redux-saga

What is redux-saga?

redux-saga is a library that aims to make application side effects (i.e., asynchronous things like data fetching and impure things like accessing the browser cache) easier to manage, more efficient to execute, and better at handling failures. It uses an ES6 feature called Generators to make those asynchronous flows easy to read, write, and test.

What are redux-saga's main functionalities?

Handling Asynchronous Actions

This feature allows you to handle asynchronous actions in a more readable and maintainable way. The code sample demonstrates how to fetch user data asynchronously using the `call` effect to call the API and `put` effect to dispatch actions.

function* fetchUser(action) {
  try {
    const user = yield call(Api.fetchUser, action.payload.userId);
    yield put({type: 'USER_FETCH_SUCCEEDED', user: user});
  } catch (e) {
    yield put({type: 'USER_FETCH_FAILED', message: e.message});
  }
}

function* mySaga() {
  yield takeEvery('USER_FETCH_REQUESTED', fetchUser);
}

Managing Side Effects

redux-saga helps manage side effects like delays, API calls, and more. The code sample shows how to delay an increment action by 1 second using the `delay` effect.

import { delay } from 'redux-saga/effects';

function* incrementAsync() {
  yield delay(1000);
  yield put({ type: 'INCREMENT' });
}

function* watchIncrementAsync() {
  yield takeEvery('INCREMENT_ASYNC', incrementAsync);
}

Handling Concurrency

redux-saga provides tools to handle concurrency, ensuring that only the latest action is processed. The code sample demonstrates using `takeLatest` to handle only the most recent fetch request.

import { takeLatest } from 'redux-saga/effects';

function* fetchData(action) {
  try {
    const data = yield call(Api.fetchData, action.payload);
    yield put({ type: 'FETCH_SUCCEEDED', data });
  } catch (error) {
    yield put({ type: 'FETCH_FAILED', error });
  }
}

function* mySaga() {
  yield takeLatest('FETCH_REQUESTED', fetchData);
}

Other packages similar to redux-saga

redux-thunk

redux-thunk is a middleware that allows you to write action creators that return a function instead of an action. It is simpler and more lightweight compared to redux-saga, but it doesn't offer the same level of control over complex asynchronous flows.

redux-observable

redux-observable is an RxJS-based middleware for Redux that allows you to work with async actions using Observables. It is more powerful and flexible than redux-saga for handling complex async logic, but it has a steeper learning curve due to its reliance on RxJS.

rematch

rematch is a Redux framework that abstracts away much of the boilerplate associated with Redux. It includes built-in support for side effects and async actions, making it easier to use than redux-saga, but it may not offer the same level of customization and control.

README

redux-saga

An alternative Side Effect model for Redux applications. Instead of dispatching thunks which get handled by the redux-thunk middleware. You create Sagas to gather all your Side Effects logic in a central place.

This means the logic of the application lives in 2 places

• Reducers are responsible of handling state transitions between actions

• Sagas are responsible of orchestrating complex/asynchronous operations.

Sagas are created using Generator functions.

This middleware is not only about handling asynchronous flow. If all what matters is simplifying asynchronous control flow, one could simply use async/await functions with some promise middleware.

What this middleware proposes is

• A composable abstraction Effect: waiting for an action, triggering State updates (by dispatching actions to the store), calling a remote service are all different forms of Effects. A Saga composes those Effects using familiar control flow constructs (if, while, for, try/catch).

• The Saga is itself an Effect. It can be combined with other Effects using combinators. It can also be called from inside other Sagas, providing the full power of Subroutines and Structured Programming

• Effects may be yielded declaratively. You yield a description of the Effect which will be executed by the middleware. This makes your operational logic inside Generators fully testable.

• You can implement complex operations with logic that spans across multiple actions (e.g. User onBoarding, Wizard dialogs, complex Game rules ...), which are not trivial to express using other effects middlewares.

• Getting started

• How is this different from other asynchronous middlewares

• Declarative Effects

• Error handling

• Effect Combinators

• Sequencing Sagas via yield*

• Composing Sagas

• Non blocking calls with fork/join

• Task cancellation

• Building examples from sources

• Using umd build in the browser

#Getting started

Install

npm install redux-saga

Create the Saga (using the counter example from Redux)

import { take, put } from 'redux-saga'
// sagas/index.js

function* incrementAsync() {

  while(true) {

    // wait for each INCREMENT_ASYNC action  
    const nextAction = yield take(INCREMENT_ASYNC)

    // delay is a sample function
    // return a Promise that resolves after (ms) milliseconds
    yield delay(1000)

    // dispatch INCREMENT_COUNTER
    yield put( increment() )
  }

}

export default [incrementAsync]

Plug redux-saga in the middleware pipeline

// store/configureStore.js
import sagaMiddleware from 'redux-saga'
import sagas from '../sagas'

const createStoreWithSaga = applyMiddleware(
  // ...,
  sagaMiddleware(...sagas)
)(createStore)

export default function configureStore(initialState) {
  return createStoreWithSaga(reducer, initialState)
}

#How is this different from the others

In the previous example we created an incrementAsync Saga. The call yield take(action) is a typical illustration of how Sagas work.

