urql

Schema Awareness

Previously, on the "Normalized Caching" page we've seen how Graphcache stores normalized data in its store and how it traverses GraphQL documents to do so. What we've seen is that just using the GraphQL document for traversal and the __typename introspection field Graphcache is able to build a normalized caching structure that keeps our application up-to-date across API results, allows it to store data by entities and keys, and provides us configuration options to write manual cache updates and local resolvers.

While this is all possible without any information about a GraphQL API's schema, the schema option on cacheExchange allows us to pass an introspected schema to Graphcache:

const introspectedSchema = {
  __schema: {
    queryType: { name: 'Query', },
    mutationType: { name: 'Mutation', },
    subscriptionType: { name: 'Subscription', },
  },
};


cacheExchange({ schema: introspectedSchema });

In GraphQL, APIs allow for the entire schema to be "introspected", which are special GraphQL queries that give us information on what the API supports. This information can either be retrieved from a GraphQL API directly or from the GraphQL.js Schema and contains a list of all types, the types' fields, scalars, and other information.

In Graphcache we can pass this schema information to enable several features that aren't enabled if we don't pass any information to this option:

  • Fragments will be matched deterministically: A fragment can be written to be on an interface type or multiple fragments can be spread for separate union'ed types in a selection set. In many cases, if Graphcache doesn't have any schema information then it won't know what possible types a field can return and may sometimes make a best guess and issue a warning. If we pass Graphcache a schema then it'll be able to match fragments deterministically.
  • A schema may have non-default names for its root types; Query, Mutation, and Subscription. The names can be changed by passing schema information to cacheExchange which is important if the root type appears elsewhere in the schema, e.g. if the Query can be accessed on a Mutation field's result.
  • We may write a lot of configuration for our cacheExchange but if we pass a schema then it'll start checking whether any of the configuration options actually don't exist, maybe because we've typo'd them. This is a small detail but can make a large different in a longer configuration.
  • Lastly; a schema contains information on which fields are optional or required. When Graphcache has a schema it knows which fields can be made optional and it'll be able to generate "partial results".

Partial Results

As we navigate an app that uses Graphcache we may be in states where some of our data is already cached and some isn't. Graphcache normalizes data and stores it in tables for links and records for each entity, which means that sometimes it can maybe even execute a query against its cache that it hasn't sent to the API before.

On the "Local Resolvers" page we've seen how to write resolvers that resolver entities without having to have seen a link from an API result before. If Graphcache uses these resolvers and previously cached data we often run into situations where a "partial result" could already be generated, which is what Graphcache does when it has schema information.

A "partial result" is an incomplete result of information that Graphcache already had cached before it sent an API result.

Without a schema and information on which fields are optional, Graphcache will consider a "partial result" as a cache miss. If we don't have all of the information for a query then we can't execute it against the locally cached data after all. However, an API's schema contains information on which fields are required and which fields are optional, and if our apps are typed with this schema and TypeScript, can't we then use and handle these partial results before a request is sent to the API?

This is the idea behind "Schema Awareness" and "Partial Results". When Graphcache has schema information it may give us partial results with the stale flag set while it fetches the full result from the API in the background. This allows our apps to show some information while more is loading.

Getting your schema

But how do you get an introspected schema? The process of introspecting a schema is running an introspection query on the GraphQL API, which will give us our IntrospectionQuery result. So an introspection is just another query we can run against our GraphQL APIs or schemas.

As long as introspection is turned on and permitted, we can download an introspection schema by running a normal GraphQL query against the API and save the result in a JSON file.

import { getIntrospectionQuery } from 'graphql';
import fetch from 'node-fetch'; // or your preferred request in Node.js
import * as fs from 'fs';


fetch('http://localhost:3000/graphql', {
  method: 'POST',
  headers: { 'Content-Type': 'application/json' },
  body: JSON.stringify({
    variables: {},
    query: getIntrospectionQuery({ descriptions: false }),
  }),
})
  .then(result => result.json())
  .then(({ data }) => {
    fs.writeFile('./schema.json', JSON.stringify(data), err => {
      if (err) {
        console.error('Writing failed:', err);
        return;
      }
      console.log('Schema written!');
    });
  });

Alternatively, if you're already using GraphQL Code Generator you can use their @graphql-codegen/introspection plugin to do the same automatically against a local schema. Furthermore it's also possible to execute the introspection query directly against your GraphQLSchema.

Optimizing a schema

An IntrospectionQuery JSON blob from a GraphQL API can without modification become quite large. The shape of this data is { "__schema": ... } and this schema data will contain information on all directives, types, input objects, scalars, deprecation, enums, and more. This can quickly add up and one of the largest schemas, the GitHub GraphQL API's schema, has an introspection size of about 1.1MB, or about 50KB gzipped.

However, we can use the @urql/introspection package's minifyIntrospectionQuery helper to reduce the size of this introspection data. This helper strips out information on directives, scalars, input types, deprecation, enums, and redundant fields to only leave information that Graphcache actually requires.

In the example of the GitHub GraphQL API this reduces the introspected data to around 20kB gzipped, which is much more acceptable.

Installation & Setup

First, install the @urql/introspection package:

yarn add @urql/introspection
# or
npm install --save @urql/introspection

You'll then need to integrate it into your introspection script or in another place where it can optimise the introspection data. For this example, we'll just add it to the fetching script from above.

import { getIntrospectionQuery } from 'graphql';
import fetch from 'node-fetch'; // or your preferred request in Node.js
import * as fs from 'fs';


import { getIntrospectedSchema, minifyIntrospectionQuery } from '@urql/introspection';


fetch('http://localhost:3000/graphql', {
  method: 'POST',
  headers: { 'Content-Type': 'application/json' },
  body: JSON.stringify({
    variables: {},
    query: getIntrospectionQuery({ descriptions: false }),
  }),
})
  .then(result => result.json())
  .then(({ data }) => {
    const minified = minifyIntrospectionQuery(getIntrospectedSchema(data));
    fs.writeFileSync('./schema.json', JSON.stringify(minified));
  });

The getIntrospectionSchema doesn't only accept IntrospectionQuery JSON data as inputs, but also allows you to pass a JSON string, GraphQLSchema, or GraphQL Schema SDL strings. It's a convenience helper and not needed in the above example.

Integrating a schema

Once we have a schema that's already saved to a JSON file, we can load it and pass it to the cacheExchange's schema option:

import schema from './schema.json';


const cache = cacheExchange({ schema });

It may be worth checking what your bundler or framework does when you import a JSON file. Typically you can reduce the parsing time by making sure it's turned into a string and parsed using JSON.parse