cacheThis feature is available in the latest Canary

Canary

cache lets you cache the result of a data fetch or computation.

const cachedFn = cache(fn);

Reference

cache(fn)

Call cache outside of any components to create a version of the function with caching.

import {cache} from 'react';
import calculateMetrics from 'lib/metrics';

const getMetrics = cache(calculateMetrics);

function Chart({data}) {
const report = getMetrics(data);
// ...
}

When getMetrics is first called with data, getMetrics will call calculateMetrics(data) and store the result in cache. If getMetrics is called again with the same data, it will return the cached result instead of calling calculateMetrics(data) again.

See more examples below.

Parameters

  • fn: The function you want to cache results for. fn can take any arguments and return any value.

Returns

cache returns a cached version of fn with the same type signature. It does not call fn in the process.

When calling cachedFn with given arguments, it first checks if a cached result exists in the cache. If a cached result exists, it returns the result. If not, it calls fn with the arguments, stores the result in the cache, and returns the result. The only time fn is called is when there is a cache miss.

補足

The optimization of caching return values based on inputs is known as memoization. We refer to the function returned from cache as a memoized function.

Caveats

  • React will invalidate the cache for all memoized functions for each server request.
  • Each call to cache creates a new function. This means that calling cache with the same function multiple times will return different memoized functions that do not share the same cache.
  • cachedFn will also cache errors. If fn throws an error for certain arguments, it will be cached, and the same error is re-thrown when cachedFn is called with those same arguments.
  • cache is for use in Server Components only.

Usage

Cache an expensive computation

Use cache to skip duplicate work.

import {cache} from 'react';
import calculateUserMetrics from 'lib/user';

const getUserMetrics = cache(calculateUserMetrics);

function Profile({user}) {
const metrics = getUserMetrics(user);
// ...
}

function TeamReport({users}) {
for (let user in users) {
const metrics = getUserMetrics(user);
// ...
}
// ...
}

If the same user object is rendered in both Profile and TeamReport, the two components can share work and only call calculateUserMetrics once for that user.

Assume Profile is rendered first. It will call getUserMetrics, and check if there is a cached result. Since it is the first time getUserMetrics is called with that user, there will be a cache miss. getUserMetrics will then call calculateUserMetrics with that user and write the result to cache.

When TeamReport renders its list of users and reaches the same user object, it will call getUserMetrics and read the result from cache.

落とし穴

Calling different memoized functions will read from different caches.

To access the same cache, components must call the same memoized function.

// Temperature.js
import {cache} from 'react';
import {calculateWeekReport} from './report';

export function Temperature({cityData}) {
// 🚩 Wrong: Calling `cache` in component creates new `getWeekReport` for each render
const getWeekReport = cache(calculateWeekReport);
const report = getWeekReport(cityData);
// ...
}
// Precipitation.js
import {cache} from 'react';
import {calculateWeekReport} from './report';

// 🚩 Wrong: `getWeekReport` is only accessible for `Precipitation` component.
const getWeekReport = cache(calculateWeekReport);

export function Precipitation({cityData}) {
const report = getWeekReport(cityData);
// ...
}

In the above example, Precipitation and Temperature each call cache to create a new memoized function with their own cache look-up. If both components render for the same cityData, they will do duplicate work to call calculateWeekReport.

In addition, Temperature creates a new memoized function each time the component is rendered which doesn’t allow for any cache sharing.

To maximize cache hits and reduce work, the two components should call the same memoized function to access the same cache. Instead, define the memoized function in a dedicated module that can be import-ed across components.

// getWeekReport.js
import {cache} from 'react';
import {calculateWeekReport} from './report';

export default cache(calculateWeekReport);
// Temperature.js
import getWeekReport from './getWeekReport';

export default function Temperature({cityData}) {
const report = getWeekReport(cityData);
// ...
}
// Precipitation.js
import getWeekReport from './getWeekReport';

export default function Precipitation({cityData}) {
const report = getWeekReport(cityData);
// ...
}

Here, both components call the same memoized function exported from ./getWeekReport.js to read and write to the same cache.

