Promises and Asynchron calls in Metro Application and JavaScript

After I wrote this post I realize that I should write a post about promises in general. Because of this I wrote the following post.
The new Windows brought a new type of applications that are called Metro Applications. These applications are design to use fewer resources as possible and to be used very easily on tablets. Metro Applications can be written not only in XAML and C# but also in HTML5 and JavaScript. This is something new for Microsoft world; we can develop native application in JavaScript. Now JavaScript is a first-class citizen in Metro world.
Of course we cannot talk about Metro Applications without talking about asynchrony methods. We need this type of methods as we need water. When we make a call to a service or we want to write something to the disk we need these calls to be asynchronous.
The most used patterns for asynchronous calls in JavaScript is the callback pattern, where we specify as parameter the callbacks functions that will be called when the given actions ends. This pattern is very used in jQuery worlds.
Microsoft decides to use another well know standard for asynchron calls. The guys from Common/JS propose another standard that was adopted by Microsoft. The name of this standards is Promises/A and is very similar with how asynchronous task work on .NET 4.5. Also this standard is based on fluent pattern, because of this we can read and understand any code written more easily. If you already worked with Node.js, Ruby or jQuery 1.5 maybe you already used this standard without knowing.
Basically a promise represents a value that will be fulfilled in the future. The exact moment cannot be defined. In a Metro Application written in JavaScript any promise return an object that have two methods that are very important and similar – then and done.

promise.then(onComplete, onError, onProgress)

The then method have three parameters where we can specify the functions (handlers) that will be called when:

• onComplete - the call end with success
• onError – is called when during the execution of a code an error appeared
• onProgress - a function that is called when the promise reports progress.

Of course we can omit any of the 2th a 3th parameter. The only one that is mandatory is the first one. The done function has the same signature as the then, the difference is the error handling. This function will throw any error that appeared in the call execution, in comparison with then function that will only return an error state.
Using these promises we don’t need to rely on anonymous functions that are defined inline. We can define functions in our code that can be specified in the then or done. When an error is thrown is more easily to detect the original location of the exceptions.
In Metro Applications all the Windows Runtime and Windows Library for JavaScript is defined using this standard. Because of this this knowing this standard is a have to and is not a nice to have (know).

fooWebService.get("http://www.microsoft.com")
    .then(function(result) {
    // process result
        return fooWebService.get("http://windows.azure.com/")
    })
    .then(function(result) {
    // process result
    }

We can have as many then appended one after another. The condition is to have in each onComplete function to return another promise. If we don’t do this, the onComplete of the second then will be immediately called, without waiting the response of the first call.
The WinJS.Promise object defined by the framework came with some helper methods. Besides then and done functions the most important functions are:

• join – create a promise that permits us to specify a list of promises and only when all this promises are fulfilled the onComplete is called.
• any – create a promise that is fulfilled when one of the promises is fullfilled
• wrap – wrap a non-promise function into a promise. It is very useful when we have a chain of promises.

For example we can use any to wait any specified promises from the list to be fullfiled. When one of them is fullfiled the callback is automatically caled.

WinJS.Promise.any([someMethod1Async(), someMethod2Async("someParam")])
    .then(function() { console.log("done");});

Promises are not only defined by the system, we can also defined promises very easily. When we create a promise we need to create and return a WinJS.Promise object. The constructor accepts a function that has three parameters (the onComplete, onError, onProgress parameters). In the next example we defined a promise that wait for 10 seconds and call the complete function (this a one of the classic examples):

   function wait10SecondsAsync() {
        return new WinJS.Promise(function (onComplete, onError, onProgress) {
            var seconds = 10;
            var intervalId = window.setInterval(function () {
                seconds--;
                onProgress(seconds);
                if (seconds < 1) {
                    window.clearInterval(intervalId);
                    onComplete();
                }
            }, 1000);
        });
    }

In the end some small hints when working with promises:

• always name suffix an asynchrone method with “Async”
• when more than one promise are bound don’t forget to return the promises on each onComplete function
• for the last call of an asynchrone chain call is recommended to use done and not then (done throws any exceptions to the upper code)
• use WinJS.Promise.any and WinJs.Promise.wait when you need to wait for one or more promises

In conclusion promises are very powerful when we are talking about asynchrone calls in JavaScript language. When are used clevered, promises can help us not only to write beautiful cod but the code it almost writes by himself.

Read More

What are the base async patterns in .NET world

In momentul de fata exista diferite paternuri care sunt folosite pentru apeluri asyncrone. Am observat ca destul de multa lume incurca aceste paternuri si se face un mix intre acestea.
Din cate am observat pana acuma exista 3 paternuri principale care sunt folosite in general in .NET (mai putin 4.5, despre care o sa vorbim separat).
EAP (Event based Asynchronous Pattern) - in acest patern orice schimbare de stare (actiune se termina de executat, apare o eroare, progresul) sunt notificate prin intermediul unor evenimente.

public class FooEAP
{   
    public void CalculateAsync(int value1, int value2)
{
    int sum = value1 + value2;
    OnCalculateComplete(sum);
}
public event CalculateCompleteEventHandler CalculateComplete;
private void OnCalculateComplete(int result)
{
    if( CalculateComplete != null )
    {
        CalculateComplete(this, new FooCalculateEventArgs(result));
    }
}
}

