Dotnet Csharp Concurrency Patterns

.NET Concurrency: Choosing the Right Tool

Choose the right concurrency abstraction in .NET — from async/await for I/O to Channels for producer/consumer to Akka.NET for stateful entity management. Avoid locks and manual synchronization unless absolutely necessary.

Acronyms: I/O (Input/Output), CPU (Central Processing Unit), Rx (Reactive Extensions), DI (Dependency Injection), UI (User Interface).

When to Use This Skill

• Deciding how to handle concurrent I/O-bound or CPU-bound operations in .NET applications
• Evaluating whether to use async/await, Channels, Akka.NET Streams, or Actors for a scenario
• Implementing producer/consumer patterns with backpressure and bounded buffering in C#
• Managing independent state for thousands of entities using the actor model pattern
• Designing server-side stream processing pipelines with batching, throttling, or merging
• Composing UI event streams with debouncing, throttling, or combining multiple sources

Related Skills

Core Principles

1. Start Simple, Escalate When Needed — Use async/await by default. Only reach for Channels, Streams, or Actors when you have a concrete need that simpler tools cannot address cleanly.
2. Avoid Shared Mutable State — Immutable data, message passing, and isolated state (actors) eliminate entire categories of concurrency bugs. Redesign before reaching for locks.
3. Use the Right Level of Abstraction — Match the tool to the problem: I/O waits → async/await, work queues → Channels, stream processing → Akka.NET Streams, stateful entities → Actors.
4. Prefer Message Passing Over Synchronization — Channels and Actors serialize access through messages. Locks should be the last resort, reserved for low-level infrastructure code.
5. Always Support Cancellation — Accept CancellationToken in every async method. Cooperative cancellation is essential for responsive, well-behaved .NET applications.

Values: 基礎と型の追求（async/await という基礎を徹底し、必要な時だけ高度な抽象へ段階的に進む）, ニュートラルな視点（ツールに偏らず、問題に応じて最適な抽象度を選択する）

Workflow: Choose and Apply Concurrency Patterns

Step 1: Use async/await for I/O Operations

Apply async/await as the default choice for all I/O-bound operations. Use Task.WhenAll for independent parallel operations. Why: async/await handles most concurrency needs without additional complexity.

// Simple async I/O — always accept CancellationToken
public async Task<Order> GetOrderAsync(string orderId, CancellationToken ct)
{
    var order = await _database.GetAsync(orderId, ct);
    var customer = await _customerService.GetAsync(order.CustomerId, ct);
    return order with { Customer = customer };
}

// Parallel independent operations with Task.WhenAll
public async Task<Dashboard> LoadDashboardAsync(string userId, CancellationToken ct)
{
    var ordersTask = _orderService.GetRecentOrdersAsync(userId, ct);
    var notificationsTask = _notificationService.GetUnreadAsync(userId, ct);
    var statsTask = _statsService.GetUserStatsAsync(userId, ct);

    await Task.WhenAll(ordersTask, notificationsTask, statsTask);

    return new Dashboard(
        Orders: await ordersTask,
        Notifications: await notificationsTask,
        Stats: await statsTask);
}

Key rules: Always accept CancellationToken. Use ConfigureAwait(false) in library code. Never block on async code (no .Result or .Wait()).

Values: 基礎と型の追求（async/await を基本の型として徹底し、すべての応用の土台とする）

Step 2: Scale CPU Work with Parallel.ForEachAsync

Apply Parallel.ForEachAsync for CPU-bound parallel processing with controlled concurrency. Why: it combines async support with bounded parallelism, preventing thread pool starvation.

public async Task ProcessOrdersAsync(IEnumerable<Order> orders, CancellationToken ct)
{
    await Parallel.ForEachAsync(
        orders,
        new ParallelOptions
        {
            MaxDegreeOfParallelism = Environment.ProcessorCount,
            CancellationToken = ct
        },
        async (order, token) =>
        {
            await ProcessOrderAsync(order, token);
        });
}

Use when: CPU-bound work across collections. Avoid when: pure I/O (async/await suffices), order matters, or you need backpressure.