Share a snapshot of data

To share a snapshot of data between components, call cache with a data-fetching function like fetch. When multiple components make the same data fetch, only one request is made and the data returned is cached and shared across components. All components refer to the same snapshot of data across the server render.

import {cache} from 'react';
import {fetchTemperature} from './api.js';

const getTemperature = cache(async (city) => {
return await fetchTemperature(city);
});

async function AnimatedWeatherCard({city}) {
const temperature = await getTemperature(city);
// ...
}

async function MinimalWeatherCard({city}) {
const temperature = await getTemperature(city);
// ...
}

If AnimatedWeatherCard and MinimalWeatherCard both render for the same city, they will receive the same snapshot of data from the memoized function.

If AnimatedWeatherCard and MinimalWeatherCard supply different city arguments to getTemperature, then fetchTemperature will be called twice and each call site will receive different data.

The city acts as a cache key.

補足

Asynchronous rendering is only supported for Server Components.

async function AnimatedWeatherCard({city}) {
const temperature = await getTemperature(city);
// ...
}

Preload data

By caching a long-running data fetch, you can kick off asynchronous work prior to rendering the component.

const getUser = cache(async (id) => {
return await db.user.query(id);
}

async function Profile({id}) {
const user = await getUser(id);
return (
<section>
<img src={user.profilePic} />
<h2>{user.name}</h2>
</section>
);
}

function Page({id}) {
// ✅ Good: start fetching the user data
getUser(id);
// ... some computational work
return (
<>
<Profile id={id} />
</>
);
}

When rendering Page, the component calls getUser but note that it doesn’t use the returned data. This early getUser call kicks off the asynchronous database query that occurs while Page is doing other computational work and rendering children.

When rendering Profile, we call getUser again. If the initial getUser call has already returned and cached the user data, when when Profile asks and waits for this data, it can simply read from the cache without requiring another remote procedure call. If the initial data request hasn’t been completed, preloading data in this pattern reduces delay in data-fetching.

さらに深く知る

Caching asynchronous work

When evaluating an asynchronous function, you will receive a Promise for that work. The promise holds the state of that work (pending, fulfilled, failed) and its eventual settled result.

In this example, the asynchronous function fetchData returns a promise that is awaiting the fetch.

async function fetchData() {
return await fetch(`https://...`);
}

const getData = cache(fetchData);

async function MyComponent() {
getData();
// ... some computational work
await getData();
// ...
}

In calling getData the first time, the promise returned from fetchData is cached. Subsequent look-ups will then return the same promise.

Notice that the first getData call does not await whereas the second does. await is a JavaScript operator that will wait and return the settled result of the promise. The first getData call simply initiates the fetch to cache the promise for the second getData to look-up.

If by the second call the promise is still pending, then await will pause for the result. The optimization is that while we wait on the fetch, React can continue with computational work, thus reducing the wait time for the second call.

If the promise is already settled, either to an error or the fulfilled result, await will return that value immediately. In both outcomes, there is a performance benefit.

落とし穴

Calling a memoized function outside of a component will not use the cache.
import {cache} from 'react';

const getUser = cache(async (userId) => {
return await db.user.query(userId);
});

// 🚩 Wrong: Calling memoized function outside of component will not memoize.
getUser('demo-id');

async function DemoProfile() {
// ✅ Good: `getUser` will memoize.
const user = await getUser('demo-id');
return <Profile user={user} />;
}

React only provides cache access to the memoized function in a component. When calling getUser outside of a component, it will still evaluate the function but not read or update the cache.

This is because cache access is provided through a context which is only accessibile from a component.

さらに深く知る

When should I use cache, memo, or useMemo?

All mentioned APIs offer memoization but the difference is what they’re intended to memoize, who can access the cache, and when their cache is invalidated.

useMemo

In general, you should use useMemo for caching a expensive computation in a Client Component across renders. As an example, to memoize a transformation of data within a component.

