async.mdx

title	description	nav
Async	This doc describes about the behavior with async.	1.03

Using async atoms, you gain access to real-world data while still managing them directly from your atoms and with incredible ease.

We can seperate them in two main categories:

• Async read atoms: async request is started instantly as soon as you try to get its value, you could relate to them as "smart getters"
• Async write atoms: async request is started at a specific moment, you could relate to them as "actions"

Async read atom

The read function of an atom can return a promise.

const countAtom = atom(1)
const asyncAtom = atom(async (get) => get(countAtom) * 2)

Jotai is inherently leveraging Suspense to handle asynchronous flows.

const ComponentUsingAsyncAtoms = () => {
  const [num] = useAtom(asyncAtom)
  // here `num` is always `number` even though asyncAtom returns a Promise
}
const App = () => {
  return (
    <Suspense fallback={/* What to show while suspended */}>
      <ComponentUsingAsyncAtoms />
    </Suspense>
  )
}

Alternatively, you could avoid the inherent suspending that Jotai does for you, by wrapping your atoms with the loadable API.

Note: An atom becomes async not only if the atom read function is async, but also if one or more of its dependencies are async.

const anotherAtom = atom((get) => get(asyncAtom) / 2)
// even though this atom doesn't return a promise,
// it's a read async atom because `asyncAtom` is async.

Note: You cannot get the value of an async atom in a write-atom (whether the write function is async or not) before its value has been resolved

const asyncAtom = atom(async (get) => ...)
const writeAtom = atom(null, (get, set, payload) => {
  get(asyncAtom) // This throws an error if "asyncAtom" is still in pending state
})

If you want to make sure the action will never fail, you could preload atoms at root level of your app directly:

const Preloader = () => {
  useAtomValue(asyncAtom) // The value will be pre-loaded
  return null
}

const Root = () => {
  return (
    <Suspense fallback={<Text>Loading...<Text>}>
      <Preloader /> {/* Wait for atoms to preload */}
      <App /> {/* Rest of your app */}
    </Suspense>
  )
}

Async write atom

Async write atoms are another kind of async atom. When the write function of atom returns a promise.

const countAtom = atom(1)
const asyncIncrementAtom = atom(null, async (get, set) => {
  // await something
  set(countAtom, get(countAtom) + 1)
})

const Component = () => {
  const [, increment] = useAtom(asyncIncrementAtom)
  React.useEffect(() => {
    increment()
  }, [])
}

Async sometimes

An interesting pattern that can be achieved with Jotai are is switching from async to sync to trigger suspending when wanted.

const request = async () => fetch('https://...').then((res) => res.json())
const baseAtom = atom(0)
const Component = () => {
  const [value, setValue] = useAtom(baseAtom)
  React.useEffect(() => {
    setValue(request()) // Will suspend until request resolves
  }, [])
}

Async forever

Sometimes you may want to suspend until an unpredetermined moment (or never).

const baseAtom = atom(new Promise(() => {})) // Will be suspend until set otherwise

Suspense

Async support is first class in Jotai. It fully leverages React Suspense at its core.

Technically, Suspense usage other than React.lazy is still unsupported / undocumented in React 17. If this is blocking, so you can still use the loadable API to avoid suspending

To use async atoms, you need to wrap your component tree with <Suspense>.

If you have a <Provider>, place at least one <Suspense> inside said <Provider>; otherwise, it may cause an endless loop while rendering the components.

const App = () => (
  <Provider>
    <Suspense fallback="Loading...">
      <Layout />
    </Suspense>
  </Provider>
)

Having more <Suspense>s in the component tree is also possible and must be considered to profit from Jotai inherent handling at best.

Async write atom behavior until v1.3.9

(This is no longer the case since v1.4.0.)

Triggering Suspense fallback of write atom

This section applies only for "async write atom" not "async read atom", which works differently with Suspense.

A write atom will trigger the Suspense fallback if:

* the atom's write argument (2nd one) is async
* the awaited call is made directly, and not from inside another containing function

This will trigger the Suspense fallback:

const writeAtom = atom(null, async (get, set) => {
  const response = await new Promise<string>((resolve, _reject) => {
    setTimeout(() => {
      resolve('some returned value')
    }, 2000)
  })
  set(somePrimitiveAtom, 'The returned value is: ' + response)
})

This will not trigger the Suspense fallback:

const writeAtom = atom(null, (get, set) => {
  const getResponse = async () => {
    const response = await new Promise<string>((resolve, _reject) => {
      setTimeout(() => {
        resolve('some returned value')
      }, 2000)
    })
    set(somePrimitiveAtom, 'The returned value is: ' + response)
  }
  getResponse()
})

But both of the above will still set somePrimitiveAtom to the correct values.