Typically, actual middlewares handle some Effect form triggered by an Action Creator. For example, redux-thunk handles thunks by calling them with (getState, dispatch) as arguments, redux-promise handles Promises by dispatching their resolved values. redux-gen handles generators by dispatching all yielded actions to the store. The common thing that all those middlewares share is the same 'call on each action' pattern. They will be called again and again each time an action happens, i.e. they are scoped by the root action that triggered them.

Sagas work differently, they are not fired from within Action Creators but are started with your application and choose what user actions to watch. They are like daemon tasks that run in the background and choose their own logic of progression. In the example above, incrementAsync pulls the INCREMENT_ASYNC action using yield take(...). This is a blocking call, which means the Saga will not progress until it receives a matching action.

After receiving the queried action, the Saga triggers a call to delay(1000), which in our example returns a Promise that will be resolved after 1 second. Again, this is a blocking call, so the Saga will wait for 1 second before continuing on (a better way is call(delay, 1000), as we'll see in the section on declarative Effects).

After the delay, the Saga dispatches an INCREMENT_COUNTER action using the put(action) function. Here also, the Saga will wait for the dispatch result. If the dispatch call returns a normal value, the Saga resumes immediately (asap), but if the result value is a Promise then the Saga will wait until the Promise is resolved (or rejected).

To generalize, waiting for a future action (yield take(MY_ACTION)), waiting for the future result of a function call (yield delay(1000)) or waiting for the result of a dispatch (yield put(myAction())) all are the same concept. In all cases, we are yielding some form of side effects.

Note also how incrementAsync uses an infinite loop while(true) which means it will stay alive for all the application lifetime. You can also create Sagas that last only for a limited amount of time. For example, the following Saga waits for the first 3 INCREMENT_COUNTER actions, triggers a showCongratulation() action and then finishes.

function* onBoarding() {

  for(let i = 0; i < 3; i++)
    yield take(INCREMENT_COUNTER)

  yield put( showCongratulation() )
}

#Declarative Effects

Sagas Generators can yield Effects in multiple forms. The simplest way is to yield a Promise

function* fetchSaga() {

  // fetch is a sample function
  // returns a Promise that will resolve with the GET response
  const products = yield fetch('/products')

  // dispatch a RECEIVE_PRODUCTS action
  yield put( receiveProducts(products) )
}

In the example above, fetch('/products') returns a Promise that will resolve with the GET response. So the 'fetch effect' will be executed immediately . Simple and idiomatic but ...

Suppose we want to test generator above

const iterator = fetchSaga()
assert.deepEqual( iterator.next().value, ?? ) // what do we expect ?

We want to check the result of the first value yielded by the generator, which is in our case the result of running fetch('/products'). Executing the real service during tests is not a viable nor a practical approach, so we have to mock the fetch service, i.e. we'll have to replace the real fetch method with a fake one which doesn't actually run the GET request but only checks that we've called fetch with the right arguments ('/products' in our case).

Mocks make testing more difficult and less reliable. On the other hand, functions that simply return values are easier to test, we can use a simple equal() to check the result.This is the way to write the most reliable tests.

Not convinced ? I encourage you to read this [Eric Elliott' article] (https://medium.com/javascript-scene/what-every-unit-test-needs-f6cd34d9836d#.4ttnnzpgc)

(...)equal(), by nature answers the two most important questions every unit test must answer, but most don’t:

• What is the actual output?
• What is the expected output?

If you finish a test without answering those two questions, you don’t have a real unit test. You have a sloppy, half-baked test.

What we need actually, is just to make sure the fetchSaga yields a call with the right function and the right arguments. For this reason, the library provides some declarative ways to yield Side Effects while still making it easy to test the Saga logic

import { call } from 'redux-saga'

function* fetchSaga() {
  const products = yield call( fetch, '/products' ) // don't run the effect
}

We're using now call(fn, ...args) function. The difference from the precedent example is that now we're not executing the fetch call immediately, instead, call creates a description of the effect. Just as in Redux you use action creators to create a plain object describing the action that will get executed by the Store, call creates a plain object describing the function call. The redux-saga middleware takes care of executing the function call and resuming the generator with the resolved response.

