Middleware
You've seen middleware in action in the Async Actions example. If you've used server-side libraries like Express and Koa, you were also probably already familiar with the concept of middleware. In these frameworks, middleware is some code you can put between the framework receiving a request, and the framework generating a response. For example, Express or Koa middleware may add CORS headers, logging, compression, and more. The best feature of middleware is that it's composable in a chain. You can use multiple independent third-party middleware in a single project.
Redux middleware solves different problems than Express or Koa middleware, but in a conceptually similar way. It provides a third-party extension point between dispatching an action, and the moment it reaches the reducer. People use Redux middleware for logging, crash reporting, talking to an asynchronous API, routing, and more.
This article is divided into an in-depth intro to help you grok the concept, and a few practical examples to show the power of middleware at the very end. You may find it helpful to switch back and forth between them, as you flip between feeling bored and inspired.
Understanding Middleware
While middleware can be used for a variety of things, including asynchronous API calls, it's really important that you understand where it comes from. We'll guide you through the thought process leading to middleware, by using logging and crash reporting as examples.
Problem: Logging
One of the benefits of Redux is that it makes state changes predictable and transparent. Every time an action is dispatched, the new state is computed and saved. The state cannot change by itself, it can only change as a consequence of a specific action.
Wouldn't it be nice if we logged every action that happens in the app, together with the state computed after it? When something goes wrong, we can look back at our log, and figure out which action corrupted the state.
How do we approach this with Redux?
Attempt #1: Logging Manually
The most naïve solution is just to log the action and the next state yourself every time you call store.dispatch(action)
. It's not really a solution, but just a first step towards understanding the problem.
Note
If you're using react-redux or similar bindings, you likely won't have direct access to the store instance in your components. For the next few paragraphs, just assume you pass the store down explicitly.
Say, you call this when creating a todo:
To log the action and state, you can change it to something like this:
This produces the desired effect, but you wouldn't want to do it every time.
Attempt #2: Wrapping Dispatch
You can extract logging into a function:
You can then use it everywhere instead of store.dispatch()
:
We could end this here, but it's not very convenient to import a special function every time.
Attempt #3: Monkeypatching Dispatch
What if we just replace the dispatch
function on the store instance? The Redux store is just a plain object with a few methods, and we're writing JavaScript, so we can just monkeypatch the dispatch
implementation:
This is already closer to what we want! No matter where we dispatch an action, it is guaranteed to be logged. Monkeypatching never feels right, but we can live with this for now.
Problem: Crash Reporting
What if we want to apply more than one such transformation to dispatch
?
A different useful transformation that comes to my mind is reporting JavaScript errors in production. The global window.onerror
event is not reliable because it doesn't provide stack information in some older browsers, which is crucial to understand why an error is happening.
Wouldn't it be useful if, any time an error is thrown as a result of dispatching an action, we would send it to a crash reporting service like Sentry with the stack trace, the action that caused the error, and the current state? This way it's much easier to reproduce the error in development.
However, it is important that we keep logging and crash reporting separate. Ideally we want them to be different modules, potentially in different packages. Otherwise we can't have an ecosystem of such utilities. (Hint: we're slowly getting to what middleware is!)
If logging and crash reporting are separate utilities, they might look like this:
If these functions are published as separate modules, we can later use them to patch our store:
Still, this isn't nice.
Attempt #4: Hiding Monkeypatching
Monkeypatching is a hack. “Replace any method you like”, what kind of API is that? Let's figure out the essence of it instead. Previously, our functions replaced store.dispatch
. What if they returned the new dispatch
function instead?
We could provide a helper inside Redux that would apply the actual monkeypatching as an implementation detail:
We could use it to apply multiple middleware like this:
However, it is still monkeypatching. The fact that we hide it inside the library doesn't alter this fact.
Attempt #5: Removing Monkeypatching
Why do we even overwrite dispatch
? Of course, to be able to call it later, but there's also another reason: so that every middleware can access (and call) the previously wrapped store.dispatch
:
It is essential to chaining middleware!
If applyMiddlewareByMonkeypatching
doesn't assign store.dispatch
immediately after processing the first middleware, store.dispatch
will keep pointing to the original dispatch
function. Then the second middleware will also be bound to the original dispatch
function.
But there's also a different way to enable chaining. The middleware could accept the next()
dispatch function as a parameter instead of reading it from the store
instance.
It's a “we need to go deeper” kind of moment, so it might take a while for this to make sense. The function cascade feels intimidating. ES6 arrow functions make this currying easier on eyes:
This is exactly what Redux middleware looks like.
Now middleware takes the next()
dispatch function, and returns a dispatch function, which in turn serves as next()
to the middleware to the left, and so on. It's still useful to have access to some store methods like getState()
, so store
stays available as the top-level argument.
Attempt #6: Naïvely Applying the Middleware
Instead of applyMiddlewareByMonkeypatching()
, we could write applyMiddleware()
that first obtains the final, fully wrapped dispatch()
function, and returns a copy of the store using it:
The implementation of applyMiddleware()
that ships with Redux is similar, but different in three important aspects:
It only exposes a subset of the store API to the middleware:
dispatch(action)
andgetState()
.It does a bit of trickery to make sure that if you call
store.dispatch(action)
from your middleware instead ofnext(action)
, the action will actually travel the whole middleware chain again, including the current middleware. This is useful for asynchronous middleware, as we have seen previously. There is one caveat when callingdispatch
during setup, described below.To ensure that you may only apply middleware once, it operates on
createStore()
rather than onstore
itself. Instead of(store, middlewares) => store
, its signature is(...middlewares) => (createStore) => createStore
.
Because it is cumbersome to apply functions to createStore()
before using it, createStore()
accepts an optional last argument to specify such functions.
Caveat: Dispatching During Setup
While applyMiddleware
executes and sets up your middleware, the store.dispatch
function will point to the vanilla version provided by createStore
. Dispatching would result in no other middleware being applied. If you are expecting an interaction with another middleware during setup, you will probably be disappointed. Because of this unexpected behavior, applyMiddleware
will throw an error if you try to dispatch an action before the set up completes. Instead, you should either communicate directly with that other middleware via a common object (for an API-calling middleware, this may be your API client object) or waiting until after the middleware is constructed with a callback.
The Final Approach
Given this middleware we just wrote:
Here's how to apply it to a Redux store:
That's it! Now any actions dispatched to the store instance will flow through logger
and crashReporter
:
Seven Examples
If your head boiled from reading the above section, imagine what it was like to write it. This section is meant to be a relaxation for you and me, and will help get your gears turning.
Each function below is a valid Redux middleware. They are not equally useful, but at least they are equally fun.