shareReplayLatestWhileConnected with RxJS

Question

I'm creating my source observable like this (make api call every 5s):

const obs$ = Observable.interval(5000).switchMap(() => makeApiCall());

And I want to modify $obs so that it has the following characteristics:

• start the observable only when at there's at least 1 subscriber
• multicast. I.e. if I obs$.subscribe(...) twice, the underlying code makeApiCall() should only run once.
• any subscriber which subscribes at any time should have immediately the last emitted value (and not wait ~5s until the next value emits)
• retryable. If one makeApiCall() errors, I want (if possible) all subscribers to get an error notification, but reconnect to $obs, and continue doing makeApiCall() every 5s

So far I found the following leads:

It seems like I'd need to create a BehaviorSubject myBehaviorSubject, do a single subscription obs$.subscribe(myBehaviorSubject), and any other observers should subscribe to myBehaviorSubject. Not sure if that answers the "retryable" part.

I also looked at shareReplay, seems like $obs.shareReplay(1) would do the trick (for the 4 requirements). If I understood correctly it subscribes a ReplaySubject(1) to the source observable, and future observers subscribe to this ReplaySubject. Is there an equivalent shareBehavior?

In RxSwift, I found shareReplayLatestWhileConnected, which seems like the shareBehavior I was imagining. But it doesn't exist in RxJS.

Any ideas what is the best way to achieve this?

Answer 1

As you mentioned, shareReplay(1) pretty much gets you there. It will multicast the response to current subscribers and replay the last value (if there is one) to new subscribers. That seems like what you would want rather than shareBehavior (if it existed) since you are calling an api and there isn't an initial value.

You should know that shareReplay will create a subscription to the source stream but will only unsubscribe when refCount === 0 AND the source stream terminates (error or complete). This means that after the first subscription that the interval will start and even when there are no more subscriptions it will continue.

If you want to stop the interval when no-one is subscribed then use multicast(new ReplaySubject(1)).refCount(). The multicast operator will create a single subscription to the source stream and push all values into the subject provided as an instance (multicast(new Subject())) or by the factory (multicast(() => new Subject())). All subscribers to the stream after the multicast will subscribe to the multicast subject. So when a value flows through the multicast operator all of its subscribers will get that value. You can change the type of subject that you pass to multicast to change its behavior. In your case you probably want a ReplaySubject so that it will replay the last value to a new subscriber. You could use a BehaviorSubject too if you felt that met your need.

Now the multicast operator is connectable meaning that you would have to call connect() on the stream to make it hot. The refCount operator basically makes a connectable observable act like an ordinary observable in that it will become hot when subscribed but will become cold when there are no subscribers. It does this be keeping an internal reference count (hence the name refCount). When refCount === 0 it will disconnect.

This is the same thing as shareReplay(1) with one minor but important difference which is that when there are no more subscribers that it will unsubscribe from the source stream. If you are using a factory method to create a new subject when subscribing to the source (ex: multicast(() => new ReplaySubject(1))) then you will lose your value when the stream goes from hot to cold to hot since it will create a new subject each time it goes hot. If you want to keep the same subject between source subscriptions then you can pass in a subject instead of a factory (ex: multicast(new ReplaySubject(1)) or use its alias publishReplay(1).

As far as your last requirement of providing errors to your subscribers and then resubscribing, you can't call the error callback on a subscription and then continue getting values on the next callback. An unhandled error will end a subscription if it reaches it. So you have to catch it before it gets there and turn it into a normal message if you want your subscription to see it and still live. You can do this like so: catch((err) => of(err)) and just flag it somehow. If you want to mute it then return empty().

If you want to retry immediately then you could use the retryWhen operator but you probably want to put that before the sharing operator to make it universal. However this also prevents your subscribers from knowing about an error. Since the root of your stream is an interval and the error came from the inner observable returned from the switchMap, the error will not kill the source of the stream but it could kill the subscription. So as long as you handle the error (catch/catchError) the api call will be retried on the next interval.

Also, you may want timer(0, 5000) instead of interval so that your api call immediately fires and then fires on a 5 second interval after that.

So I would suggest something like the following:

let count = 0;
function makeApiCall() {
  return Rx.Observable.of(count++).delay(1000);
}

const obs$ = Rx.Observable.timer(0, 5000)
  .switchMap(() => makeApiCall().catch(() => Rx.Observable.empty()))
  .publishReplay(1)
  .refCount();
  
console.log('1 subscribe');
let firstSub = obs$.subscribe((x) => { console.log('1', x); });
let secondSub;
let thirdSub;

setTimeout(() => {
  console.log('2 subscribe');
  secondSub = obs$.subscribe((x) => { console.log('2', x); });
}, 7500);

setTimeout(() => {
  console.log('1 unsubscribe');
  firstSub.unsubscribe();
  console.log('2 unsubscribe');
  secondSub.unsubscribe();
}, 12000);

setTimeout(() => {
  console.log('3 subscribe');
  thirdSub = obs$.subscribe((x) => { console.log('3', x); });
}, 17000);

setTimeout(() => {
  console.log('3 unsubscribe');
  thirdSub.unsubscribe();
}, 30000);

<script src="https://cdnjs.cloudflare.com/ajax/libs/rxjs/5.5.10/Rx.min.js"></script>

For convenience, here are aliases for multicast:

publish() === multicast(new Subject())
publishReplay(#) === multicast(new ReplaySubject(#))
publishBehavior(value) === multicast(new BehaviorSubject(value))

Answer 2

I just tried to implement this with rxjs 6, but the implementation feels kinda hacky. I think there should be a much cleaner way to achieve this.

The expected behavior is:

• As long as there are observers, they all get the same values.
• When there are 0 observers, the source subscription is closed but the ReplaySubject is not completed.
• When new observers subscribe again they get the last N values and a new subscription to source is established.
• When the source completes or throws an error, current observers are completed resp. notified.
• After source completion or source error, new subscribers don't get replayed values any more and are completed immediately.

export function shareReplayLatestWhileConnected<T>(count?: number) {
  return function (source: Observable<T>): Observable<T> {
    let done = false;
    return source.pipe(
      // Identify when source is completed or throws an error.
      tap(
        null,
        () => (done = true),
        () => (done = true),
      ),
      multicast(
        // Subject for multicasting
        new ReplaySubject<T>(count),
        // Selector function. Stop subscription on subject, when source is done, to kill all subscriptions.
        (shared) => shared.pipe(takeWhile(() => !done)),
      ),
      // I was not able to get rid of duplicate subscriptions. Multicast subscribed multiple times on the source.
      share(),
    );
  };
}

Any tips on how I could improve this solution are very appreciated.

Use it like this:

const shared$ = source$.pipe(shareReplayLatestWhileConnected(1));