问题
Before coming here I have read the official documentation of Rxjs and some other pages but I am still not clear. What I understood is this:
It is used to "join" 2 observables and thus obtain a single observable as a result, I also saw that it is used to "flatten" an observable (I am also not very clear).
Now ... I have days trying to program a user registry using Angular and Node.js with Express and I found a little tutorial which I decided to use and it has this code:
import { Injectable, Injector } from '@angular/core';
import { HttpClient, HttpInterceptor, HttpRequest, HttpHandler, HttpEvent, HttpErrorResponse } from '@angular/common/http';
import { Observable, throwError } from 'rxjs';
import { catchError, retry, mergeMap } from 'rxjs/operators'
import { AuthenticationService } from './authentication.service';
@Injectable({
providedIn: 'root'
})
export class AppInterceptor implements HttpInterceptor {
constructor(private injector: Injector) { }
intercept(req: HttpRequest<any>, next: HttpHandler): Observable<HttpEvent<any>> {
let accessToken = "", refreshToken = ""
const tokens = JSON.parse(sessionStorage.getItem("tokens"))
if (tokens) {
accessToken = tokens.accessToken
refreshToken = tokens.refreshToken
}
let clonHttp: HttpRequest<any>
clonHttp = tokens ? req.clone({ headers: req.headers.append("Authorization", `Bearer ${accessToken}`) }) : req
let auth = this.injector.get(AuthenticationService);
return next.handle(clonHttp)
.pipe(
catchError((error: HttpErrorResponse) => {
if (error.error instanceof ErrorEvent) {
console.log("error event")
} else if (error.status == 401) {
return auth.getNewAccessToken(refreshToken)
.pipe(
retry(3),
mergeMap(
(response: any) => {
tokens.accessToken = response.accessToken
sessionStorage.setItem("tokens", JSON.stringify(tokens))
clonHttp = req.clone({ headers: req.headers.append("Authorization", `Bearer ${response.accessToken}`) })
return next.handle(clonHttp)
}
)
)
} else if (error.status == 409) {
return throwError("User not logged")
} else {
if (error.error && error.error.message) {
return throwError(error.error.message)
} else {
return throwError("Check your connection")
}
}
})
)
}
}If you see, when you use the MergeMap operator they only pass you the answer (a single observable), or at least that's what I can see. What I'm trying to say is that I don't see that they are using it with 2 observables or to mix 2 observables, which is what I have read in their official documentation, in fact, in the examples they show they always use it with 2 observables.
Honestly it has been too difficult for me to understand this operator, if someone could help me understand it in a simple way, I would be extremely grateful, in addition to understanding its use in that code that I show earlier. Greetings in advance. Thank you!
回答1:
mergeMap, like many other so-called higher order mapping operators, maintains one or multiple inner observables.
An inner observable is created with the outer value and the provided function. The outer value essentially is just the value received from its source. For example:
of(1, 2, 3).pipe(
mergeMap((outerValue, index) => /* ... return an observable ... */)
).subscribe(); // `outerValue`: 1, 2, 3 (separately)
When an outer value comes in, a new inner observable will be created. I think the best way to understand this is to have a look at the source code:
// `value` - the `outerValue`
protected _next(value: T): void {
if (this.active < this.concurrent) {
this._tryNext(value);
} else {
this.buffer.push(value);
}
}
protected _tryNext(value: T) {
let result: ObservableInput<R>;
const index = this.index++;
try {
// Create the inner observable based on the `outerValue` and the provided function (`this.project`)
// `mergeMap(project)`
result = this.project(value, index);
} catch (err) {
this.destination.error(err);
return;
}
this.active++;
// Subscribe to the inner observable
this._innerSub(result, value, index);
}
Please disregard for now concurrent and buffer, we'll have a look at them a bit later.
Now, what happens when an inner observable emits ? Before going any further, it's worth mentioning that, although it's obvious, an inner observable requires an inner subscriber. We can see this in the _innerSub method from above:
private _innerSub(ish: ObservableInput<R>, value: T, index: number): void {
const innerSubscriber = new InnerSubscriber(this, value, index);
const destination = this.destination as Subscription;
destination.add(innerSubscriber);
// This is where the subscription takes place
subscribeToResult<T, R>(this, ish, undefined, undefined, innerSubscriber);
}
When an inner observable emits, the notifyNext method will be called:
notifyNext(outerValue: T, innerValue: R,
outerIndex: number, innerIndex: number,
innerSub: InnerSubscriber<T, R>): void {
this.destination.next(innerValue);
}
Where destination points to the next subscriber in the chain. For example, it can be this:
of(1)
.pipe(
mergeMap(/* ... */)
)
.subscribe({} /* <- this is the `destination` for `mergeMap` */)
This will be explained in more detail in What about the next subscriber in the chain below.
So, what does it mean to to mix 2 observables ?
Let's see this example:
of(2, 3, 1)
.pipe(
mergeMap(outerValue => timer(outerValue).pipe(mapTo(outerValue)))
)
.subscribe(console.log)
/* 1 \n 2 \n 3 */
When 2 arrives, mergeMap will subscribe to an inner observable that will emit in 200ms. This is an asynchronous action, but notice that the outer values(2, 3, 1) arrive synchronously. Next, 3 arrives and will create an inner obs. that will emit in 300ms. Since the current script has not finished executing yet, the callback queue is not yet considered. Now 1 arrives, and will create an inner obs. that will emit in 100 ms.
mergeMap has now 3 inner observables and will pass along the inner value of whichever inner observable emits.
