View Source Safety Limits

A misbehaving client might send a very complex GraphQL query that would require considerable resources to handle. There are two variations of this problem:

  • Complex queries that overwhelm resolution resources.
  • Extremely long queries that could take considerable resources to parse. (For example, an attacker could craft a long query including thousands of undefined fields or directives.)

Either of these could be a vector for a denial-of-service attack in any GraphQL service. Absinthe includes mechanisms to protect services each of these.

Complexity Analysis

To protect against queries that could overwhelm available resources, the complexity of a query can be estimated before it is resolved and limited to a specified maximum.

For example, to enable complexity analysis and limit the complexity to a value of 50 -- if we were using Absinthe.run/3 directly -- we would do this:

Absinthe.run(doc, MyAppWeb.Schema, analyze_complexity: true, max_complexity: 50)

That would translate to the following configuration when using absinthe_plug (>= v1.2.3):

plug Absinthe.Plug,
  schema: MyAppWeb.Schema,
  analyze_complexity: true,
  max_complexity: 50

The maximum value, 50, is compared to complexity values calculated for each request.

Complexity Analysis

Here's how the complexity value is calculated:

By default each field in a query will increase the complexity by 1. However, it can be useful to customize how the complexity value for a field is calculated. This is done in your schema using the complexity/1 macro, which can accept a function or an explicit integer value.

As an example, when a field is a list, the complexity is often correlated to the size of the list. To prevent large selections, a field can use a limit argument with a suitable default (think, for instance, of page sizes during pagination), and complexity can be calculated keeping that in mind. Here is a schema that supports analyzing (and limiting) complexity using that approach:

defmodule MyAppWeb.Schema do

  use Absinthe.Schema

  query do
    field :people, list_of(:person) do
      arg :limit, :integer, default_value: 10
      complexity fn %{limit: limit}, child_complexity ->
        # set complexity based on maximum number of items in the list and
        # complexity of a child.
        limit * child_complexity
      end
    end
  end

  object :person do
    field :name, :string
    field :age, :integer
  end

end

For a field, the first argument to the function you supply to complexity/1 is the user arguments -- just as a field's resolver can use user arguments to resolve its value, the complexity function that you provide can use the same arguments to calculate the field's complexity.

The second argument passed to your complexity function is the sum of all the complexity scores of all the fields nested below the current field.

(If a complexity function accepts three arguments, the third will be an %Absinthe.Resolution{} struct, just as with resolvers.)

If the value of a document's :limit argument was 10 and both name and age were queried for, the complexity of the :people field would be calculated as 20:

  • 10, the value of :limit
  • times 2, the sum of the complexity of the fields requested on the :person

A field's complexity will default to 1 if it's not set explicitly.

So this would be okay:

{
  people(limit: 10) {
    name
  }
}

But this, at a complexity of 60, wouldn't:

{
  people(limit: 30) {
    name
  }
}

Complexity limiting

If a document's calculated complexity exceeds the configured limit, resolution will be skipped and an error will be returned in the result detailing the calculated and maximum complexities.

Token Limits

To protect a service from invalid queries that could take considerable resources to parse, Absinthe offers the option to configure a maximum limit on tokens in the GraphQL request document. If the lexer encounters more tokens than this, it will stop the parse phase and return a phase error with the message "Token limit exceeded". This limit is :infinity by default (no limit) and can be overridden by providing an integer to the option token_limit to Absinthe.run.