• 周日. 7月 3rd, 2022




RxJs的switchMap,mergeMap,concatMap (and exhaustMap)


11月 28, 2021

大家还是看原文吧:RxJs Mapping: switchMap vs mergeMap vs concatMap vs exhaustMap (angular-university.io)


我们日常发现的一些最常用的 RxJs 操作符是 RxJs 高阶映射操作符:switchMap、mergeMap、concatMap 和exhaustMap。





这样做是有原因的:为了理解这些操作符,我们首先需要了解每个内部使用的 Observable 组合策略。

与其试图自己理解switchMap,不如先了解什么是Observable切换; 我们需要先学习 Observable 连接等,而不是直接深入 concatMap。

这就是我们在这篇文章中要做的事情,我们将按逻辑顺序学习 concat、merge、switch 和exhaust 策略及其对应的映射运算符:concatMap、mergeMap、switchMap 和exhaustMap。



Table of Contents

In this post, we will cover the following topics:

  • The RxJs Map Operator
  • What is higher-order Observable Mapping
  • Observable Concatenation
  • The RxJs concatMap Operator
  • Observable Merging
  • The RxJs mergeMap Operator
  • Observable Switching
  • The RxJs switchMap Operator
  • The Exhaust strategy
  • The RxJs exhaustMap Operator
  • How to choose the right mapping Operator?
  • Running GitHub repo (with code samples)
  • Conclusions

Note that this post is part of our ongoing RxJs Series. So without further ado, let’s get started with our RxJs mapping operators deep dive!

基本 Map Operator 的工作原理


正如运算符的名称所暗示的那样,他们正在做某种映射:但究竟是什么被映射了? 我们先来看看 RxJs Map 操作符的弹珠图:

RxJs Map Operator marble diagram

基本 Map Operator 的工作原理

使用 map 运算符,我们可以获取输入流(值为 1、2、3),并从中创建派生的映射输出流(值为 10、20、30)。

底部输出流的值是通过获取输入流的值并将它们应用到一个函数来获得的:这个函数只是将这些值乘以 10。

所以 map 操作符就是映射输入 observable 的值。 以下是我们如何使用它来处理 HTTP 请求的示例:

const http$ : Observable<Course[]> = this.http.get('/api/courses');

        tap(() => console.log('HTTP request executed')),
        map(res => Object.values(res['payload']))
        courses => console.log("courses", courses)

在这个例子中,我们正在创建一个 HTTP observable 来进行后端调用,我们正在订阅它。 observable 将发出后端 HTTP 响应的值,这是一个 JSON 对象。

在这种情况下,HTTP 响应将数据包装在有效负载属性中,因此为了获取数据,我们应用了 RxJs 映射运算符。 然后映射函数将映射 JSON 响应负载并提取负载属性的值。



在高阶映射中,我们不是将像 1 这样的普通值映射到另一个像 10 这样的值,而是将一个值映射到一个 Observable 中!

结果是一个高阶的 Observable。 它和其他任何一个 Observable 一样只是一个 Observable,但它的值本身也是 Observables,我们可以单独订阅。

这听起来可能有些牵强,但实际上,这种类型的映射一直在发生。 让我们举一个这种类型映射的实际例子。 比方说,例如,我们有一个 Angular Reactive Form,它通过 Observable 随时间发出有效的表单值:

    selector: 'course-dialog',
    templateUrl: './course-dialog.component.html'
export class CourseDialogComponent implements AfterViewInit {

    form: FormGroup;

    @ViewChild('saveButton') saveButton: ElementRef;

        private fb: FormBuilder,
        private dialogRef: MatDialogRef<CourseDialogComponent>,
        @Inject(MAT_DIALOG_DATA) course:Course) {

        this.course = course;

        this.form = fb.group({
            description: [course.description, 
            category: [course.category, Validators.required],
            releasedAt: [moment(), Validators.required],
            longDescription: [course.longDescription,

Reactive Form 提供了一个 Observable this.form.valueChanges,它在用户与表单交互时发出最新的表单值。这将是我们的源 Observable。

我们想要做的是在这些值随着时间的推移发出时至少保存其中一些值,以实现表单草稿预保存功能。这样,随着用户填写表单,数据会逐渐保存,从而避免由于意外重新加载而丢失整个表单数据。Why Higher-Order Observables?