Values: 継続は力（並列処理の基本パターンを身につけ、日常的な処理の高速化に活かす）

Step 3: Decouple with System.Threading.Channels

Apply Channel<T> for producer/consumer patterns, work queues, and decoupling component speeds. Use bounded channels for backpressure. Why: Channels provide thread-safe, lock-free communication between producers and consumers.

public class OrderProcessor
{
    private readonly Channel<Order> _channel;

    public OrderProcessor()
    {
        // Bounded channel provides backpressure
        _channel = Channel.CreateBounded<Order>(new BoundedChannelOptions(100)
        {
            FullMode = BoundedChannelFullMode.Wait
        });
    }

    // Producer
    public async Task EnqueueOrderAsync(Order order, CancellationToken ct)
        => await _channel.Writer.WriteAsync(order, ct);

    // Consumer (run as background task)
    public async Task ProcessOrdersAsync(CancellationToken ct)
    {
        await foreach (var order in _channel.Reader.ReadAllAsync(ct))
        {
            await ProcessOrderAsync(order, ct);
        }
    }

    public void Complete() => _channel.Writer.Complete();
}

Use when: decoupling producer/consumer speeds, buffering work, fan-out to multiple workers. Avoid when: complex stream operations (batching, windowing) or stateful entity processing.

See references/channel-patterns.md for multi-consumer worker pools and advanced patterns.

Values: 成長の複利（Channel の習得がメッセージパッシングの理解を深め、Actor モデルへの橋渡しになる）

Step 4: Process Streams with Akka.NET Streams or Rx

Apply Akka.NET Streams for server-side pipelines needing backpressure, batching, or throttling. Apply Reactive Extensions (Rx) for UI event composition. Why: these tools handle complex stream operations that Channels cannot express cleanly.

using Akka.Streams;
using Akka.Streams.Dsl;

// Server-side: batching with timeout
public Source<IReadOnlyList<Event>, NotUsed> BatchEvents(
    Source<Event, NotUsed> events)
{
    return events
        .GroupedWithin(100, TimeSpan.FromSeconds(1))
        .Select(batch => batch.ToList() as IReadOnlyList<Event>);
}

// Client-side: search-as-you-type with Rx
public IObservable<IList<SearchResult>> CreateSearch(
    IObservable<string> searchText, ISearchService searchService)
{
    return searchText
        .Throttle(TimeSpan.FromMilliseconds(300))
        .DistinctUntilChanged()
        .Where(text => text.Length >= 3)
        .SelectMany(text => searchService.SearchAsync(text).ToObservable());
}

See references/stream-patterns.md for throttling, parallel processing, complex pipelines, and Rx examples.

Values: 温故知新（Rx の成熟したイベント合成技術と Akka.NET の分散ストリーム処理を、用途に応じて使い分ける）

Step 5: Manage Stateful Entities with Akka.NET Actors

Apply Akka.NET Actors when you need isolated state per entity, state machines with Become(), or distributed entity management. Why: actors provide natural isolation — each entity gets its own actor with no locks needed.

// Entity-per-actor: each order has isolated state
public class OrderActor : ReceiveActor
{
    private OrderState _state;

    public OrderActor(string orderId)
    {
        _state = new OrderState(orderId);

        Receive<AddItem>(msg =>
        {
            _state = _state.AddItem(msg.Item);
            Sender.Tell(new ItemAdded(msg.Item));
        });

        Receive<Checkout>(msg =>
        {
            if (_state.CanCheckout)
            {
                _state = _state.Checkout();
                Sender.Tell(new CheckoutSucceeded(_state.Total));
            }
            else
            {
                Sender.Tell(new CheckoutFailed("Cart is empty"));
            }
        });

        Receive<GetState>(_ => Sender.Tell(_state));
    }
}

The Actor Mindset: Use when you have thousands of entities needing independent state, lifecycle behaviors, push-based updates, or fault isolation. If none of these apply, use simpler tools.

See references/actor-patterns.md for state machines with Become, Cluster Sharding, and async local functions in actors.

