Skip to main content


👻 Primitive and flexible state management for React
Go to Latest
---title: Performancedescription: How to limit extra re-rendersnav: 7.08keywords: performance---
**Note**: This guide has room for improvement. Consider it as FYI for now.
Jotai & React gives us quite a few tools to manage the re-renders that happen in the app lifecycle.First, please read about the difference [between render & commit](, because that's very important to understand before going further.
### Cheap renders
As seen in the [core section](../core/atom.mdx), due to React 18 default behaviour (but overall good practice), you have to make sure your component functions are _idempotent_.They will be called multiple times during the render phase, even at mount. So we need to keep our renders cheap at all cost!
#### Heavy computation
Always make heavy computation outside of the React lifecycle (in actions for example)
```js// Heavy computation for each itemconst selector = (s) => s.filter(heavyComputation)const Profile = () => { const [computed] = useAtom(selectAtom(friendsAtom, selector))}```
```jsconst friendsAtom = atom([])const fetchFriendsAtom = atom(null, async (get, set, payload) => { // Fetch all friends const res = await fetch('https://...') // Make heavy computation once only const computed = res.filter(heavyComputation) set(friendsAtom, computed)})// Usage in componentsconst Profile = () => { const [friends] = useAtom(friendsAtom)}```
#### Small components
Observed atoms should only re-render small parts of your application that required an update. The less comparison React has to make, the shorter your render time will be.
```jsxconst Profile = () => { const [name] = useAtom(nameAtom) const [ageAtom] = useAtom(ageAtom) return ( <> <div>{name}</div> <div>{age}</div> </> )}```
```jsxconst NameComponent = () => { const [name] = useAtom(nameAtom) return <div>{name}</div>}const AgeComponent = () => { const [age] = useAtom(ageAtom) return <div>{age}</div>}const Profile = () => { return ( <> <NameComponent /> <AgeComponent /> </> )}```
### Render on demand
Usually, the main performance overhead will come from re-rendering parts of your app that did not need to, or way more than they should.
We have a few tools to deal with "when" React should render our components. If you have not seen the usage of `useMemo` and `useCallback`, please check the offcial React documentation for more info before going further.They are of great use to reduce un-necessary renders where your app is not fluid.
But Jotai also provides its set of tools to handle the "when" our atoms should trigger a re-render.
- Out of the box, Jotai encourages you to split your data into atomic parts, hence each atom is stored separately and will only trigger a re-render when their own value change- `selectAtom` allows you to subscribe to specific part of a large object and only re-render on value change- `focusAtom` same as selectAtom, but creating a new atom for the part, giving a setter to update that specific part easily- `splitAtom` does the work of selectAtom/focus atom for a dynamic listWhile this seems simplistic, it is simple to reason about. That was the goal, let's keep it simple to keep it fast.
#### Frequent or rare updates
Ask yourself whether your atom is usually going to be frequently update or more rarely.Let's imagine an atom containing an object that changes almost every second, it may not be best suited to "focus" on a specific properties of this object using `focusAtom`, because anyway they will all re-render in the same time, so best adding no overhead and not create any more atoms.
On the other hand, if your object has properties that rarely change, and most importantly, that change independently from the other properties, then you may want to use `focusAtom` or `selectAtom` to prevent un-necessary renders.
### "Stop observing" pattern
An example of pattern that can be interesting is to use `useMemo` to read an atom value only once, in order to prevent further renders even if the atom changes down the line.
Let's imagine a case, you have a list of toggles. Let's view 2 approaches for it:
### Standard pattern
We create our store of 3 toggles set to false
```tsxconst togglesAtom = atom([false, false, false])```
Then, when the user clicks one toggle, we update it
```tsxconst Item = ({ index, val }) => { const setToggles = useSetAtom(togglesAtom) const onPress = () => { setToggles(old => [...old, [index]: !val]) }}
const List = () => { const [toggles] = useAtom(togglesAtom) return, index) => <Item id={index} val={val} />)}```
With this approach, updating any toggle will cause all `<Item />` to re-render.
### Memoized pattern
Now let's try to memoize the value on the first render
```tsxconst List = () => { const [toggles] = useMemo(() => useAtom(togglesAtom), []) return, index) => <Item id={index} initialValue={val} />)}```
But now it means we have to handle the changes locally in each `Item`
```tsxconst Item = ({ index, initialValue }) => { const setToggles = useSetAtom(togglesAtom) const [toggle, setToggle] = React.useMemo(() => useAtom(atom(val)), []) const onPress = () => { setToggle(!toggle) // Update locally setToggles(old => [...old, [index]: !val]) // Update the main atom }}```
Now if you update one toggle, it will only re-render the corresponding `<Item />`