In order to implement the form draft save functionality, we need to take the form value, and then create a second HTTP observable that performs a backend save, and then subscribe to it.

We could try to do all of this manually, but then we would fall in the nested subscribes anti-pattern:

  formValue => {
  const httpPost$ =
  this.http.put(`/api/course/${courseId}`, formValue);
  res => … handle successful save …
  err => … handle save error …
view raw02.ts hosted with ❤ by GitHub

As we can see, this would cause our code to nest at multiple levels quite quickly, which was one of the problems that we were trying to avoid while using RxJs in the first place.

Let’s call this new httpPost$ Observable the inner Observable, as it was created in an inner nested code block.

Avoiding nested subscriptions

We would like to do all this process in a much more convenient way: we would like to take the form value, and map it into a save Observable. And this would effectively create a higher-order Observable, where each value corresponds to a save request.

We want to then transparently subscribe to each of these network Observables, and directly receive the network response all in one go, to avoid any nesting.

And we could do all this if we would have available some sort of a higher order RxJs mapping operator! Why do we need four different operators then?

To understand that, imagine what happens if multiple form values are emitted by the valueChanges observable in quick succession and the save operation takes some time to complete:

  • should we wait for one save request to complete before doing another save?
  • should we do multiple saves in parallel?
  • should we cancel an ongoing save and start a new one?
  • should we ignore new save attempts while one is already ongoing?

Before exploring each one of these use cases, let’s go back to the nested subscribes code above.

In the nested subscribes example, we are actually triggering the save operations in parallel, which is not what we want because there is no strong guarantee that the backend will handle the saves sequentially and that the last valid form value is indeed the one stored on the backend.

Let’s see what it would take to ensure that a save request is done only after the previous save is completed.

Understanding Observable Concatenation

In order to implement sequential saves, we are going to introduce the new notion of Observable concatenation. In this code example, we are concatenating two example observables using the concat() RxJs function:

  const series1$ = of(‘a’, ‘b’);
  const series2$ = of(‘x’, ‘y’);
  const result$ = concat(series1$, series2$);
view raw03.ts hosted with ❤ by GitHub

After creating two Observables series1$ and series2$ using the of creation function, we have then created a third result$ Observable, which is the result of concatenating series1$ and series2$.

Here is the console output of this program, showing the values emitted by the result Observable:


As we can see, the values are the result of concatenating the values of series1$ with series2$ together. But here is the catch: this only works because these Observables are completing!!

The of() function will create Observables that emit values passed to of() and then it will complete the Observables after all values are emitted.

Observable Concatenation Marble Diagram

To really understand what is going on, we need to look at the Observable concatenation marble diagram:

RxJs Map Operator marble diagram

Do you notice the vertical bar after the value b on the first Observable? That marks the point in time when the first Observable with values a and b (series1$) is completed.

Let’s break down what is going on here by following step-by-step the timeline:

  • the two Observables series1$ and series2$ are passed to the concat() function
  • concat() will then subscribe to the first Observable series1$, but not to the second Observable series2$ (this is critical to understand concatenation)
  • source1$ emits the value a, which gets immediately reflected in the output result$ Observable
  • note that the source2$ Observable is not yet emitting values, because it has not yet been subscribed to
  • source1$ will then emit the b value, which gets reflected in the output
  • source1$ will then complete, and only after that will concat() now subscribe to source2$
  • the source2$ values will then start getting reflected in the output, until source2$ completes
  • when source2$ completes, then the result$ Observable will also complete
  • note that we can pass to concat() as many Observables as we want, and not only two like in this example

The key point about Observable Concatenation

As we can see, Observable concatenation is all about Observable completion! We take the first Observable and use its values, wait for it to complete and then we use the next Observable, etc. until all Observables complete.

