Skip to content

Update serving over HTTP docs. #1813

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 3 commits into from
Nov 15, 2024
Merged
Changes from 1 commit
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
123 changes: 96 additions & 27 deletions src/pages/learn/serving-over-http.mdx
Original file line number Diff line number Diff line change
@@ -1,22 +1,65 @@
import { Callout } from "nextra/components"

# Serving over HTTP

HTTP is the most common choice for client-server protocol when using GraphQL because of its ubiquity. Here are some guidelines for setting up a GraphQL server to operate over HTTP.
<p className="learn-subtitle">Respond to GraphQL requests using an HTTP server</p>

The GraphQL specification doesn't require paricular client-server protocols when sending API requests and responses, but HTTP is the most common choice because of its ubiquity. On this page, we'll review some key guidelines to follow when setting up a GraphQL server to operate over HTTP.

Note that the guidelines that follow only apply to stateless query and mutation operations. Visit the [Subscriptions page](/learn/subscriptions) for more information on transport protocols that commonly support these requests.

<Callout type="info">
The recommendations on this page align with the detailed [HTTP transport specification](https://github.com/graphql/graphql-over-http) currently in development. Though not yet finalized, this draft specification acts as a single source of truth for GraphQL client and library maintainers, detailing how to expose and consume a GraphQL API using an HTTP transport. Unlike the language specification, adherence is not mandatory, but most implementations are moving towards these standards to maximize interoperability.
</Callout>

## API endpoint

HTTP is commonly associated with REST, which uses "resources" as its core concept. In contrast, GraphQL's conceptual model is an [entity graph](/learn/thinking-in-graphs/). As a result, entities in GraphQL are not identified by URLs. Instead, a GraphQL server operates on a single URL/endpoint, usually `/graphql`, and all GraphQL requests for a given service should be directed at this endpoint.

## Where auth happens

Most modern web frameworks use a pipeline model where requests are passed through a stack of _middleware_, also known as _filters_ or _plugins_. As the request flows through the pipeline, it can be inspected, transformed, modified, or terminated with a response. GraphQL should be placed after all authentication middleware so that you have access to the same session and user information you would in your HTTP endpoint handlers.

After authentication, the server should not make any [authorization](/learn/authorization/) decisions about the request until GraphQL execution begins. Specifically, field-level authorization should be done by the GraphQL schema during [GraphQL's ExecuteRequest()](https://spec.graphql.org/draft/#ExecuteRequest()), allowing for a partial response to be generated if authorization-related errors are raised.

## Request format

## Web Request Pipeline
### Headers

Most modern web frameworks use a pipeline model where requests are passed through a stack of middleware (AKA filters/plugins). As the request flows through the pipeline, it can be inspected, transformed, modified, or terminated with a response. GraphQL should be placed after all authentication middleware, so that you have access to the same session and user information you would in your HTTP endpoint handlers.
GraphQL clients and servers should support JSON for serialization and may support other formats. Clients should also indicate what media types they support in responses using the `Accept` HTTP header. Specifically, the client should include the `application/graphql-response+json` in the `Accept` header.

## URIs, Routes
If no encoding information is included with this header, then it will be assumed to be `utf-8`. However, the server may respond with an error if a client doesn't provide the `Accept` header with a request.

<Callout type="info">
The `application/graphql-response+json` is described in the draft GraphQL over HTTP specification. To ensure compatibility, if a client sends a request to a legacy GraphQL server before 1st January 2025, the `Accept` header should also include the `application/json` media type as follows: `application/graphql-response+json; charset=utf-8, application/json; charset=utf-8`
</Callout>

### Methods

Your GraphQL HTTP server must handle the HTTP `POST` method, and may also accept the `GET` method for incoming API requests.

#### `POST` request and body

A standard GraphQL POST request should set `application/json` for its `Content-type` header, and include a JSON-encoded body of the following form:

```json
{
"query": "...",
"operationName": "...",
"variables": { "myVariable": "someValue", ... },
"extensions": { "myExtension": "someValue", ... }
}
```

HTTP is commonly associated with REST, which uses "resources" as its core concept. In contrast, GraphQL's conceptual model is an entity graph. As a result, entities in GraphQL are not identified by URLs. Instead, a GraphQL server operates on a single URL/endpoint, usually `/graphql`, and all GraphQL requests for a given service should be directed at this endpoint.
The `query` parameter is required and will contain the source text of the GraphQL document. Note that this parameter may contain documents for query or mutation operations.

## HTTP Methods, Headers, and Body
The `operationName`, `variables`, and `extensions` parameters are optional fields. `operationName` is only required if multiple operations are present in the query.

Your GraphQL HTTP server should handle the HTTP GET and POST methods.
Note that if the `Content-type` header is missing in the client's request, then the server should respond with a `4xx` status code. As with the `Accept` header, `utf-8` encoding is assumed for a request body with an `application/json` media type when this information is not explicitly provided.