This allows us to easily test the Generator outside the Redux environment.

import { call } from 'redux-saga'

const iterator = fetchSaga()
assert.deepEqual(iterator.next().value, call(fetch, '/products')) // expects a call(...) value

Now, we don't need to mock anything, a simple equality test will suffice.

The advantage of declarative effects is that we can test all the logic inside a Saga/Generator by simply iterating over the resulting iterator and doing a simple equality tests on the values yielded successively. This is a real benefit, as your complex asynchronous operations are no longer black boxes, you can test in detail their logic of operation no matter how complex it is.

The call method is well suited for functions that return Promise results. Another function cps can be used to handle Node style functions (e.g. fn(...args, callback) where callback is of the form (error, result) => ()). For example

import { cps } from 'redux-saga'

const content = yield cps(readFile, '/path/to/file')

and of course you can test it just like you test call

import { cps } from 'redux-saga'

const iterator = fetchSaga()
assert.deepEqual(iterator.next().value, cps(readFile, '/path/to/file') )

#Error handling

You can catch errors inside the Generator using the simple try/catch syntax. In the following example, the Saga catch errors from the api.buyProducts call (i.e. a rejected Promise)

function* checkout(getState) {

  while( yield take(types.CHECKOUT_REQUEST) ) {
    try {
      const cart = getState().cart
      yield call(api.buyProducts, cart)
      yield put(actions.checkoutSuccess(cart))
    } catch(error) {
      yield put(actions.checkoutFailure(error))
    }
  }
}

Of course you're not forced to handle you API errors inside try/catch blocks, you can also make your API service return a normal value with some error flag on it

function buyProducts(cart) {
  return doPost(...)
    .then(result => {result})
    .catch(error => {error})
}

function* checkout(getState) {
  while( yield take(types.CHECKOUT_REQUEST) ) {
    const cart = getState().cart
    const {result, error} = yield call(api.buyProducts, cart)
    if(!error)
      yield put(actions.checkoutSuccess(result))
    else
      yield put(actions.checkoutFailure(error))
  }
}

#Effect Combinators

The yield statements are great for representing asynchronous control flow in a simple and linear style. But we also need to do things in parallel. We can't simply write

// Wrong, effects will be executed in sequence
const users  = yield call(fetch, '/users'),
      repose = yield call(fetch, '/repose')

Because the 2nd effect will not get executed until the first call resolves. Instead we have to write

import { call } from 'redux-saga'

// correct, effects will get executed in parallel
const [users, repose]  = yield [
  call(fetch, '/users'),
  call(fetch, '/repose')
]

When we yield an array of effects, the generator is blocked until all the effects are resolved (or as soon as one is rejected, just like how Promise.all behaves).

Sometimes we start multiple tasks in parallel but we don't want to wait for all of them, we just need to get the winner: the first one that resolves (or rejects). The race function offers a way of triggering a race between multiple effects.

The following sample shows a Saga that triggers a remote fetch request, and constrain the response with a 1 second timeout.

import { race, take, put } from 'redux-saga'

function* fetchPostsWithTimeout() {
  while( yield take(FETCH_POSTS) ) {
    // starts a race between 2 effects
    const {posts, timeout} = yield race({
      posts   : call(fetchApi, '/posts'),
      timeout : call(delay, 1000)
    })

    if(posts)
      put( actions.receivePosts(posts) )
    else
      put( actions.timeoutError() )
  }
}

#Sequencing Sagas via yield*

You can use the builtin yield* operator to compose multiple sagas in a sequential way. This allows you to sequence your macro-tasks in a simple procedural style.

function* playLevelOne(getState) { ... }

function* playLevelTwo(getState) { ... }

function* playLevelThree(getState) { ... }

function* game(getState) {

  const score1 = yield* playLevelOne(getState)
  put(showScore(score1))

  const score2 = yield* playLevelTwo(getState)
  put(showScore(score2))

  const score3 = yield* playLevelThree(getState)
  put(showScore(score3))

}

Note that using yield* will cause the JavaScript runtime to spread the whole sequence. The resulting iterator (from game()) will yield all values from the nested iterators. A more powerful alternative is to use the more generic middleware composition mechanism.

#Composing Sagas

While using yield* provides an idiomatic way of composing Sagas. This approach has some limits:

• You'll likely want to test nested generators separately. This leads to some duplication in the test code as well as an overhead of the duplicated execution. We don't want to execute a nested generator but only make sure the call to it was issued with the right argument.