Going back to our higher-order Observable mapping example, let’s see how the notion of concatenation can help us.

Using Observable Concatenation to implement sequential saves

As we have seen, in order to make sure that our form values are saved sequentially, we need to take each form value and map it to an httpPost$ Observable.

We then need to subscribe to it, but we want the save to complete before subscribing to the next httpPost$ Observable.

In order to ensure sequentiality, we need to concatenate the multiple httpPost$ Observables together!

We will then subscribe to each httpPost$ and handle the result of each request sequentially. In the end, what we need is an operator that does a mixture of:

  • a higher-order mapping operation (taking the form value and turning it into an httpPost$ Observable)
  • with a concat() operation, concatenating the multiple httpPost$ Observables together to ensure that an HTTP save is not made before the previous ongoing save completes first

What we need is the aptly named RxJs concatMap Operator, which does this mixture of higher order mapping with Observable concatenation.

The RxJs concatMap Operator

Here is what our code looks like if we now use the concatMap Operator:

  concatMap(formValue => this.http.put(`/api/course/${courseId}`,
  saveResult => … handle successful save …,
  err => … handle save error …
view raw04.ts hosted with ❤ by GitHub

As we can see, the first benefit of using a higher-order mapping operator like concatMap is that now we no longer have nested subscribes.

By using concatMap, now all form values are going to be sent to the backend sequentially, as shown here in the Chrome DevTools Network tab:

RxJs concatMap Example

Breaking down the concatMap network log diagram

As we can see, one save HTTP request starts only after the previous save has completed. Here is how the concatMap operator is ensuring that the requests always happen in sequence:

  • concatMap is taking each form value and transforming it into a save HTTP Observable, called an inner Observable
  • concatMap then subscribes to the inner Observable and sends its output to the result Observable
  • a second form value might come faster than what it takes to save the previous form value in the backend
  • If that happens, that new form value will not be immediately mapped to an HTTP request
  • instead, concatMap will wait for previous HTTP Observable to complete before mapping the new value to an HTTP Observable, subscribing to it and therefore triggering the next save

Notice that the code here is just the basis of an implementation to save draft form values. You can combine this with other operators to for example save only valid form values, and throttle the saves to make sure that they don’t occur too frequently.

Observable Merging

Applying Observable concatenation to a series of HTTP save operations seems like a good way to ensure that the saves happen in the intended order.

But there are other situations where we would like to instead run things in parallel, without waiting for the previous inner Observable to complete.

And for that, we have the merge Observable combination strategy! Merge, unlike concat, will not wait for an Observable to complete before subscribing to the next Observable.

Instead, merge subscribes to every merged Observable at the same time, and then it outputs the values of each source Observable to the combined result Observable as the multiple values arrive over time.

Practical Merge Example

To make it clear that merging does not rely on completion, let’s merge two Observables that never complete, as these are interval Observables:

  const series1$ = interval(1000).pipe(map(val => val*10));
  const series2$ = interval(1000).pipe(map(val => val*100));
  const result$ = merge(series1$, series2$);
view raw05.ts hosted with ❤ by GitHub

The Observables created with interval() will emit the values 0, 1, 2, etc. at a one second interval and will never complete.

Notice that we are applying a couple of map operator to these interval Observables, just to make it easier to distinguish them in the console output.

Here are the first few values visible in the console:


Merging and Observable Completion

As we can see, the values of the merged source Observables show up in the result Observable immediately as they are emitted. If one of the merged Observables completes, merge will continue to emit the values of the other Observables as they arrive over time.

Notice that if the source Observables do complete, merge will still work in the same way.