'use client';

function WeatherReport({record}) {
const avgTemp = useMemo(() => calculateAvg(record)), record);
// ...
}

function App() {
const record = getRecord();
return (
<>
<WeatherReport record={record} />
<WeatherReport record={record} />
</>
);
}

In this example, App renders two WeatherReports with the same record. Even though both components do the same work, they cannot share work. useMemo’s cache is only local to the component.

However, useMemo does ensure that if App re-renders and the record object doesn’t change, each component instance would skip work and use the memoized value of avgTemp. useMemo will only cache the last computation of avgTemp with the given dependencies.

cache

In general, you should use cache in Server Components to memoize work that can be shared across components.

const cachedFetchReport = cache(fetchReport);

function WeatherReport({city}) {
const report = cachedFetchReport(city);
// ...
}

function App() {
const city = "Los Angeles";
return (
<>
<WeatherReport city={city} />
<WeatherReport city={city} />
</>
);
}

Re-writing the previous example to use cache, in this case the second instance of WeatherReport will be able to skip duplicate work and read from the same cache as the first WeatherReport. Another difference from the previous example is that cache is also recommended for memoizing data fetches, unlike useMemo which should only be used for computations.

At this time, cache should only be used in Server Components and the cache will be invalidated across server requests.

memo

You should use memo to prevent a component re-rendering if its props are unchanged.

'use client';

function WeatherReport({record}) {
const avgTemp = calculateAvg(record);
// ...
}

const MemoWeatherReport = memo(WeatherReport);

function App() {
const record = getRecord();
return (
<>
<MemoWeatherReport record={record} />
<MemoWeatherReport record={record} />
</>
);
}

In this example, both MemoWeatherReport components will call calculateAvg when first rendered. However, if App re-renders, with no changes to record, none of the props have changed and MemoWeatherReport will not re-render.

Compared to useMemo, memo memoizes the component render based on props vs. specific computations. Similar to useMemo, the memoized component only caches the last render with the last prop values. Once the props change, the cache invalidates and the component re-renders.


Troubleshooting

My memoized function still runs even though I’ve called it with the same arguments

See prior mentioned pitfalls

If none of the above apply, it may be a problem with how React checks if something exists in cache.

If your arguments are not primatives (ex. objects, functions, arrays), ensure you’re passing the same object reference.

When calling a memoized function, React will look up the input arguments to see if a result is already cached. React will use shallow equality of the arguments to determine if there is a cache hit.

import {cache} from 'react';

const calculateNorm = cache((vector) => {
// ...
});

function MapMarker(props) {
// 🚩 Wrong: props is an object that changes every render.
const length = calculateNorm(props);
// ...
}

function App() {
return (
<>
<MapMarker x={10} y={10} z={10} />
<MapMarker x={10} y={10} z={10} />
</>
);
}

In this case the two MapMarkers look like they’re doing the same work and calling calculateNorm with the same value of {x: 10, y: 10, z:10}. Even though the objects contain the same values, they are not the same object reference as each component creates its own props object.

React will call Object.is on the input to verify if there is a cache hit.

import {cache} from 'react';

const calculateNorm = cache((x, y, z) => {
// ...
});

function MapMarker(props) {
// ✅ Good: Pass primitives to memoized function
const length = calculateNorm(props.x, props.y, props.z);
// ...
}

function App() {
return (
<>
<MapMarker x={10} y={10} z={10} />
<MapMarker x={10} y={10} z={10} />
</>
);
}

One way to address this could be to pass the vector dimensions to calculateNorm. This works because the dimensions themselves are primitives.

Another solution may be to pass the vector object itself as a prop to the component. We’ll need to pass the same object to both component instances.

import {cache} from 'react';

const calculateNorm = cache((vector) => {
// ...
});

function MapMarker(props) {
// ✅ Good: Pass the same `vector` object
const length = calculateNorm(props.vector);
// ...
}

function App() {
const vector = [10, 10, 10];
return (
<>
<MapMarker vector={vector} />
<MapMarker vector={vector} />
</>
);
}