As a GraphQL API gives you a lot of extra power and possibilities, it also introduces some new attack vectors.
For example, you have build some client applications consuming your GraphQL API. These applications are using a range of queries that you carefully created to implement the client functionality. These queries are well tested, optimized and don't stretch your datasource too much.
But nothing prevents the user of your (web) application to open the developer console and start creating and sending other queries to your GraphQL backend. By using the authentication token already available, he/she can call your API. So without further mitigations a user can create and run any query he/she can think of.
Luckily there are multiple ways to control this attack vector. In this blog series I want to share some ways on how you can improve the security of your GraphQL API's. In first post I'll focus on the concept of a complexity budget.
A single GraphQL query can potentially generate huge workload for a server, like thousands of database operations which can be used to cause DDoS attacks. By assigning a "cost" per field and then analyze the AST we can estimate the total cost of the GraphQL query. If this cost exceeds the complexity budget we have given to a user, the query is not executed.
Let me further explain this with an example. In HotChocolate the default cost of every field is 1. By using the custom @cost directive we can change this to any value we want.
If we now execute the following query, the total query cost is 17:
| Field | Cost | ||
|---|---|---|---|
| books | 10 | ||
| title | 1 | ||
| author | 5 | ||
| name | 1 | ||
In the example above, the cost calculation is done by static analysis. But sometimes the cost calculation is more dynamic. For example what if we decide to fetch a 100 items instead of the default 10?
To handle this scenario we can configure a multiplier:
As you can see we are using the take argument as a multiplier.
Limit the operational complexity
Now that we have explained the concept, let’s apply it in practice.
I typically use the code first syntax. There we could set the cost through the Cost function:
Now to limit the complexity we have to update our GraphQL configuration:
Log the operational complexity
To monitor the operational complexity, you can tweak the OpenTelemetry integration and include the AnalyzeComplexity scope: