dotnet-resilience（点网弹性）

dotnet-resilience

Modern resilience patterns for.NET applications using Polly v8 and Microsoft.Extensions.Http.Resilience. Covers the standard resilience pipeline (rate limiter, total timeout, retry, circuit breaker, attempt timeout), custom pipeline configuration, and integration with the.NET dependency injection system.

Superseded package: Microsoft.Extensions.Http.Polly is superseded by Microsoft.Extensions.Http.Resilience. Do not use Microsoft.Extensions.Http.Polly for new projects. See the migration guide for upgrading existing code.

Out of scope: DI container mechanics and service lifetimes -- see [skill:dotnet-csharp-dependency-injection]. Async/await patterns and cancellation token propagation -- see [skill:dotnet-csharp-async-patterns]. HTTP client factory patterns (typed clients, named clients, DelegatingHandlers) are covered in [skill:dotnet-http-client]. Testing resilience policies -- see [skill:dotnet-integration-testing] for testing with WebApplicationFactory and [skill:dotnet-xunit] for unit testing resilience handlers.

Cross-references: [skill:dotnet-csharp-dependency-injection] for service registration, [skill:dotnet-csharp-async-patterns] for cancellation token propagation, [skill:dotnet-http-client] for applying resilience to HTTP clients.

Package Landscape

Install the modern stack:

<PackageReference Include="Microsoft.Extensions.Http.Resilience" Version="9.*" />
<!-- Transitively brings in Polly v8 and Microsoft.Extensions.Resilience -->

For non-HTTP scenarios only:

<PackageReference Include="Microsoft.Extensions.Resilience" Version="9.*" />

Standard Resilience Pipeline

Microsoft.Extensions.Http.Resilience provides a standard resilience pipeline that follows the recommended order. The pipeline layers execute from outermost to innermost:

Request
  --> Rate Limiter        (1. shed excess load)
    --> Total Timeout      (2. cap total wall-clock time)
      --> Retry             (3. retry transient failures)
        --> Circuit Breaker  (4. stop calling failing services)
          --> Attempt Timeout (5. cap individual attempt time)
            --> HTTP call

Why This Order Matters

• Rate limiter first: prevents retry storms from overwhelming downstream services
• Total timeout wraps retry: ensures the entire operation (including all retries) has a deadline
• Retry wraps circuit breaker: retries can try again after the breaker resets; a broken circuit counts as a retriable failure
• Circuit breaker wraps attempt timeout: timed-out attempts count toward the breaker's failure threshold
• Attempt timeout innermost: each individual HTTP call has its own deadline

Standard Pipeline with Defaults

builder.Services
    .AddHttpClient("catalog-api", client =>
    {
        client.BaseAddress = new Uri("https://catalog.internal");
    })
    .AddStandardResilienceHandler();

This applies the standard pipeline with sensible defaults:

• Rate limiter: 1000 concurrent requests
• Total timeout: 30 seconds
• Retry: 3 attempts, exponential backoff (2s base), jitter
• Circuit breaker: 10% failure ratio, 100 sample size, 5s break duration
• Attempt timeout: 10 seconds

Standard Pipeline with Custom Options

builder.Services
    .AddHttpClient("catalog-api", client =>
    {
        client.BaseAddress = new Uri("https://catalog.internal");
    })
    .AddStandardResilienceHandler(options =>
    {
        // Total timeout for the entire operation including retries
        options.TotalRequestTimeout.Timeout = TimeSpan.FromSeconds(60);

        // Retry strategy
        options.Retry.MaxRetryAttempts = 5;
        options.Retry.Delay = TimeSpan.FromSeconds(1);
        options.Retry.BackoffType = DelayBackoffType.Exponential;
        options.Retry.UseJitter = true;
        options.Retry.ShouldHandle = args => ValueTask.FromResult(
            args.Outcome.Result?.StatusCode is
                HttpStatusCode.RequestTimeout or
                HttpStatusCode.TooManyRequests or
                >= HttpStatusCode.InternalServerError);

        // Circuit breaker
        options.CircuitBreaker.SamplingDuration = TimeSpan.FromSeconds(30);
        options.CircuitBreaker.FailureRatio = 0.1;
        options.CircuitBreaker.MinimumThroughput = 20;
        options.CircuitBreaker.BreakDuration = TimeSpan.FromSeconds(10);

        // Per-attempt timeout
        options.AttemptTimeout.Timeout = TimeSpan.FromSeconds(5);
    });

Configuration via appsettings.json

Bind resilience options from configuration for environment-specific tuning:

builder.Services
    .AddHttpClient("catalog-api", client =>
    {
        client.BaseAddress = new Uri("https://catalog.internal");
    })
    .AddStandardResilienceHandler(options =>
    {
        builder.Configuration
            .GetSection("Resilience:CatalogApi")
            .Bind(options);
    });

{
  "Resilience": {
    "CatalogApi": {
      "Retry": {
        "MaxRetryAttempts": 5,
        "Delay": "00:00:02",
        "BackoffType": "Exponential"
      },
      "CircuitBreaker": {
        "BreakDuration": "00:00:15"
      },
      "TotalRequestTimeout": {
        "Timeout": "00:01:00"
      }
    }
  }
}

Custom Resilience Pipelines

When the standard pipeline does not fit, build custom pipelines with Polly v8 directly.

