Create observable from periodic async request

早过忘川 提交于 2020-12-03 06:34:50

问题


I want a generic way to convert an asynchronous method to an observable. In my case, I'm dealing with methods that uses HttpClient to fetch data from an API.

Let's say we have the method Task<string> GetSomeData() that needs to become a single Observable<string> where the values is generated as a combination of:

  • Repeated periodic calls to GetSomeData() (for example every x seconds)
  • Manually triggered calls to GetSomeData() at any given time (for example when user hits refresh).

Since there is two ways to trigger execution of GetSomeData() concurrency can be an issue. To avoid demanding that GetSomeData() is thread-safe, I want to limit the concurrency so that only one thread is executing the method at the same time. As a consequence I need to handle overlapping requests with some strategy. I made a (kind of) marble diagram trying to describe the problem and wanted outcome

My instinct tells me there is a simple way to achieve this, so please give me some insights :)

This is the solution I've got so far. It unfortunately doesn't solve the concurrency problem.

    public class ObservableCreationWrapper<T>
    {
        private Subject<Unit> _manualCallsSubject = new Subject<Unit>();
        private Func<Task<T>> _methodToCall;
        private IObservable<T> _manualCalls;

        public IObservable<T> Stream { get; private set; }

        public ObservableCreationWrapper(Func<Task<T>> methodToCall, TimeSpan period)
        {
            _methodToCall = methodToCall;
            _manualCalls = _manualCallsSubject.AsObservable()
                .Select(x => Observable.FromAsync(x => methodToCall()))
                .Merge(1);

            Stream = Observable.FromAsync(() => _methodToCall())
                .DelayRepeat(period)
                .Merge(_manualCalls);
        }

        public void TriggerAdditionalCall()
        {
            _manualCallsSubject.OnNext(Unit.Default);
        }
    }

Extension method for repeating with delay:

static class Extensions
{
    public static IObservable<T> DelayRepeat<T>(this IObservable<T> source, TimeSpan delay) => source
        .Concat(
            Observable.Create<T>(async observer =>
            {
                await Task.Delay(delay);
                observer.OnCompleted();
            }))
        .Repeat();
}

An example of a service containing the method to generate the observable

class SomeService
{
    private int _ticks = 0;

    public async Task<string> GetSomeValueAsync()
    {
        //Just a hack to dermine if request was triggered manuall or by timer
        var initiatationWay = (new StackTrace()).GetFrame(4).GetMethod().ToString().Contains("System.Threading.CancellationToken") ? "manually" : "by timer";

        //Here we have a data race! We would like to limit access to this method 
        var valueToReturn = $"{_ticks} ({initiatationWay})";

        await Task.Delay(500);
        _ticks += 1; 
        return valueToReturn;
    }
}

Used like this (data race will occur):

static async Task Main(string[] args)
{
    //Running this program will yield non deterministic results due to data-race in GetSomeValueAsync
    var someService = new SomeService();
    var stopwatch = Stopwatch.StartNew();
    var observableWrapper = new ObservableCreationWrapper<string>(someService.GetSomeValueAsync, TimeSpan.FromMilliseconds(2000));
    observableWrapper.Stream
        .Take(6)
        .Subscribe(x => 
            {
                Console.WriteLine($"{stopwatch.ElapsedMilliseconds} | Request: {x} fininshed");
            });

    await Task.Delay(4000);
    observableWrapper.TriggerAdditionalCall();
    observableWrapper.TriggerAdditionalCall();
    Console.ReadLine();
}

回答1:


Here is my take on this problem:


Update: I was able to simplify greatly my suggested solution by borrowing ideas from Enigmativity's answer. The Observable.StartAsync method handles the messy business of cancellation automatically, and the requirement of non-overlapping execution can be enforced simply by using a SemaphoreSlim.

/// <summary>
/// Creates an observable sequence containing the results of an asynchronous
/// function that is invoked periodically and manually. Overlapping invocations
/// are prevented. Timer ticks that would cause overlapping are ignored.
/// Manual invocations cancel previous invocations, and restart the timer.
/// </summary>
public static IObservable<T> PeriodicAndManual<T>(
    Func<bool, CancellationToken, Task<T>> functionAsync,
    TimeSpan period,
    out Action manualInvocation)
{
    // Arguments validation omitted
    var manualSubject = new Subject<bool>();
    manualInvocation = () => manualSubject.OnNext(true);
    var semaphore = new SemaphoreSlim(1);
    return Observable
        .Interval(period)
        .Select(_ => false) // Not manual
        .Merge(manualSubject)
        .TakeUntil(isManual => isManual) // Stop on first manual
        .Repeat() // ... and restart the timer
        .Prepend(false) // Skip the initial interval delay
        .Scan(seed: (
            // Both representations of an operation are needed
            // The Observable provides automatic cancellation on unsubscription
            // The Task maintains the IsCompleted state
            Operation: (IObservable<T>)null,
            AsTask: Task.FromResult(default(T))
        ), accumulator: (previous, isManual) =>
        {
            // Start a new operation only if the previous operation is completed,
            // or if the call is manual. Otherwise return the previous operation.
            if (!previous.AsTask.IsCompleted && !isManual) return previous;
            // Start a new operation as hot observable
            var operation = Observable.StartAsync(async ct =>
            {
                await semaphore.WaitAsync(ct); // Ensure no overlapping
                try { return await functionAsync(isManual, ct); }
                finally { semaphore.Release(); }
            }, Scheduler.Immediate); // Propagate the task status synchronously
            return (operation, operation.ToTask());
        })
        .Select(entry => entry.Operation) // Discard the AsTask representation
        .DistinctUntilChanged() // Ignore duplicate operations
        .Switch(); // Cancel pending operations and ignore them
}

The out Action manualInvocation argument is the mechanism that triggers a manual invocation.