### `GET` request
#### `GET` request and parameters

When receiving an HTTP GET request, the GraphQL query should be specified in the "query" query string. For example, if we wanted to execute the following GraphQL query:
When receiving an HTTP GET request, the GraphQL document should be provided in the `query` parameter of the query string. For example, if we wanted to execute the following GraphQL query:

```graphql
{
Expand All @@ -34,37 +77,63 @@ http://myapi/graphql?query={me{name}}

Query variables can be sent as a JSON-encoded string in an additional query parameter called `variables`. If the query contains several named operations, an `operationName` query parameter can be used to control which one should be executed.

### `POST` request
#### Choosing an HTTP method

A standard GraphQL POST request should use the `application/json` content type, and include a JSON-encoded body of the following form:
When choosing an HTTP method for a GraphQL request, there are a few points to consider. First, support for HTTP methods other than `POST` will be at the discretion of the GraphQL server, so a client will be limited to the supported verbs.

```json
{
"query": "...",
"operationName": "...",
"variables": { "myVariable": "someValue", ... }
}
```
Additionally, the `GET` HTTP method may only be used for query operations, so if a client sends a mutation operation in a document then it must use the `POST` method instead.

When servers that do support the `GET` method for query operations, a client may be encouraged to leverage this option to facilitate HTTP caching or edge caching in a content delivery network (CDN). However, because GraphQL documents strings may be quite long for complex operations, the query parameters may exceed the limits that browsers and CDNs impose on URL lengths.

In these cases, it would be necessary to use something like _trusted documents_ (sometimes referred to as _persisted queries_) or _automatic persisted queries_, where the server stores hashes of known GraphQL documents and the clients can send a hash as the `query` parameter value in a `GET` request instead of the full source text of the query operation.

`operationName` and `variables` are optional fields. `operationName` is only required if multiple operations are present in the query.
## Response format

## Response
### Headers

Regardless of the method by which the query and variables were sent, the response should be returned in the body of the request in JSON format. As mentioned in the spec, a query might result in some data and some errors, and those should be returned in a JSON object of the form:
The response type should be `application/graphql-response+json` but can be `application/json` to support legacy clients. The server will indicate the media type in the `Content-type` header of the response. It should also indicate the encoding, otherwise `utf-8` will be assumed.

### Body

Regardless of the HTTP method used to send the query and variables, the response should be returned in the body of the request in JSON format. As mentioned in the GraphQL specification, a query might result in some data and some errors, and those should be returned in a JSON object of the form:

```json
{
"data": { ... },
"errors": [ ... ]
"errors": [ ... ],
"extensions": { ... }
}
```

If there were no errors returned, the `"errors"` field should not be present on the response. If no data is returned, [according to the GraphQL spec](https://spec.graphql.org/October2021/#sec-Data), the `"data"` field should only be included if no errors occurred during execution.
If no errors were returned, the `errors` field should not be present in the response. If no data is returned, [according to the GraphQL spec](https://spec.graphql.org/October2021/#sec-Data), the `data` field should only be included if no errors occurred before execution. The `extensions` field is optional and information provided here will be at the discretion of the GraphQL implementation.

You can read more about GraphQL spec-compliant response formats on the [Response page](/learn/response/).

### Status Codes

If a response contains a `data` key and its value is not `null`, then the server should respond with a `2xx` status code for either the `application/graphql-response+json` or `application/json` media types.

For validation errors that prevent the execution of a GraphQL operation, the server will typically send a `400` status code, although some legacy servers may return a `2xx` status code when the `application/json` media type is used.

For legacy servers that respond with the `application/json` media type, the use of `4xx` and `5xx` status codes for valid GraphQL requests that fail to execute is discouraged but may be used depending on the implementation. Because of potential ambiguities regarding the source of the server-side error, a `2xx` code should be used with this media type instead.

However, for responses with the `application/graphql-response+json` media type, the server will reply with a `4xx` or `5xx` status code if a valid GraphQL request fails to execute.

## Server implementations

If you use Node.js, we recommend looking at the [list of JS server implementations](/community/tools-and-libraries/?tags=javascript_server).

## Node
You can view a [list of servers written in many other languages here](/community/tools-and-libraries/?tags=server).

If you are using Node.js, we recommend looking at the [list of server implementations](/code/#javascript-server).
## Recap

## Draft Transport Specification
To recap these recommendations for serving GraphQL over HTTP:

A detailed [HTTP transport specification](https://github.com/graphql/graphql-over-http) is in development. Though it is not yet finalized, these draft specifications act as a single source of truth for GraphQL client & library maintainers, detailing how to expose and consume a GraphQL API using an HTTP transport. Unlike the language specification, adherence is not mandatory, but most implementations are moving towards these standards to maximize interoperability.
- The server should handle user authentication before a GraphQL request is validated; authorization should happen during request execution against the schema
- GraphQL APIs are exposed at a single endpoint, typically ending with `/graphql`
- GraphQL clients and servers should support JSON, but may support other formats
- Client should indicate a media type on the `Accept` header of `application/graphql-response+json`
- GraphQL requests are sent using the `POST` HTTP method, but query operations may also be sent using the `GET` method
- Clients should also provide a `Content-type` header with a value of `application/json` for `POST` requests
- The full or partial result of an executed GraphQL operation will be available on the `data` key in the JSON-encoded body of the response, while information about any errors raised during validation or execution will be available on the `errors` key
- Response status codes for valid GraphQL operations that fail may vary depending on the indicated media type of the response, but any GraphQL response with a non-null `data` key will provide a `2xx` response