
strapi.app.use(async (ctx, next) => {
await next();
const status = ctx.status;
if (status < 200 || status >= 300) {
console.log(ctx.body);
}
});
},
};
};
Step 2: Enable the middleware
Path — config/environments/development/middleware.json
{
"log": {
"enabled": true
}
}
Step 3: Set the middleware in the right order
Path — config/middleware.json
{
"timeout": 100,
"load": {
"before": [
"log",
"responseTime",
"logger",
"cors",
"responses",
"gzip"
],
"order": [
"Define the middlewares' load order by putting their name in this array is the right order"
],
"after": [
"parser",
"router"
]
}
}
Receiving data with AppSync directly from DynamoDB seems working for my case, but when I try to put a lambda function in between, I receive errors that says "Can't resolve value (/issueNewMasterCard/masterCards) : type mismatch error, expected type LIST"
Looking to the AppSync cloudwatch response mapping output, I get this:
"context": {
"arguments": {
"userId": "18e946df-d3de-49a8-98b3-8b6d74dfd652"
},
"result": {
"Item": {
"masterCards": {
"L": [
{
"M": {
"cardId": {
"S": "95d67f80-b486-11e8-ba85-c3623f6847af"
},
"cardImage": {
"S": "https://s3.eu-central-1.amazonaws.com/logo.png"
},
"cardWallet": {
"S": "0xFDB17d12057b6Fe8c8c434653456435634565"
},...............
here is how I configured my response mapping template:
$utils.toJson($context.result.Item)
I'm doing this mutation:
mutation IssueNewMasterCard {
issueNewMasterCard(userId:"18e946df-d3de-49a8-98b3-8b6d74dfd652"){
masterCards {
cardId
}
}
}
and this is my schema :
type User {
userId: ID!
masterCards: [MasterCard]
}
type MasterCard {
cardId: String
}
type Mutation {
issueNewMasterCard(userId: ID!): User
}
The Lambda function:
exports.handler = (event, context, callback) => {
const userId = event.arguments.userId;
const userParam = {
Key: {
"userId":{S:userId}
},
TableName:"FidelityCardsUsers"
}
dynamoDB.getItem(userParam, function(err, data) {
if (err) {
console.log('error from DynamDB: ',err)
callback(err);
} else {
console.log('mastercards: ',JSON.stringify(data));
callback(null,data)
}
})
I think the problem is that the getItem you use when you use the DynamoDB datasource is not the same as the the DynamoDB.getItem function in the aws-sdk.
Specifically it seems like the datasource version returns an already marshalled response (that is, instead of something: { L: [ list of things ] } it just returns something: [ list of things]).
This is important, because it means that $utils.toJson($context.result.Item) in your current setup is returning { masterCards: { L: [ ... which is why you are seeing the type error- masterCards in this case is an object with a key L, rather than an array/list.
To solve this in the resolver, you can use the $util.dynamodb.toDynamoDBJson(Object) macro (https://docs.aws.amazon.com/appsync/latest/devguide/resolver-util-reference.html#dynamodb-helpers-in-util-dynamodb). i.e. your resolver should be:
$util.dynamodb.toDynamoDBJson($context.result.Item)
Alternatively you might want to look at the AWS.DynamoDB.DocumentClient class (https://docs.aws.amazon.com/AWSJavaScriptSDK/latest/AWS/DynamoDB/DocumentClient.html). This includes versions of getItem, etc. that automatically marshal and unmarshall the proprietary DynamoDB typing back into native JSON. (Frankly I find this much nicer to work with and use it all the time).
In that case you can keep your old resolver, because you'll be returning an object where masterCards is just a JSON array.
Let's say I'm trying to create a bike as a mutation
var createBike = (wheelSize) => {
if (!factoryHasEnoughMetal(wheelSize)) {
return supplierError('Not enough metal');
}
return factoryBuild(wheelSize);
}
What happens when there's not enough steel for them shiny wheels? We'll probably need an error for the client side. How do I get that to them from my graphQL server with the below mutation:
// Mutations
mutation: new graphql.GraphQLObjectType({
name: 'BikeMutation',
fields: () => ({
createBike: {
type: bikeType,
args: {
wheelSize: {
description: 'Wheel size',
type: new graphql.GraphQLNonNull(graphql.Int)
},
},
resolve: (_, args) => createBike(args.wheelSize)
}
})
})
Is it as simple as returning some error type which the server/I have defined?
Not exactly sure if this is what you after...
Just throw a new error, it should return something like
{
"data": {
"createBike": null
},
"errors": [
{
"message": "Not enough metal",
"originalError": {}
}
]
}
your client side should just handle the response
if (res.errors) {res.errors[0].message}
What I do is passing a object with errorCode and message, at this stage, the best way to do is stringify it.
throw new Errors(JSON.stringify({
code:409,
message:"Duplicate request......"
}))
NOTE: also you might be interested at this library https://github.com/kadirahq/graphql-errors
you can mask all errors (turn message to "Internal Error"), or define userError('message to the client'), which these will not get replaced.