Retry Strategy

builder.Services.AddResiliencePipeline("db-retry", pipelineBuilder =>
{
    pipelineBuilder.AddRetry(new RetryStrategyOptions
    {
        MaxRetryAttempts = 3,
        Delay = TimeSpan.FromMilliseconds(500),
        BackoffType = DelayBackoffType.Exponential,
        UseJitter = true,
        ShouldHandle = new PredicateBuilder()
            .Handle<DbUpdateConcurrencyException>()
            .Handle<TimeoutException>(),
        OnRetry = args =>
        {
            // Structured logging of retry attempts
            var logger = args.Context.Properties
                .GetValue(new ResiliencePropertyKey<ILogger>("logger"), null!);
            logger?.LogWarning(
                args.Outcome.Exception,
                "Retry attempt {AttemptNumber} after {Delay}ms",
                args.AttemptNumber,
                args.RetryDelay.TotalMilliseconds);
            return ValueTask.CompletedTask;
        }
    });
});

// Inject and use
public sealed class OrderRepository(
    [FromKeyedServices("db-retry")] ResiliencePipeline pipeline,
    AppDbContext db)
{
    public async Task<Order> UpdateAsync(Order order, CancellationToken ct)
    {
        return await pipeline.ExecuteAsync(async token =>
        {
            db.Orders.Update(order);
            await db.SaveChangesAsync(token);
            return order;
        }, ct);
    }
}

Circuit Breaker Strategy

builder.Services.AddResiliencePipeline("payment-gateway", pipelineBuilder =>
{
    pipelineBuilder.AddCircuitBreaker(new CircuitBreakerStrategyOptions
    {
        SamplingDuration = TimeSpan.FromSeconds(30),
        FailureRatio = 0.25,           // Open after 25% failure rate
        MinimumThroughput = 10,        // Need at least 10 calls to evaluate
        BreakDuration = TimeSpan.FromSeconds(15),
        ShouldHandle = new PredicateBuilder()
            .Handle<HttpRequestException>()
            .Handle<TimeoutException>()
    });
});

Timeout Strategy

builder.Services.AddResiliencePipeline("external-api", pipelineBuilder =>
{
    // Total timeout for the entire pipeline execution
    pipelineBuilder.AddTimeout(new TimeoutStrategyOptions
    {
        Timeout = TimeSpan.FromSeconds(30),
        OnTimeout = args =>
        {
            // Log timeout details for diagnostics
            return ValueTask.CompletedTask;
        }
    });
});

Composing Multiple Strategies

Build a composite pipeline by chaining strategies. Order matters -- outermost strategy is added first:

builder.Services.AddResiliencePipeline("composed", pipelineBuilder =>
{
    // 1. Total timeout (outermost -- caps entire operation)
    pipelineBuilder.AddTimeout(new TimeoutStrategyOptions
    {
        Timeout = TimeSpan.FromSeconds(45)
    });

    // 2. Retry (retries on transient failures)
    pipelineBuilder.AddRetry(new RetryStrategyOptions
    {
        MaxRetryAttempts = 3,
        Delay = TimeSpan.FromSeconds(1),
        BackoffType = DelayBackoffType.Exponential,
        UseJitter = true,
        ShouldHandle = new PredicateBuilder()
            .Handle<HttpRequestException>()
            .Handle<TimeoutException>()
    });

    // 3. Circuit breaker (stops calling failing services)
    pipelineBuilder.AddCircuitBreaker(new CircuitBreakerStrategyOptions
    {
        FailureRatio = 0.1,
        MinimumThroughput = 20,
        SamplingDuration = TimeSpan.FromSeconds(30),
        BreakDuration = TimeSpan.FromSeconds(10),
        ShouldHandle = new PredicateBuilder()
            .Handle<HttpRequestException>()
            .Handle<TimeoutException>()
    });

    // 4. Attempt timeout (innermost -- caps single attempt)
    pipelineBuilder.AddTimeout(new TimeoutStrategyOptions
    {
        Timeout = TimeSpan.FromSeconds(10)
    });
});

Typed Resilience Pipelines

For result-bearing operations, use ResiliencePipeline<T>:

builder.Services.AddResiliencePipeline<string, HttpResponseMessage>(
    "typed-http",
    pipelineBuilder =>
    {
        pipelineBuilder.AddRetry(new RetryStrategyOptions<HttpResponseMessage>
        {
            MaxRetryAttempts = 3,
            Delay = TimeSpan.FromSeconds(1),
            BackoffType = DelayBackoffType.Exponential,
            ShouldHandle = new PredicateBuilder<HttpResponseMessage>()
                .Handle<HttpRequestException>()
                .HandleResult(r => r.StatusCode >= HttpStatusCode.InternalServerError)
        });
    });

Hedging Strategy

Send parallel requests to reduce tail latency. The hedging strategy dispatches additional attempts if the initial request is slow:

builder.Services
    .AddHttpClient("search-api")
    .AddStandardHedgingHandler(options =>
    {
        options.Hedging.MaxHedgedAttempts = 2;
        options.Hedging.Delay = TimeSpan.FromMilliseconds(500);
        // Hedging sends a parallel request if the first hasn't
        // responded within 500ms
    });

Use hedging when:

• Operations are idempotent (GET requests, read-only queries)
• Tail latency reduction matters more than extra load

Do not use hedging when:

• Operations have side effects (POST, PUT, DELETE)
• Downstream services cannot handle increased load

Telemetry and Observability

Polly v8 emits metrics and traces via System.Diagnostics out of the box when using the DI integration.

Built-in Metrics

The Microsoft.Extensions.Resilience package automatically reports:

These integrate with OpenTelemetry automatically when the OpenTelemetry SDK is configured in your application -- see [skill:dotnet-observability] for collector setup.

Enabling Telemetry

Resilience telemetry is enabled automatically when using the DI-based registration (AddResiliencePipeline, AddStandardResilienceHandler). The Microsoft.Extensions.Resilience package registers a MeteringEnricher and LoggingEnricher that emit structured logs and metrics through the standard ILoggerFactory and IMeterFactory from DI:

// Telemetry is automatic -- no extra configuration needed.
// Structured logs appear via ILogger; metrics via IMeter.
builder.Services
    .AddHttpClient("catalog-api")
    .AddStandardResilienceHandler();

// To see resilience logs, set the Polly category to Information:
// appsettings.json
// {
//   "Logging": {
//     "LogLevel": {
//       "Polly": "Information"
//     }
//   }
// }

Migrating from Microsoft.Extensions.Http.Polly

If upgrading from the superseded Microsoft.Extensions.Http.Polly package:

Before (Legacy)

// Using Microsoft.Extensions.Http.Polly (superseded)
builder.Services
    .AddHttpClient("catalog-api")
    .AddTransientHttpErrorPolicy(p =>
        p.WaitAndRetryAsync(3, attempt =>
            TimeSpan.FromSeconds(Math.Pow(2, attempt))))
    .AddTransientHttpErrorPolicy(p =>
        p.CircuitBreakerAsync(5, TimeSpan.FromSeconds(30)));

After (Modern)

// Using Microsoft.Extensions.Http.Resilience (current)
builder.Services
    .AddHttpClient("catalog-api")
    .AddStandardResilienceHandler(options =>
    {
        options.Retry.MaxRetryAttempts = 3;
        options.Retry.Delay = TimeSpan.FromSeconds(2);
        options.Retry.BackoffType = DelayBackoffType.Exponential;

        options.CircuitBreaker.MinimumThroughput = 5;
        options.CircuitBreaker.BreakDuration = TimeSpan.FromSeconds(30);
    });

Migration Steps

1. Replace Microsoft.Extensions.Http.Polly NuGet reference with Microsoft.Extensions.Http.Resilience
2. Replace AddTransientHttpErrorPolicy calls with AddStandardResilienceHandler or custom pipeline
3. Translate Polly v7 policy configuration to v8 strategy options
4. Remove explicit IAsyncPolicy<HttpResponseMessage> type references
5. Verify retry/circuit breaker behavior matches previous configuration

Key Principles

• Use the standard pipeline for HTTP -- AddStandardResilienceHandler() covers the most common case with battle-tested defaults
• Customize via options, not custom code -- the standard handler options cover retry count, backoff, circuit breaker thresholds, and timeouts
• Use AddResiliencePipeline for non-HTTP -- database calls, message queues, file I/O
• Always add jitter to retries -- prevents thundering herd when multiple clients retry simultaneously
• Configure via appsettings.json -- different environments need different thresholds; avoid hardcoded values in production
• Do not suppress TimeoutRejectedException -- let it propagate through the pipeline so outer strategies (retry, circuit breaker) can react
• Do not use Microsoft.Extensions.Http.Polly for new projects -- it wraps Polly v7 and is superseded

Agent Gotchas

1. Do not mix Polly v7 and v8 APIs -- v8 uses ResiliencePipeline and strategy options; v7 uses IAsyncPolicy. They are not interchangeable.
2. Do not add AddStandardResilienceHandler twice -- it composes a full pipeline; adding it twice doubles every strategy layer.
3. Do not set attempt timeout higher than total timeout -- the attempt timeout must be shorter than the total timeout or it has no effect.
4. Do not retry non-idempotent operations -- only retry operations that are safe to repeat (GETs, idempotent writes with idempotency keys).
5. Do not use hardcoded retry delays in production -- always use UseJitter = true and configure delays via appsettings.

References

• Polly v8 documentation
• Microsoft.Extensions.Http.Resilience
• Microsoft.Extensions.Resilience
• Migration from Http.Polly to Http.Resilience
• Standard resilience pipeline
• Polly v8 strategy options