The Merge Marble Diagram

Let’s take another merge example, depicted in the following marble diagram:

RxJs merge Example

As we can see, the values of the merged source Observables show up immediately in the output. The result Observable will not be completed until all the merged Observables are completed.

Now that we understand the merge strategy, let’s see how it how it can be used in the context of higher-order Observable mapping.

The RxJs mergeMap Operator

If we combine the merge strategy with the notion of higher-order Observable mapping, we get the RxJs mergeMap Operator. Let’s have a look at the marble diagram for this operator:

RxJs mergeMap Example

Here is how the mergeMap operator works:

  • each value of the source Observable is still being mapped into an inner Observable, just like the case of concatMap
  • Like concatMap, that inner Observable is also subscribed to by mergeMap
  • as the inner Observables emit new values, they are immediately reflected in the output Observable
  • but unlike concatMap, in the case of mergeMap we don’t have to wait for the previous inner Observable to complete before triggering the next innner Observable
  • this means that with mergeMap (unlike concatMap) we can have multiple inner Observables overlapping over time, emitting values in parallel like we see highlighted in red in the picture

Checking the mergeMap Network Log

Going back to our previous form draft save example, its clear that what we need concatMap in that case and not mergeMap, because we don’t want the saves to happen in parallel.

Let’s see what happens if we would accidentally choose mergeMap instead:

  mergeMap(formValue =>
  saveResult => … handle successful save …,
  err => … handle save error …
view raw06.ts hosted with ❤ by GitHub

Let’s now say that the user interacts with the form and starts inputting data rather quickly. In that case, we would now see multiple save requests running in parallel in the network log:

RxJs mergeMap Example

As we can see, the requests are happening in parallel, which in this case is an error! Under heavy load, it’s possible that these requests would be processed out of order.

Observable Switching

Let’s now talk about another Observable combination strategy: switching. The notion of switching is closer to merging than to concatenation, in the sense that we don’t wait for any Observable to terminate.

But in switching, unlike merging, if a new Observable starts emitting values we are then going to unsubscribe from the previous Observable, before subscribing to the new Observable.

Observable switching is all about ensuring that the unsubscription logic of unused Observables gets triggered, so that resources can be released!

Switch Marble Diagram

Let’s have a look at the marble diagram for switching:

RxJs switch Example

Notice the diagonal lines, these are not accidental! In the case of the switch strategy, it was important to represent the higher-order Observable in the diagram, which is the top line of the image.

This higher-order Observable emits values which are themselves Observables.

The moment that a diagonal line forks from the higher-order Observable top line, is the moment when a value Observable was emitted and subscribed to by switch.

Breaking down the switch Marble Diagram

Here is what is going on in this diagram:

  • the higher-order Observable emits its first inner Observable (a-b-c-d), that gets subscribed to (by the switch strategy implementation)
  • the first inner Observable (a-b-c-d) emits values a and b, that get immediately reflected in the output
  • but then the second inner Observable (e-f-g) gets emitted, which triggers the unsubscription from the first inner Observable (a-b-c-d), and this is the key part of switching
  • the second inner Observable (e-f-g) then starts emitting new values, that get reflected in the output
  • but notice that the first inner Observable (a-b-c-d) is meanwhile still emitting the new values c and d
  • these later values, however, are not reflected in the output, and that is because we had meanwhile unsubscribed from the first inner Observable (a-b-c-d)

We can now understand why the diagram had to be drawn in this unusual way, with diagonal lines: its because we need to represent visually when each inner Observable gets subscribed (or unsubscribed) from, which happens at the points the diagonal lines fork from the source higher-order Observable.

The RxJs switchMap Operator

Let’s then take the switch strategy and apply it to higher order mapping. Let’s say that we have a plain input stream that is emitting the values 1, 3 and 5.

We are then going to map each value to an Observable, just like we did in the cases of concatMap and mergeMap and obtain a higher-order Observable.