Values: 余白の設計（Actor の隔離境界が、各エンティティに独立した変更・障害の余白を与える）

Good Practices

• Use async/await as the default concurrency tool for I/O operations
• Apply CancellationToken in every async method signature
• Use ConfigureAwait(false) in library code to avoid deadlocks
• Apply bounded Channel<T> for producer/consumer with backpressure
• Use async local functions instead of Task.Run(async () => ...) for clarity
• Avoid raw List<T> in parallel code; use ConcurrentBag<T> or ConcurrentDictionary<K,V>
• Use Rx for UI events (debounce, throttle) and Akka.NET Streams for server-side
• Consider actors when entities need independent state, lifecycle, or fault isolation

Common Pitfalls

1. Blocking on async — Calling .Result or .Wait() causes deadlocks. Use async all the way through the call chain.
2. Unbounded channels — Using Channel.CreateUnbounded<T>() without considering memory. Use bounded channels with BoundedChannelFullMode.Wait.
3. Over-engineering — Reaching for Actors or Streams when async/await would suffice. Start simple and escalate only with concrete need.
4. Manual thread creation — Using new Thread() instead of Task.Run or higher abstractions. Let the thread pool manage threads.
5. Shared mutable state in parallel loops — Using List<T>.Add() inside Parallel.ForEachAsync. Use ConcurrentBag<T> instead.
6. ContinueWith chains — Using .ContinueWith() for sequencing instead of await. Async local functions are clearer and safer.

Anti-Patterns

❌ Locks for Business Logic → ✅ Actor or Channel

// ❌ BAD — locks to protect shared state
private readonly object _lock = new();
private Dictionary<string, Order> _orders = new();
public void UpdateOrder(string id, Action<Order> update)
{
    lock (_lock) { if (_orders.TryGetValue(id, out var order)) update(order); }
}
// ✅ GOOD — actor per entity, no locks needed
// Each order gets its own OrderActor with isolated state

❌ Blocking in Async Code → ✅ Async All the Way

// ❌ BAD — deadlock risk
var result = GetDataAsync().Result;
// ✅ GOOD — async all the way
var result = await GetDataAsync();

❌ Shared Mutable Collections → ✅ Thread-Safe Alternatives

// ❌ BAD — race condition in parallel loop
var results = new List<Result>();
await Parallel.ForEachAsync(items, async (item, ct) =>
{
    results.Add(await ProcessAsync(item, ct)); // Race condition!
});
// ✅ GOOD — use ConcurrentBag
var results = new ConcurrentBag<Result>();

❌ Manual Threads → ✅ Task-Based Abstractions

// ❌ BAD — manual thread management
var thread = new Thread(() => ProcessOrders());
thread.Start();
// ✅ GOOD — task-based abstraction
_ = Task.Run(() => ProcessOrdersAsync(ct));

❌ Anonymous Async Lambda → ✅ Async Local Function

// ❌ BAD — anonymous lambda obscures intent
_ = Task.Run(async () => { var r = await DoWorkAsync(); return new WorkCompleted(r); }).PipeTo(Self);
// ✅ GOOD — named async local function
async Task<WorkCompleted> ExecuteAsync()
{
    var result = await DoWorkAsync();
    return new WorkCompleted(result);
}
ExecuteAsync().PipeTo(Self);

Quick Reference

Decision Tree

What are you trying to do?
│
├─► Wait for I/O?                    → async/await
├─► Process collection in parallel?  → Parallel.ForEachAsync
├─► Producer/consumer work queue?    → Channel<T>
├─► UI event composition?            → Reactive Extensions (Rx)
├─► Server-side stream processing?   → Akka.NET Streams
├─► State machines / entity state?   → Akka.NET Actors
├─► Fire multiple async ops?         → Task.WhenAll
├─► Race multiple async ops?         → Task.WhenAny
└─► Periodic work?                   → PeriodicTimer

Which Tool When

Escalation Path

async/await (start here)
    ├─► Need parallelism?            → Parallel.ForEachAsync
    ├─► Need producer/consumer?      → Channel<T>
    ├─► Need UI event composition?   → Reactive Extensions
    ├─► Need server-side streaming?  → Akka.NET Streams
    └─► Need stateful entities?      → Akka.NET Actors

Resources

• Async/Await Best Practices
• System.Threading.Channels
• Akka.NET Documentation
• Reactive Extensions
• references/channel-patterns.md — Multi-consumer pools, advanced Channel patterns
• references/stream-patterns.md — Akka.NET Streams pipelines, Rx UI patterns
• references/actor-patterns.md — Become state machines, Cluster Sharding, async local functions