[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content

tommasobertoni/NScatterGather

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

NScatterGather logo

NScatterGather


Table of Contents

Intro
How to use
When to use
Special cases
Samples

Intro

The Scatter-Gather pattern: send a request to multiple recipients, and aggregate the results into a single response.

scatter-gather-diagram

This pattern helps to limit the coupling between the consumer and the recipients in integration scenarios, and provides standard error-handling and timeout capabilities.

How to use

Use a RecipientsCollection to register the eligible recipients:

var collection = new RecipientsCollection();
collection.Add<Foo>(); // type
collection.Add(new Bar()); // instance
collection.Add((int n) => n.ToString()); // delegate

Use an Aggregator for sending the requests to all the available recipients that can support the desired request/response types, and for aggregating the results:

var aggregator = new Aggregator(collection);

// Send a request to all the recipients
// capable of accepting and int.
// The results are then combined in the response:
AggregatedResponse<object> objects = await aggregator.Send(42);

// The following overload can be used when
// the return type is either known or binding.
// Only the recipients that accept an int and
// return a string will be invoked:
AggregatedResponse<string> strings = await aggregator.Send<string>(42);

Inspect the AggregatedResponse containing the results of the scatter-gather operation, grouped by completed, faulted and incomplete:

var response = await aggregator.Send<string>(42, timeout: TimeSpan.FromSeconds(5));

var completed = response.Completed[0];
// (_, _, string result, TimeSpan duration) = completed;

var faulted = response.Faulted[0];
// (_, _, Exception? exception, TimeSpan duration) = faulted;

var incomplete = response.Incomplete[0];
// (string? recipientName, Type? recipientType) = incomplete;

Embrace duck typing

No binding contracts are used (e.g. IRecipient).
A recipient is invoked if it defines a method matching the request:

class Foo
{
    public string ThisIsInvoked(int n) => n.ToString();
}

class Bar
{
    public int ThisIsInvokedToo(int n) => n * 2;
}

// Invoke every recipient that accepts an int.
_ = await aggregator.Send(42);

When to use

Competing Tasks

The recipients compete in order to provide the best, or the fastest, response to the request. The consumer will then pick the best value from the aggregated response.

sample: get an item's best price from a collection of suppliers:

competing-tasks-diagram

Task parallelization

Different operations are computed concurrently, and their results combined or used together. The result types could be different.

sample: get a user's data from different services, and then compose into a model:

tasks-parallelization-diagram

Special cases

Handling async methods

The Aggregator exposes async-only methods for sending requests.

Even if the consumer requested only results of type TResponse, a recipient that returns Task<TResponse>, or ValueTask<TResponse>, or any task-like type, will still be invoked and its result awaited:

class Foo { public int Echo(int n) => n; }
class Bar { public Task<int> EchoAsync(int n) => Task.FromResult(n); }

// Nothing changes!
var response = await aggregator.Send(42);
// [ 42, 42 ]

Handling conflicts

Sometimes, a recipient can have two or more methods conflicting, given a request type:

class Foo
{
    public int Double(int n) => n * 2;
    public long Triple(int n) => n * 3L;
}

In this case, the aggregator will be able to invoke the recipient only if the return type of the conflicting methods is different, and it's explicitely defined by the consumer:

// The recipient won't be used.
_ = await aggregator.Send(42);

// Method "Triple" will be invoked.
var response = await aggregator.Send<long>(42);

Samples

Hello world

class Foo { public int Double(int n) => n * 2; }
class Bar { public long Square(int n) => n * 1L * n; }

var response = await aggregator.Send(42);
// [ 84, 1764L ]

Specify the response type

class Foo { public string Stringify(int n) => n.ToString(); }
class Bar { public long Longify(int n) => n * 1L; }

var onlyStrings = await aggregator.Send<string>(42);
// [ "42" ]

Invoke async methods

class Foo { public string Stringify(int n) => n.ToString(); }

class Bar
{
    public async Task<long> Longify(int n)
    {
        await Task.Yield();
        return n * 1L;
    }
}

var response = await aggregator.Send(42);
// [ "42", 42L ]

Error handling

class Foo
{
    public string Todo(string s) =>
        throw new NotImplementedException("TODO");
}

var response = await aggregator.Send("Don't Panic");
var (recipientType, exception) = response.Faulted[0];
// ( typeof(Foo), NotImplementedException("TODO") )

Timeout

class Foo
{
    public Task<int> Block(int n)
    {
        var tcs = new TaskCompletionSource<int>();
        return tcs.Task; // It will never complete.
    }
}

var response = await aggregator.Send(42, timeout: TimeSpan.FromSeconds(5));
Type recipientType = response.Incomplete[0];
// typeof(Foo)

Typeless recipients

// Register typeless recipients using delegates:
collection.Add((int n) => n.ToString());
collection.Add((int n) => n * 2);

// Use them as normal recipients:
var aggregator = new Aggregator(collection);
var responseOfInt = await aggregator.Send(42);
var resultsOfInt = responseOfInt.AsResultsList(); // 84, "42"

For more, take a look at the samples project in solution.

Continuous Integration

github-actions xUnit coverlet coveralls.io codefactor.io