If we now switch between the emitted inner Observables, instead of concatenating them or merging them, we end up with the switchMap Operator:

RxJs switchMap Example

Breaking down the switchMap Marble Diagram

Here is how this operator works:

  • the source observable emits values 1, 3 and 5
  • these values are then turned into Observables by applying a mapping function
  • the mapped inner Observables get subscribed to by switchMap
  • when the inner Observables emit a value, the value gets immediately reflected in the output
  • but if a new value like 5 gets emitted before the previous Observable got a chance to complete, the previous inner Observable (30-30-30) will be unsubscribed from, and its values will no longer be reflected in the output
  • notice the 30-30-30 inner Observable in red in the diagram above: the last 30 value was not emitted because the 30-30-30 inner Observable got unsubscribed from

So as we can see, Observable switching is all about making sure that we trigger that unsubscription logic from unused Observables. Let’s now see switchMap in action!

Search TypeAhead – switchMap Operator Example

A very common use case for switchMap is a search Typeahead. First let’s define our source Observable, whose values are themselves going to trigger search requests.

This source Observable is going to emit values which are the search text that the user types in an input:

  const searchText$: Observable<string> =
  fromEvent<any>(this.input.nativeElement, ‘keyup’)
  map(event => event.target.value),
view raw07.ts hosted with ❤ by GitHub

This source Observable is linked to an input text field where the user types its search. As the user types the words “Hello World” as a search, these are the values emitted by searchText$:

Hello W
Hello W
Hello Wo
Hello Wor
Hello Worl
Hello World

Debouncing and removing duplicates from a Typeahead

Notice the duplicate values, either caused by the use of the space between the two words, or the use of the Shift key for capitalizing the letters H and W.

In order to avoid sending all these values as separate search requests to the backend, let’s wait for the user input to stabilize by using the debounceTime operator:

  const searchText$: Observable<string> =
  fromEvent<any>(this.input.nativeElement, ‘keyup’)
  map(event => event.target.value),
view raw08.ts hosted with ❤ by GitHub

With the use of this operator, if the user types at a normal speed, we now have only one value in the output of searchText$:

Hello World

This is already much better than what we had before, now a value will only be emitted if its stable for at least 400ms!

But if the user types slowly as he is thinking about the search, to the point that it takes more than 400 ms between two values, then the search stream could look like this:

Hello World

Also, the user could type a value, hit backspace and type it again, which might lead to duplicate search values. We can prevent the occurrence of duplicate searches by adding the distinctUntilChanged operator.

Cancelling obsolete searches in a Typeahead

But more than that, we would need a way to cancel previous searches, as a new search get’s started.

What we want to do here is to transform each search string into a backend search request and subscribe to it, and apply the switch strategy between two consecutive search requests, causing the previous search to be canceled if a new search gets triggered.

And that is exactly what the switchMap operator will do! Here is the final implementation of our Typeahead logic that uses it:

  const searchText$: Observable<string> =
  fromEvent<any>(this.input.nativeElement, ‘keyup’)
  map(event => event.target.value),
  const lessons$: Observable<Lesson[]> = searchText$
  switchMap(search => this.loadLessons(search))
  function loadLessons(search:string): Observable<Lesson[]> {
  const params = new HttpParams().set(‘search’, search);
  return this.http.get(`/api/lessons/${coursesId}`, {params});
view raw09.ts hosted with ❤ by GitHub

switchMap Demo with a Typeahead

Let’s now see the switchMap operator in action! If the user types on the search bar, and then hesitates and types something else, here is what we can typically see in the network log:

RxJs switchMap Example

As we can see, several of the previous searches have been canceled as they where ongoing, which is awesome because that will release server resources that can then be used for other things.

The Exhaust Strategy

The switchMap operator is ideal for the typeahead scenario, but there are other situations where what we want to do is to ignore new values in the source Observable until the previous value is completely processed.