• More importantly, yield* allows only for sequential composition of tasks, you can only yield* to one generator at a time.

You can simply use yield to start one or more subtasks in parallel. When yielding a call to a generator, the Saga will wait for the generator to terminate before progressing, then resumes with the returned value (or throws if an error propagates from the subtask).

function* fetchPosts() {
  yield put( actions.requestPosts() )
  const products = yield call(fetchApi, '/products')
  yield put( actions.receivePosts(products) )
}

function* watchFetch() {
  while ( yield take(FETCH_POSTS) ) {
    yield call(fetchPosts) // waits for the fetchPosts task to terminate
  }
}

Yielding to an array of nested generators will start all the sub-generators in parallel and wait for them to finish. Then resume with all the results

function* mainSaga(getState) {
  const results = yield [ call(task1), call(task2), ...]
  yield put( showResults(results) )
}

In fact, yielding Sagas is no more different than yielding other effects (future actions, timeouts ...). It means you can combine those Sagas with all the other types using the effect combinators.

For example you may want the user finish some game in a limited amount of time

function* game(getState) {

  let finished
  while(!finished) {
    // has to finish in 60 seconds
    const {score, timeout}  = yield race({
      score  : call( play, getState),
      timeout : call(delay, 60000)
    })

    if(!timeout) {
      finished = true
      yield put( showScore(score) )
    }
  }

}

#Non blocking calls with fork/join

the yield statement causes the generator to pause until the yielded effect resolves or rejects. If you look closely at this example

function* watchFetch() {
  while ( yield take(FETCH_POSTS) ) {
    yield put( actions.requestPosts() )
    const posts = yield call(fetchApi, '/posts') // blocking call
    yield put( actions.receivePosts(posts) )
  }
}

the watchFetch generator will wait until yield call(fetchApi, '/posts') terminates. Imagine that the FETCH_POSTS action is fired from a Refresh button. If our application disables the button between each fetch (no concurrent fetches) then there is no issue, because we know that no FETCH_POSTS action will occur until we get the response from the fetchApi call.

But what happens if the application allows the user to click on Refresh without waiting for the current request to terminate ?

The following example illustrates a possible sequence of the events

UI                              watchFetch
--------------------------------------------------------
FETCH_POSTS.....................call fetchApi........... waiting to resolve
........................................................
........................................................                     
FETCH_POSTS............................................. missed
........................................................
FETCH_POSTS............................................. missed
................................fetchApi returned.......
........................................................

When watchFetch is blocked on the fetchApi call, all FETCH_POSTS occurring in between the call and the response are missed.

To express non blocking calls, we can use the fork function. A possible rewrite of the previous example with fork can be

import { fork, call, take, put } from 'redux-saga'

function* fetchPosts() {
  yield put( actions.requestPosts() )
  const posts = yield call(fetchApi, '/posts')
  yield put( actions.receivePosts(posts) )
}

function* watchFetch() {
  while ( yield take(FETCH_POSTS) ) {
    yield fork(fetchPosts) // non blocking call
  }
}

fork accepts function/generator calls as well as simple effects

yield fork(func, ...args)       // simple async functions (...) -> Promise
yield fork(generator, ...args)  // Generator functions
yield fork( put(someActions) )  // Simple effects

The result of yield fork(api) is a Task descriptor. To get the result of a forked Task in a later time, we use the join function

import { fork, join } from 'redux-saga'

function* child() { ... }

function *parent() {
  // non blocking call
  const task = yield fork(subtask, ...args)

  // ... later
  // now a blocking call, will resume with the outcome of task
  const result = yield join(task)

}

the task object exposes some useful methods

method	return value
task.isRunning()	true if the task hasn't yet returned or throwed an error
task.result()	task return value. `undefined` if task is still running
task.error()	task thrown error. `undefined` if task is still running
task.done	a Promise which is either resolved with task's return value rejected with task's thrown error

#Task cancellation

Once a task is forked, you can abort its execution using yield cancel(task). Cancelling a running task will throw a SagaCancellationException inside it.

To see how it works, let's consider a simple example. A background sync which can be started/stopped by some UI commands. Upon receiving a START_BACKGROUND_SYNC action, we fork a background task that will periodically sync some data from a remote server.