APM (Asynchronous Programming Model) - paternul este destul de usor de recunoscut prin formatul BeginXXX si EndXXX. In momentul in care un utilizator apeleaza BeginXXX, task-ul incepe sa se execute. Aceasta metoda o sa returneze o implementare a interfetei IAsyncResult care are urmatoarea forma:

public interface IAsyncResult
{
WaitHandle AsyncWaitHandle { get; }
Boolean    IsCompleted     { get; }
Object     AsyncState      { get; }
Boolean    CompletedSynchronously { get; }
}

Aceasta interfata se poate folosii pentru a verifica in ce stare se afla actiunea noastra sau sa vedem progresul. Acest patern este destul de intalnit in WCF.

public class FooAPM
{
    IAsyncResult calculateAsyncResult;
    public IAsyncResult BeginCalculate(int value1, int value2)
{
    ...   
}
    public void EndInvoke()
{
    // Wait the task to finish
}
}

TPL (Task Parallel Library) - se bazeaza pe task-uri care au fost introduse cu .NET 4.0. Nu reprezinta un patern in sine, dar din .NET 4.5 mai ales o sa fim nevoiti sa le folosim din ce in ce mai des. In general acestea se foloseste cand lucram cu API care se bazeaza pe task-uri. TPL simplifica foarte mult paternul asyncron. Folosind TPL ne este mult mai usor sa scriem si sa folosim cod asyncron.

public class FooTPL
{
    public Task<int> Calculate(int value1, int value2)
{
    return new Task<int>(() => { return value1 + value2; });
}
}

Task-urile sunt destul de complexe din puct de vedere a API-ului. Sunt extrem de usor de folosit, dar trebuie folosite cu cap deoarece pot sa faca mai mult rau decat bine.
Odata cu .NET 4.5 au aparut async si await, iar orice apel poate sa fie executat asyncron fara sa mai scrim nici o linie de cod. In spate ele se bazeaza pe task-uri si fac parte din TPL si vin in ajutor programatorului. Folosinf aceste doua chei ne este multi mai usor sa scriem si sa facem apeluri asyncrone. In general aceste metode sunt denumite sub forma XXXAsync.

public class FooTPL
{
    public async Task<int> CalculateAsync(int value1, int value2)
{   
    int sum = await CalculateHelper.SumAsync(value1, value2);
return sum;
}
}

Despre cum functioneaza await, o sa revin intr-un post viitor.
Sper ca dupa citirea acestui post va este putin mai clar fiecare din aceste "paternuri". Toate fac acelasi lucru, doar ca in diferite moduri.

Read More

Short brief - Request-Response, Asynchron and Fire and Forget patterns

Un apel spre un serviciu poate sa fie facut in diferite moduri. Cele mai uzuale moduri sunt cele care respecta paternul Request-Response si Asynchron. Pe langa aceste doua paternuri mai exista inca unul, dar care este mai rar folosit Fire and Forget.
Request-Response pattern- se refera la faptul ca pentru fiecare request facut se asteapta raspunsul.

Response response = service.CallService(request)

Asynchron pattern - este un patern bazat pe evenimente. Odata ce un apel a fost facut, aplicatia nu asteapta raspunsul de la server. Cel care face apelul trebuie sa se inregistreze la un eveniment care o sa fie declansat cand raspunsul soseste.

service.ResponseEvent += new EventHandler(ResponseReceived);
service.CallServiceAsync(request);

Fire and Forget pattern - este destul de asemanator cu paternul Asynchron, doar ca odata comanda trimisa nu ne mai intereseaza ce se intampla cu aceasta.

service.CalService(request);    // Metoda nu returneaza nimica

Exista diferite scenarii cand Fire and Forget este recomandat sa fie folosit. Conditia principala care trebuie sa fie implinita ca sa putem folostii acest patern este ca sa ne putem asigura ca request-ul ajunge la server - avem un protocol ce ne garanteaza ca mesajele au ajuns la destinatie( de exemplu daca folosim cozi de mesaje care ne asigura ca mesajele nu se pierd - de exemplul MQ).
In cazul in care avem servicii cu un volum mare de request-uri, atunci poate pentru unele tipuri de request-uri putem sa folosim acest patern. De exemplu pentru cele in care rezultatul este format din mesaje precum Okay.
Cele mai mari dezavantaje apar in momentul in care ceva nu functioneaza bine. Debug-ul este destul de complicat si de obicei necesita interventia umana pentru a putea gasi problema. Problemele care pot sa apara sunt la nivelului protocolului care ar trebui sa ne garanteze ca mesajele ajung la destinatie.
Un alt mod de a apela un serviciu este de a face unu sau mai multe request-uri de tipul Fire and Forget, iar apoi la un moment dat sa se verifice daca a venit un raspuns( intr-un cache, in baza de date etc) - in cazul in care raspunsul se pune intr-o locatie unde atat clientul cat si serverul le pot accesa
Nu trebuie confundat un apel de forma:

service.CallService(request,()=>{ ... });

ca si cum ar fi un apel de tip Fire and Forget. Orice metoda sau lambda expresion specificata ca si al doilea parametru care este apelata in momentul in care raspunsul soseste de la server reprezinta un apel de tip asincron.

Read More