For example, let’s say that we are triggering a backend save request in response to a click in a save button. We might try first to implement this using the concatMap operator, in order to ensure that the save operations happen in sequence:

  fromEvent(this.saveButton.nativeElement, ‘click’)
  concatMap(() => this.saveCourse(this.form.value))
view raw10.ts hosted with ❤ by GitHub

This ensures the saves are done in sequence, but what happens now if the user clicks the save button multiple times? Here is what we will see in the network log:

RxJs Exhaust Strategy

As we can see, each click triggers its own save: if we click 20 times, we get 20 saves! In this case, we would like something more than just ensuring that the saves happen in sequence.

We want also to be able to ignore a click, but only if a save is already ongoing. The exhaust Observable combination strategy will allow us to do just that.

Exhaust Marble Diagram

To understand how exhaust works, let’s have a look at this marble diagram:

RxJs Exhaust Strategy

Just like before, we have here a higher-order Observable on the first line, whose values are themselves Observables, forking from that top line. Here is what is going on in this diagram:

  • Just like in the case of switch, exhaust is subscribing to the first inner Observable (a-b-c)
  • The value a, b and c get immediately reflected in the output, as usual
  • then a second inner Observable (d-e-f) is emitted, while the first Observable (a-b-c) is still ongoing
  • This second Observable gets discarded by the exhaust strategy, and it will not be subscribed to (this is the key part of exhaust)
  • only after the first Observable (a-b-c) completes, will the exhaust strategy subscribe to new Observables
  • when the third Observable (g-h-i) is emitted, the first Observable (a-b-c) has already completed, and so this third Observable will not be discarded and will be subscribed to
  • the values g-h-i of the third Observable will then show up in the output of the result Observable, unlike to values d-e-f that are not present in the output

Just like the case of concat, merge and switch, we can now apply the exhaust strategy in the context of higher-order mapping.

The RxJs exhaustMap Operator

Let’s now have a look at the marble diagram of the exhaustMap operator. Let’s remember, unlike the top line of the previous diagram, the source Observable 1-3-5 is emitting values that are not Observables.

Instead, these values could for example be mouse clicks:

RxJs exhaustMap Example

So here is what is going on in the case of the exhaustMap diagram:

  • the value 1 gets emitted, and a inner Observable 10-10-10 is created
  • the Observable 10-10-10 emits all values and completes before the value 3 gets emitted in the source Observable, so all 10-10-10 values where emitted in the output
  • a new value 3 gets emitted in the input, that triggers a new 30-30-30 inner Observable
  • but now, while 30-30-30 is still running, we get a new value 5 emitted in the source Observable
  • this value 5 is discarded by the exhaust strategy, meaning that a 50-50-50 Observable was never created, and so the 50-50-50 values never showed up in the output

A Practical Example for exhaustMap

Let’s now apply this new exhaustMap Operator to our save button scenario:

  fromEvent(this.saveButton.nativeElement, ‘click’)
  exhaustMap(() => this.saveCourse(this.form.value))
view raw11.ts hosted with ❤ by GitHub

If we now click save let’s say 5 times in a row, we are going to get the following network log:

RxJs Exhaust Strategy

As we can see, the clicks that we made while a save request was still ongoing where ignored, as expected!

Notice that if we would keep clicking for example 20 times in a row, eventually the ongoing save request would finish and a second save request would then start.

How to choose the right mapping Operator?

The behavior of concatMap, mergeMap, switchMap and exhaustMap is similar in the sense they are all higher order mapping operators.

But its also so different in so many subtle ways, that there isn’t really one operator that can be safely pointed to as a default.

Instead, we can simply choose the appropriate operator based on the use case:

  • if we need to do things in sequence while waiting for completion, then concatMap is the right choice

  • for doing things in parallel, mergeMap is the best option

  • in case we need cancellation logic, switchMap is the way to go

  • for ignoring new Observables while the current one is still ongoing, exhaustMap does just that