The task will execute continually until a STOP_BACKGROUND_SYNC action is triggered. Then we cancel the background task and wait again for the next START_BACKGROUND_SYNC action.

import { take, put, call, fork, cancel, SagaCancellationException } from 'redux-saga'
import actions from 'somewhere'
import { someApi, delay } from 'somewhere'

function* bgSync() {
  try {
    while(true) {
      yield put(actions.requestStart())
      const result = yield call(someApi)
      yield put(actions.requestSuccess(result))
      yield call(delay, 5000)
    }
  } catch(error) {
    if(error instanceof SagaCancellationException)
      yield put(actions.requestFailure('Sync cancelled!'))
  }
}

function* main() {
  while( yield take(START_BACKGROUND_SYNC) ) {
    // starts the task in the background
    const bgSyncTask = yield fork(bgSync)

    // wait for the user stop action
    yield take(STOP_BACKGROUND_SYNC)
    // user clicked stop. cancel the background task
    // this will throw a SagaCancellationException into task
    yield cancel(bgSyncTask)
  }
}

yield cancel(bgSyncTask) will throw a SagaCancellationException inside the currently running task. In the above example, the exception is caught by bgSync. Otherwise, it will propagate up to main. And it if main doesn't handle it then it will bubble up the call chain, just as normal JavaScript errors bubble up the call chain of synchronous functions.

Cancelling a running task will also cancel the current effect where the task is blocked at the moment of cancellation.

For example, suppose that at a certain point in application lifetime, we had this pending call chain

function* main() {
  const task = yield fork(subtask)
  ...
  // later
  yield cancel(task)
}

function* subtask() {
  ...
  yield call(subtask2) // currently blocked on this call
  ...
}

function* subtask2() {
  ...
  yield call(someApi) // currently blocked on this all
  ...
}

yield cancel(task) will trigger a cancellation on subtask, which in turn will trigger a cancellation on subtask2. A SagaCancellationException will be thrown inside subtask2, then another SagaCancellationException will be thrown inside subtask. If subtask omits to handle the cancellation exception, it will propagate up to main.

The main purpose of the cancellation exception is to allow cancelled tasks to perform any cleanup logic. So we wont leave the application in an inconsistent state. In the above example of background sync, by catching the cancellation exception, bgSync is able to dispatch a requestFailure action to the store. Otherwise, the store could be left in a inconsistent state (e.g. waiting for the result of a pending request)

It's important to remember that yield cancel(task) doesn't wait for the cancelled task to finish (i.e. to perform its catch block). The cancel effect behave like fork. It returns as soon as the cancel was initiated. Once cancelled, a task should normally return as soon as it finishes its cleanup logic. In some cases, the cleanup logic could involve some async operations, but the cancelled task lives now as a separate process, and there is no way for it to rejoin the main control flow (except dispatching actions other tasks via the Redux store. However this will lead to complicated control flows that ae hard to reason about. It's always preferable to terminate a cancelled task asap).

##Automatic cancellation

Besides manual cancellation. There are cases where cancellation is triggered automatically

1- In a race effect. All race competitors, except the winner, are automatically cancelled.

2- In a parallel effect (yield [...]). The parallel effect is rejected as soon as one of the sub-effects is rejected (as implied by Promise.all). In this case, all the other sub-effects are automatically cancelled.

Unlike in manual cancellations, unhandled cancellation exceptions are not propagated to the actual saga running the race/parallel effect. Nevertheless, a warning is logged into the console in case a cancelled task omitted to handle a cancellation exception.

#Building examples from sources

Pre-requisites

• browserify
• budo to serve with live-reload npm i -g budo

You can also build the examples manually, and open index.html at the root of each example directory to run.

git clone https://github.com/yelouafi/redux-saga.git
cd redux-saga
npm install
npm test

Below the examples ported (so far) from the Redux repos

Counter example

// run with live-reload server
npm run counter

// manual build
npm run build-counter

// test sample for the generator
npm run test-counter

Shopping Cart example

// run with live-reload server
npm run shop

// manual build
npm run build-shop

// test sample for the generator
npm run test-shop

async example

// run with live-reload server
npm run async

// manual build
npm run build-async

//sorry, no tests yet

real-world example (with webpack hot reloading)

cd examples/real-world
npm install
npm start

#Using umd build in the browser

There's an umd build of redux-saga available in dist/ folder. Using the umd build redux-saga is available as ReduxSaga in the window object. The umd version is useful if you don't use webpack or browserify, you can access it directly from npmcdn. The following builds are available: https://npmcdn.com/redux-saga/dist/redux-saga.js https://npmcdn.com/redux-saga/dist/redux-saga.min.js

Important! If the browser you are targeting doesn't support es2015 generators you must provide a valid polyfill, for example the one provided by babel: browser-polyfill.min.js. The polyfill must be imported before redux-saga.