Allows the use of operators for controlling the timing around when actions scheduled on workers are actually done. This makes it possible to layer additional behavior on this Scheduler. The only parameter is a function that flattens an Observable of Observable of Completables into just one Completable. There must be a chain of operators connecting the returned value to the source Observable otherwise any work scheduled on the returned Scheduler will not be executed.

When Scheduler#createWorker() is invoked a Observable of Completables is onNext'd to the combinator to be flattened. If the inner Observable is not immediately subscribed to an calls to Worker#schedule are buffered. Once the Observable is subscribed to actions are then onNext'd as Completables.

Finally the actions scheduled on the parent Scheduler when the inner most Completables are subscribed to.

When the Worker is unsubscribed the Completable emits an onComplete and triggers any behavior in the flattening operator. The Observable and all Completables give to the flattening function never onError.

Limit the amount concurrency two at a time without creating a new fix size thread pool:

Scheduler limitScheduler = Schedulers.computation().when(workers -> {
 // use merge max concurrent to limit the number of concurrent
 // callbacks two at a time
 return Completable.merge(Observable.merge(workers), 2);
});

This is a slightly different way to limit the concurrency but it has some interesting benefits and drawbacks to the method above. It works by limited the number of concurrent Workers rather than individual actions. Generally each Observable uses its own Worker. This means that this will essentially limit the number of concurrent subscribes. The danger comes from using operators like Flowable#zip(org.reactivestreams.Publisher, org.reactivestreams.Publisher, io.reactivex.rxjava3.functions.BiFunction) where subscribing to the first Observable could deadlock the subscription to the second.

Scheduler limitScheduler = Schedulers.computation().when(workers -> {
 // use merge max concurrent to limit the number of concurrent
 // Observables two at a time
 return Completable.merge(Observable.merge(workers, 2));
});

Slowing down the rate to no more than than 1 a second. This suffers from the same problem as the one above I could find an Observable operator that limits the rate without dropping the values (aka leaky bucket algorithm).

Scheduler slowScheduler = Schedulers.computation().when(workers -> {
 // use concatenate to make each worker happen one at a time.
 return Completable.concat(workers.map(actions -> {
     // delay the starting of the next worker by 1 second.
     return Completable.merge(actions.delaySubscription(1, TimeUnit.SECONDS));
 }));
});

History 2.0.1 - experimental