GraphQL Depth Limit
Dead-simple defense against unbounded GraphQL queries. Limit the complexity of the queries solely by their depth.
Why?
Suppose you have an Album
type that has a list of Song
s.
{
album(id: 42) {
songs {
title
artists
}
}
}
And perhaps you have a different entry point for a Song
and the type allows you to go back up to the Album
.
{
song(id: 1337) {
title
album {
title
}
}
}
That opens your server to the possibility of a cyclical query!
query evil {
album(id: 42) {
songs {
album {
songs {
album {
songs {
album {
songs {
album {
songs {
album {
songs {
album {
# and so on...
}
}
}
}
}
}
}
}
}
}
}
}
}
}
How would your server handle it if the query was 10,000 deep? This may become a very expensive operation, at some point pinning a CPU on the server or perhaps the database. This is a possible DOS vulnerability. We want a way to validate the complexity of incoming queries.
This implementation lets you limit the total depth of each operation.
Quantifying Complexity
Deciding exactly when a GraphQL query is "too complex" is a nuanced and subtle art. It feels a bit like deciding how many grains of sand are needed to compose a "pile of sand". Some other libraries have the developer assign costs to parts of the schema, and adds the cumulative costs for each query.
graphql-validation-complexity does this based on the types, and graphql-query-complexity does it based on each field.
Adding up costs may work for some backends, but it does not always faithfully represent the complexity.
By adding the costs at each depth, it's as if the complexity is increasing lineraly with depth.
Sometimes the complexity actually increases exponentially with depth, for example if requesting a field means doing another SQL JOIN
.
This library validates the total depth of the queries (and mutations).
Here's are some queries with a depth of 0
# simplest possible query
query shallow1 {
thing1
}
# inline fragments don't actually increase the depth
query shallow2 {
thing1
... on Query {
thing2
}
}
# neither do named fragments
query shallow3 {
...queryFragment
}
fragment queryFragment on Query {
thing1
}
Deeper queries
# depth = 1
query deep1_1 {
viewer {
name
}
}
query deep1_2 {
viewer {
... on User {
name
}
}
}
# depth = 2
query deep2 {
viewer {
albums {
title
}
}
}
# depth = 3
query deep3 {
viewer {
albums {
...musicInfo
songs{
...musicInfo
}
}
}
}
fragment musicInfo on Music {
id
title
artists
}
Usage
$ npm install graphql-depth-limit
It works with express-graphql and koa-graphql. Here is an example with Express.
import depthLimit from 'graphql-depth-limit'
import express from 'express'
import graphqlHTTP from 'express-graphql'
import schema from './schema'
const app = express()
app.use('/graphql', graphqlHTTP((req, res) => ({
schema,
validationRules: [ depthLimit(10) ]
})))
The first argument is the total depth limit. This will throw a validation error for queries (or mutations) with a depth of 11 or more.
The second argument is an options object, where you can do things like specify ignored fields. Introspection fields are ignored by default.
The third argument is a callback which receives an Object
which is a map of the depths for each operation.
depthLimit(
10,
{ ignore: [ /_trusted$/, 'idontcare' ] },
depths => console.log(depths)
)
Now the evil query from before will tell the client this:
{
"errors": [
{
"message": "'evil' exceeds maximum operation depth of 10",
"locations": [
{
"line": 13,
"column": 25
}
]
}
]
}
Future Work
- Type-specific sub-depth limits, e.g. you can only descend 3 levels from an
Album
type, 5 levels from theUser
type, etc. - More customization options, like custom errors.
Documentation
function
depthLimit(maxDepth, [options], [callback]) ⇒ Creates a validator for the GraphQL query depth
Kind: global function
Returns: function
- The validator function for GraphQL validation phase.
Param | Type | Description |
---|---|---|
maxDepth | Number |
The maximum allowed depth for any operation in a GraphQL document. |
[options] | Object |
|
options.ignore | Array.<(String|RegExp|function())> |
Stops recursive depth checking based on a field name. Either a string or regexp to match the name, or a function that reaturns a boolean. |
[callback] | function |
Called each time validation runs. Receives an Object which is a map of the depths for each operation. |