As expected, we get 1, 2, 3.
So that's what mergeMap does. Mixing observables can be thought of this way: if an outer value comes and an inner observable has already been created, then mergeMap simply says: "no problem, I'll just create a new inner obs. and subscribe to it".
What about concurrent and buffer
mergeMap can be given a second argument, concurrent which indicates how many inner observables should handle at the same time. These number of active inner observables is tracked with the active property.
As seen in _next method, if active >= concurrent, the outerValues will be added to a buffer, which is a queue(FIFO).
Then, when one active inner observable completes, mergeMap will take the oldest value from the value and will create an inner observable out of it, using the provided function:
// Called when an inner observable completes
notifyComplete(innerSub: Subscription): void {
const buffer = this.buffer;
this.remove(innerSub);
this.active--;
if (buffer.length > 0) {
this._next(buffer.shift()!); // Create a new inner obs. with the oldest buffered value
} else if (this.active === 0 && this.hasCompleted) {
this.destination.complete();
}
}
With this in mind, concatMap(project) is just mergeMap(project, 1).
So, if you have:
of(2, 3, 1)
.pipe(
mergeMap(outerValue => timer(outerValue * 100).pipe(mapTo(outerValue)), 1)
)
.subscribe(console.log)
this will be logged:
2 \n 3 \n 1.
What about the next subscriber in the chain
Operators are functions that return another function which accepts an observable as their only parameter and return another observable. When a stream is being subscribed to, each observable returned by an operator will have its own subscriber.
All these subscribers can be seen as a linked list. For example:
// S{n} -> Subscriber `n`, where `n` depends on the order in which the subscribers are created
of(/* ... */)
.pipe(
operatorA(), // S{4}
operatorB(), // S{3}
operatorC(), // S{2}
).subscribe({ /* ... */ }) // S{1}; the observer is converted into a `Subscriber`
S{n} is the parent(destination) of S{n+1}, meaning that S{1} is the destination of S{2}, S{2} is the destination of S{3} and so forth.
StackBlitz
Unexpected results
Compare these:
of(2, 1, 0)
.pipe(
mergeMap(v => timer(v * 100).pipe(mapTo(v)))
).subscribe(console.log)
// 0 1 2
of(2, 1, 0)
.pipe(
mergeMap(v => timer(v).pipe(mapTo(v)))
).subscribe(console.log)
// 1 0 2
As per MDN:
The specified amount of time (or the delay) is not the guaranteed time to execution, but rather the minimum time to execution. The callbacks you pass to these functions cannot run until the stack on the main thread is empty.
As a consequence, code like setTimeout(fn, 0) will execute as soon as the stack is empty, not immediately. If you execute code like setTimeout(fn, 0) but then immediately after run a loop that counts from 1 to 10 billion, your callback will be executed after a few seconds.
This section by MDN should clarify things as well.
I'd say this is environment-specific, rather than RxJs-specific.
In the second snippet, the delays are consecutive so that's why you're getting unexpected results. If you increase the delays just a bit, like: timer(v * 2), you should get the expected behavior.
回答2:
So merge map is mainly used to resolve multiple inner observables concurrently and when all inner observables are resolved outer observable will resolve. I hope this helps.
回答3:
Imagine you have to to read a list of ids from some async source, being it a remote service, a DB, a file on your file system.
Imagine that you have to launch an async query for each id to get the details.
Imagine you have to collect all the details for each id and do something else.
You end up having an initial Obsersable emitting a list and then a bunch of of Observables generated by that list. This is were you would use mergeMap.
The code would look like this
mySourceObs = getIdListFromSomewhere();
myStream = mySourceObs.pipe(
// after you get the list of the ids from your service, you generate a new strem
// which emits all the values of the list via from operator
concatMap(listOfIds => from(listOfIds)),
// for each id you get the details
mergeMap(id => getDetails(id),
)
If you subscribe to myStream you get a stream of details data, one for every id of the original list. The code would be simply
myStream.subscribe(
detail => {
// do what you have to do with the details of an id
}
)
MORE ON THE CODE REFERENCED IN THE QUESTION
My understanding of piece of code using mergeMap is the following:
- you fetch a new token with
auth.getNewAccessToken - If something goes wrong you retry 3 times
- When you receive a fresh token, you do some stuff and then you clone something with
next.handle(clonHttp)
Key point is that both auth.getNewAccessToken and next.handle(clonHttp) are async calls returning an Observable.
In this case you want to make sure that FIRST you get the response from auth.getNewAccessToken and ONLY THEN you call next.handle(clonHttp).
In this case the best way to code such logic is using concatMap which ensures that the second Observable is concatenated to the successful completion of the first one.
mergeMap and switchMap can also work in this scenario since auth.getNewAccessToken emits only ONCE and then completes, but the right semantic is given by concatMap (which by the way is the same as mergeMap with concurrency set to 1, but this is another story).
来源：https://stackoverflow.com/questions/61646013/how-exactly-does-the-mergemap-operator-work-and-in-which-cases-is-it-used