Version: v10.1.3

Persisted Queries

The relay compiler supports persisted queries which is useful because:

  • the client operation text becomes just an md5 hash which is usually shorter than the real query string. This saves upload bytes from the client to the server.

  • the server can now whitelist queries which improves security by restricting the operations that can be executed by a client.

Usage on the client#

The --persist-output flag#

In your npm script in package.json, run the relay compiler using the --persist-output flag:

"scripts": {
"relay": "relay-compiler --src ./src --schema ./schema.graphql --persist-output ./path/to/persisted-queries.json"

The --persist-ouput flag does 2 things:

  1. It converts all query and mutation operation texts to md5 hashes.

    For example without --persist-output, a generated ConcreteRequest might look like below:

    const node/*: ConcreteRequest*/ = (function(){
    //... excluded for brevity
    return {
    "kind": "Request",
    "operationKind": "query",
    "name": "TodoItemRefetchQuery",
    "id": null, // NOTE: id is null
    "text": "query TodoItemRefetchQuery(\n $itemID: ID!\n) {\n node(id: $itemID) {\n ...TodoItem_item_2FOrhs\n }\n}\n\nfragment TodoItem_item_2FOrhs on Todo {\n text\n isComplete\n}\n",
    //... excluded for brevity

    With --persist-output <path> this becomes:

    const node/*: ConcreteRequest*/ = (function(){
    //... excluded for brevity
    return {
    "kind": "Request",
    "operationKind": "query",
    "name": "TodoItemRefetchQuery",
    "id": "3be4abb81fa595e25eb725b2c6a87508", // NOTE: id is now an md5 hash of the query text
    "text": null, // NOTE: text is null now
    //... excluded for brevity
  2. It generates a JSON file at the <path> you specify containing a mapping from query ids to the corresponding operation texts.

"scripts": {
"relay": "relay-compiler --src ./src --schema ./schema.graphql --persist-output ./src/queryMaps/queryMap.json"

The example above writes the complete query map file to ./src/queryMaps/queryMap.json. You need to ensure all the directories leading to the queryMap.json file exist.

Network layer changes#

You'll need to modify your network layer fetch implementation to pass a documentId parameter in the POST body instead of a query parameter:

function fetchQuery(operation, variables,) {
return fetch('/graphql', {
method: 'POST',
headers: {
'content-type': 'application/json'
body: JSON.stringify({
documentId:, // NOTE: pass md5 hash to the server
// query: operation.text, // this is now obsolete because text is null
}).then(response => {
return response.json();

Executing Persisted Queries on the Server#

To execute client requests that send persisted queries instead of query text, your server will need to be able to lookup the query text corresponding to each id. Typically this will involve saving the output of the --persist-output <path> JSON file to a database or some other storage mechanism, and retrieving the corresponding text for the ID specified by a client.

For universal applications where the client and server code are in one project, this is not an issue since you can place the query map file in a common location accessible to both the client and the server.

Compile time push#

For applications where the client and server projects are separate, one option is to have an additional npm run script to push the query map at compile time to a location accessible by your server:

"scripts": {
"push-queries": "node ./pushQueries.js",
"relay": "relay-compiler --src ./src --schema ./schema.graphql --persist-ouput <path> && npm run push-queries"

Some possibilities of what you can do in ./pushQueries.js:

  • git push to your server repo

  • save the query maps to a database

Run time push#

A second more complex option is to push your query maps to the server at runtime, without the server knowing the query ids at the start. The client optimistically sends a query id to the server, which does not have the query map. The server then in turn requests for the full query text from the client so it can cache the query map for subsequent requests. This is a more complex approach requiring the client and server to interact to exchange the query maps.

Simple server example#

Once your server has access to the query map, you can perform the mapping. The solution varies depending on the server and database technologies you use, so we'll just cover the most common and basic example here.

If you use express-graphql and have access to the query map file, you can import the --persist-output JSON file directly and perform the matching using the matchQueryMiddleware from relay-compiler-plus.

import Express from 'express';
import expressGraphql from 'express-graphql';
import {matchQueryMiddleware} from 'relay-compiler-plus';
import queryMapJson from './path/to/persisted-queries.json';
const app = Express();

Using --persist-output and --watch#

It is possible to continuously generate the query map files by using the --persist-output and --watch options simultaneously. This only makes sense for universal applications i.e. if your client and server code are in a single project and you run them both together on localhost during development. Furthermore, in order for the server to pick up changes to the queryMap.json, you'll need to have server side hot-reloading set up. The details on how to set this up is out of the scope of this document.

Last updated on by Andrey Lunyov