orchestrate-side-effects/handle-side-effects

Use this when choosing between RTK Query, createAsyncThunk, handwritten thunks, and createListenerMiddleware. Covers imperative versus reactive workflows, listener middleware setup, and keeping side e

SKILL.md

GitHub

Handle Side Effects

Setup

import {
  configureStore,
  createListenerMiddleware,
  createSlice,
} from '@reduxjs/toolkit'

const docsSlice = createSlice({
  name: 'docs',
  initialState: { status: 'idle' as 'idle' | 'saved' },
  reducers: {
    saveStarted(state) {
      state.status = 'idle'
    },
    saveFinished(state) {
      state.status = 'saved'
    },
  },
})

const listenerMiddleware = createListenerMiddleware()

export const store = configureStore({
  reducer: {
    docs: docsSlice.reducer,
  },
  middleware: (getDefaultMiddleware) =>
    getDefaultMiddleware().prepend(listenerMiddleware.middleware),
})

export type RootState = ReturnType<typeof store.getState>
export type AppDispatch = typeof store.dispatch

export const startAppListening =
  listenerMiddleware.startListening.withTypes<RootState, AppDispatch>()

Core Patterns

Use RTK Query for server cache by default

import { createApi, fetchBaseQuery } from '@reduxjs/toolkit/query/react'

type Post = { id: string; title: string }

export const api = createApi({
  reducerPath: 'api',
  baseQuery: fetchBaseQuery({ baseUrl: '/api' }),
  tagTypes: ['Post'],
  endpoints: (build) => ({
    getPosts: build.query<Post[], void>({
      query: () => 'posts',
      providesTags: ['Post'],
    }),
  }),
})

If the problem is server data that should be cached and re-used, start with RTK Query instead of a thunk.

Use createAsyncThunk for imperative workflows

import { createAsyncThunk, createSlice } from '@reduxjs/toolkit'

type Draft = { title: string }

export const draftSaved = createAsyncThunk(
  'drafts/save',
  async (draft: Draft) => {
    const response = await fetch('/api/drafts', {
      method: 'POST',
      headers: { 'Content-Type': 'application/json' },
      body: JSON.stringify(draft),
    })

    return (await response.json()) as { id: string; title: string }
  },
)

const draftsSlice = createSlice({
  name: 'drafts',
  initialState: { status: 'idle' as 'idle' | 'pending' | 'failed' },
  reducers: {},
  extraReducers: (builder) => {
    builder
      .addCase(draftSaved.pending, (state) => {
        state.status = 'pending'
      })
      .addCase(draftSaved.fulfilled, (state) => {
        state.status = 'idle'
      })
      .addCase(draftSaved.rejected, (state) => {
        state.status = 'failed'
      })
  },
})

Use a thunk when you need one imperative async workflow with dispatch and getState.

Use listener middleware for reactive workflows

import { createListenerMiddleware, createSlice } from '@reduxjs/toolkit'

const docsSlice = createSlice({
  name: 'docs',
  initialState: { status: 'idle' as 'idle' | 'saved' },
  reducers: {
    saveFinished(state) {
      state.status = 'saved'
    },
  },
})

const notificationsSlice = createSlice({
  name: 'notifications',
  initialState: [] as string[],
  reducers: {
    notificationQueued(state, action: { payload: string }) {
      state.push(action.payload)
    },
  },
})

const listenerMiddleware = createListenerMiddleware()

listenerMiddleware.startListening({
  actionCreator: docsSlice.actions.saveFinished,
  effect: async (_action, listenerApi) => {
    listenerApi.dispatch(
      notificationsSlice.actions.notificationQueued('Document saved'),
    )
  },
})

Listeners fit workflows that react to future actions or state changes over time instead of driving one imperative request from a single callsite.

Common Mistakes

CRITICAL Running side effects inside reducers

Wrong:

const todosSlice = createSlice({
  name: 'todos',
  initialState: [] as { id: string }[],
  reducers: {
    todoSaved(state, action: { payload: { id: string } }) {
      fetch('/api/todos', { method: 'POST' })
      state.push(action.payload)
    },
  },
})

Correct:

import { createAsyncThunk } from '@reduxjs/toolkit'

const todoSaved = createAsyncThunk('todos/save', async (todo: { id: string }) => {
  await fetch('/api/todos', {
    method: 'POST',
    headers: { 'Content-Type': 'application/json' },
    body: JSON.stringify(todo),
  })
  return todo
})

Reducers must stay pure even when Immer is available.

Source: reduxjs/redux:docs/style-guide/style-guide.md

HIGH Using thunks to watch future state changes

Wrong:

export const waitForSave = () => async (
  _dispatch: unknown,
  getState: () => { docs: { status: string } },
) => {
  while (getState().docs.status !== 'saved') {
    await new Promise((resolve) => setTimeout(resolve, 100))
  }
}

Correct:

startAppListening({
  predicate: (_action, currentState) => currentState.docs.status === 'saved',
  effect: async () => {
    console.log('Document saved')
  },
})

Polling inside thunks fights the architecture; listener middleware is the reactive tool.

Source: reduxjs/redux-toolkit:docs/api/createListenerMiddleware.mdx

HIGH Appending listener middleware after the default checks

Wrong:

import { configureStore, createListenerMiddleware } from '@reduxjs/toolkit'

const reducer = (state = { ready: true }) => state
const listenerMiddleware = createListenerMiddleware()

const store = configureStore({
  reducer,
  middleware: (getDefaultMiddleware) =>
    getDefaultMiddleware().concat(listenerMiddleware.middleware),
})

Correct:

import { configureStore, createListenerMiddleware } from '@reduxjs/toolkit'

const reducer = (state = { ready: true }) => state
const listenerMiddleware = createListenerMiddleware()

const store = configureStore({
  reducer,
  middleware: (getDefaultMiddleware) =>
    getDefaultMiddleware().prepend(listenerMiddleware.middleware),
})

Listener add and remove actions may carry functions, so the listener middleware needs to run before serializability checks.

Source: reduxjs/redux-toolkit:docs/api/createListenerMiddleware.mdx

References