Usage example:

int ticks = 0;
var subscription = PeriodicAndManual(async (isManual, token) =>
{
    var id = $"{++ticks} " + (isManual ? "manual" : "periodic");
    Console.WriteLine($"{DateTime.Now:HH:mm:ss.fff} Begin {id}");
    await Task.Delay(500, token);
    return id;
}, TimeSpan.FromMilliseconds(1000), out var manualInvocation)
.Do(x => Console.WriteLine($"{DateTime.Now:HH:mm:ss.fff} Received #{x}"))
.Subscribe();

await Task.Delay(3200);
manualInvocation();
await Task.Delay(200);
manualInvocation();
await Task.Delay(3200);

subscription.Dispose();

Output:

19:52:43.684 Begin 1 periodic
19:52:44.208 Received #1 periodic
19:52:44.731 Begin 2 periodic
19:52:45.235 Received #2 periodic
19:52:45.729 Begin 3 periodic
19:52:46.232 Received #3 periodic
19:52:46.720 Begin 4 periodic
19:52:46.993 Begin 5 manual
19:52:47.220 Begin 6 manual
19:52:47.723 Received #6 manual
19:52:48.223 Begin 7 periodic
19:52:48.728 Received #7 periodic
19:52:49.227 Begin 8 periodic
19:52:49.730 Received #8 periodic
19:52:50.226 Begin 9 periodic

The technique of using the Scan and the DistinctUntilChanged operators in order to drop elements while the previous asynchronous operation is running, is borrowed from this question.




回答2:


Here's the query that you need:

var subject = new Subject<Unit>();
var delay = TimeSpan.FromSeconds(1.0);

IObservable<string> query =
    subject
        .StartWith(Unit.Default)
        .Select(x => Observable.Timer(TimeSpan.Zero, delay))
        .Switch()
        .SelectMany(x => Observable.FromAsync(() => GetSomeData()));

If any time you call subject.OnNext(Unit.Default) it will immediately trigger a call to GetSomeData and when then repeat the call based on the TimeSpan set in delay.

The use of .StartWith(Unit.Default) will set the query going immediately there is a subscriber.

The use of .Switch() cancels any pending operations based on a new subject.OnNext(Unit.Default) being called.

This should match your marble diagram.


The above version didn't introduce the delay between values.

Version 2 should.

var subject = new Subject<Unit>();
var delay = TimeSpan.FromSeconds(5.0);

var source = Observable.FromAsync(() => GetSomeData());

IObservable<string> query =
    subject
        .StartWith(Unit.Default)
        .Select(x => source.Expand(n => Observable.Timer(delay).SelectMany(y => source)))
        .Switch();

I've used the Expand operator to introduce a delay between values. As long as source only produces a single value (which FromAsync does) this should work just fine.




回答3:


I'd suggest not try to cancel an already started call. Things will get too messy. If the logic in GetSomeValueAsync involves database call and/or web API call, you simply cannot really cancel the call.

I think the key here is to make sure all the calls to GetSomeValueAsync are serialized.

I created the following solution based on Enigmativity's Version 1. It is tested on a webassembly blazor page on asp.net core 3.1, works fine.

private int _ticks = 0; //simulate a resource you want serialized access

//for manual event, trigger will be 0; for Timer event, trigger will be 1,2,3...
protected async Task<string> GetSomeValueAsync(string trigger)
{
    var valueToReturn = $"{DateTime.Now.Ticks.ToString()}: {_ticks.ToString()} | ({trigger})";

    await Task.Delay(1000);
    _ticks += 1;
    return valueToReturn;
}

//define two subjects
private Subject<string> _testSubject = new Subject<string>();
private Subject<string> _getDataSubject = new Subject<string>();

//driving observable, based on Enigmativity's Version 1
var delay = TimeSpan.FromSeconds(3.0);
IObservable<string> getDataObservable =
    _testSubject
   .StartWith("Init")
   .Select(x => Observable.Timer(TimeSpan.Zero, delay).Select(i => i.ToString()))
   .Switch()
   .WithLatestFrom(_getDataSubject.AsObservable().StartWith("IDLE"))
   .Where(a => a.Second == "IDLE")
   .Select(a => a.First);

//_disposables is CompositeDisposable defined in the page
_disposables.Add(getDataObservable.Subscribe(async t =>
{
     _getDataSubject.OnNext("WORKING");
     //_service.LogToConsole is my helper function to log data to console
     await _service.LogToConsole(await GetSomeValueAsync(t)); 
     _getDataSubject.OnNext("IDLE");
}));

That is it. I used a button to trigger manual events. The _ticks in output is always in sequence, that is, no overlapping happened.



来源:https://stackoverflow.com/questions/64659387/create-observable-from-periodic-async-request

易学教程内所有资源均来自网络或用户发布的内容,如有违反法律规定的内容欢迎反馈
该文章没有解决你所遇到的问题?点击提问,说说你的问题,让更多的人一起探讨吧!