@pipedrive/graphql-query-cost

Graphql cost analysis, mainly to limit query execution in graphql service before resolvers are executed.

Features

• Query cost calculation based on schema and cost mappings
• Cost directive for the schema
• Cost directive extraction from the schema
• Cost directive value extraction from the schema as cost mappings

API

In the API documentation, @pipedrive/graphql-query-cost is referred to as queryCost, as if it was imported with:

const queryCost = require('@pipedrive/graphql-query-cost');

queryCost.costDirective

A custom graphql directive for defining expenses directly in the schema.

const schema = `
  ${queryCost.costDirective}
  type Greeting {
    id: ID
    name: String
  }
  type Query {
    hello(limit: Int!): Greeting @cost(
      complexity: 10,
      network: 1,
      db: 1,
      useMultiplers: true,
      multipliers: ["limit"]
      provides: ["id"]
    )
  }
`;

Directive arguments

queryCost.extractCost

Returns cost directive arguments per type definition.

const schema = `
  ${queryCost.costDirective}
  type Greeting {
    id: ID
    name: String @cost(db: 10)
  }
  type Query {
    hello: Greeting @cost(complexity: 10)
  }
`;
const { costMap } = queryCost.extractCost(schema);
// {
//   Query: {
//     hello: {
//       complexity: 10,
//     },
//   },
//   Greeting: {
//     hello: {
//       tokens: 1000, // n * 100
//     },
//   },
// };

queryCost.calculateCost

Calculates cost of the current query based on cost mappings and schema.

const schema = `
  type Query {
    hello(limit: Int!): string
    world: string
  }
`
const query = `
  query makeQuery($limit)  {
    hello(limit: $limit)
    world
  }
`
const costMap = {
  Query: {
    hello: {
      complexity: 5
      multipliers: ['limit']
    }
  }
}
const cost = queryCost.calculateCost(
  query, schema, {
    costMap,
    defaultCost: 1,
    variables: {
      $limit: 5
    }
  }
)
// (5 * 5) + 1 = 11

Cost calculation

• Definition of cost for entity/resolver/property should be pessimistic (considered for worst case scenario, cold cache)

Complexity vs tokens(db, network)

• Complexity argument is multiplied by its own multipliers and every parent mutliplier.

  deals(limit: 100) @cost(complexity: 2) = 200
  

• Tokens define how many resources are need to resolve a field. When it's recursive it is multiplied.
• Tokens are only multiplied only by its parents multipliers. Only parent multipliers are used because it takes "parent" times resources to execute the query.
  Example query

  deals(limit: 100)
    @cost(complexity: 2, db: 1, network: 1)
  

  Example token cost

  complexity(2) * limit(100) + tokens(200) = 400
  

  If tokens were multiplied by own multipler

  complexity(200) + tokens(200) * limit(100) = 20200
  

Flat complexity

# Schema
type Query {
  field: String @cost(complexity: 3)
  default: String
}

# Query
query {
  field
  default
}

Total cost is 4:

• field cost is 3
• default cost is 1

Multipliers

• Multipliers are recursive
• Undefined(Parent.name) complexity is not multiplied

# Schema
type Query {
  parents(limit: Int!, names: [String]): [Parent]
  @cost(complexity: 3, multipliers: ["limit", "names"])
}

type Parent {
  name: String
  children(limit: Int): [Child] @cost(complexity: 5)
}

type Child {
  name: String
}

# Query
{
  parents(limit: 2, names: ["elon", "foo"]) {
    name
    children(limit: 4) {
      name
    }
  }
}

Ignoring multipliers

useMultipliers

• Definining useMultipliers: false ignores the multipliers

# Schema
type Query {
  parents(limit: Int): [Parent] @cost(complexity: 2, multipliers: ["limit"])
}

type Parent {
  name: String @cost(complexity: 8, useMultipliers: false)
}
# Query
{
  parents(limit: 5) {
    name
  }
}

provides

• If all the queried fields are in the list of provides argument, then the complexity of the field is ignored and the default cost is applied

# Schema
type Query {
  parents(limit: Int): [Parent] @cost(complexity: 3, multipliers: ["limit"], provides: ["id"])
}

type Parent {
  id: ID @cost(complexity: 1)
  name: String
}

# Query
{
  parents(limit: 5) {
    id
  }
}

Recursive queries

# Schema
type Query {
  pipelines: [Pipeline]
}

type Pipeline {
  id: ID
  deals: [Deal]
}

type Deal {
  id: ID
  pipeline: Pipeline
}

# Query
{
  pipelines {
    deals {
      pipeline {
        deals {
          pipeline {
            id
          }
        }
      }
    }
  }
}

Recursion multiplier

recursionMultiplier default value is 100. If recursion is detected, thats how much cost gets affected. recursionMultiplier value of 1 will mean that detected recursion doesn't affect cost. recursionMultiplier value gets inherited to deeper levels of the graph, so you can have different values depending on the schema

For cases where schema and resolver can have deeply nested recursive structures (trees or graphs) fetched once into memory, you can override default behaviour by setting recursionMultiplier to lower values.

Usage

With maximum cost of 5000 per request, you can see different values of recursionMultiplier affects graphq and schema below:

• It is recommended that you don't set recursion to 1, but still leave some recursion scaling limits, such that your query still fits into max cost limit.

# Schema
type Query {
  myTree: [TreeLeaf]
}

type TreeLeaf {
  id: ID
  leafs: [TreeLeaf] @cost(recursionMultiplier: 1)
}

# Query
{
  myTree {
    leafs {
      leafs {
        leafs {
          leafs {
            id
          }
        }
      }
    }
  }
}

• With great recursionMultiplier override power comes great responsibility that you don't reset it to 1 and leave holes in graph cost map.

type Query {
  myTree: [TreeLeaf]
}

type TreeLeaf {
  id: ID

  # reset recursionMultiplier back to 100, as its part of the recursive graph
  branch: Branch @cost(recursionMultiplier: 100)
  leafs: [TreeLeaf] @cost(recursionMultiplier: 1)
}

type Branch {
  id: ID
  name: String @cost(network: 1)
  leafs: [TreeLeaf]
}

Development

Cost debugging

To ease understanding how cost is calculated, you can use debug:true param, which will console.log node visiting along with how price is added. (Maybe it could be a product feature later if further )

	calculateCost(`query { a { b { c { b {c { id }}}}}}`, typeDefs, { costMap, debug: true });

Debug result:

       undefined (<OperationDefinition>)
        a <Field>
         a (<Field>)
          b <Field>
           b (<Field>)
            c <Field>
             c (<Field>)
              b <Field>
               b (<Field>)
                c <Field>
                recursion detected b=>c for 1 times
                recursion detected c=>bfor 1 times
                * recursion multiplier = (10000)
                 c (<Field>)
                  id <Field>
                  = 1
                 == 1
                = 20000
               == 20000
              = 20001
             == 20001
            = 20002
           == 20002
          = 20003
         == 20003
        = 20004
       == 20004

Contribution

• Before making PR, make sure to run npm run version & fill CHANGELOG
• npm-version-<version> – should be set when creating a pull request. It’s good to set it as soon as possible, so reviewer can validate that the proposed version bump is correct
• npm-ready-for-publish – add this label after PR has been approved, it will publish the package to NPM and merge changes to master

Authors and acknowledgment

Original author - @ErikSchults. Current maintainers - @tot-ra, @Wolg. Mention in PR, if it is stuck

See blog post