PicoLog.Abs 2026.9.3

PicoLog

High-performance structured logging for .NET Native AOT.

English | 简体中文 | 日本語 | Español | Português | 繁體中文 | 한국어 | Français | Deutsch | Русский

Performance

PicoLog vs Microsoft.Extensions.Logging on .NET 10.0.5 (Windows 10, X64, Release).

Summary: 13/30 wins, average 3.51x speedup, max 25.76x.

PicoLog is faster than MEL on cached-message and allocation-heavy paths where source generation eliminates boxing and reflection. On single-message throughput (N=1), PicoLog is deliberately slower — roughly 700ns vs MEL's 200ns per call. This is not a bug; it's the cost of an async-first pipeline architecture.

Design Rationale: Async-First Pipeline

Every Log() call traverses the same pipeline regardless of whether you call the synchronous or asynchronous overload:

Log() → CreateEntry → CategoryPipeline → Channel → ProcessEntriesAsync → DispatchEntryAsync → Sink

This is fundamentally different from MEL's synchronous model:

Log() → Logger.Log() → foreach (logger in loggers) → logger.Log()  // direct, no channel

PicoLog chose the async pipeline because production logging systems need guaranteed delivery, backpressure, and graceful shutdown more than they need the absolute lowest single-message latency. If you're logging at 1M msg/s, the per-message overhead dominates. If you're logging at 1K msg/s with flush semantics, PicoLog's guarantees matter more.

For direct throughput comparisons across all scenarios, see benchmark results.

Quick Start

dotnet add package PicoLog

using PicoLog;
using PicoLog.Abs;

var sink = new ColoredConsoleSink(new ConsoleFormatter());
using var factory = new LoggerFactory([sink],
    new LoggerFactoryOptions { MinLevel = LogLevel.Info });
var logger = factory.CreateLogger("App");
logger.Info("Application started");

Log Levels

Core API

LoggerFactory

using var factory = new LoggerFactory(
    sinks: [new ColoredConsoleSink(new ConsoleFormatter())],
    options: new LoggerFactoryOptions
    {
        MinLevel = LogLevel.Info,
        QueueCapacity = 65535,
        QueueFullMode = LogQueueFullMode.DropOldest,
        SyncWriteTimeout = TimeSpan.FromMilliseconds(250),
        ShutdownTimeout = TimeSpan.FromSeconds(5)
    });

ILogger logger = factory.CreateLogger("MyComponent");

Typed Logger

public sealed class OrderService(ILoggerFactory factory)
{
    private readonly ILogger<OrderService> _logger = new Logger<OrderService>(factory);

    public void Process(Order order)
    {
        _logger.Info($"Processing order {order.Id}");
    }
}

Extension Methods

Level-specific convenience helpers provided as extension methods on ILogger. Not part of the interface contract — consumers only implement the 8 core interface methods.

logger.Trace("Detailed diagnostic");
logger.Debug("Debug information");
logger.Info("Application event");
logger.Notice("Significant event");
logger.Warning("Unexpected condition");
logger.Error("Operation failed", exception);
logger.Critical("Severe failure");
logger.Alert("Immediate action required");
logger.Emergency("System is down");

Each has a FormattableString overload for deferred formatting and an EventId overload for structured event identification. Async variants (TraceAsync, InfoAsync, etc.) are available with CancellationToken support.

Message Templates

// FormattableString — template and arguments preserved for deferred formatting
logger.Info($"Processing order {orderId} for {customer}");

// Structured properties — explicit key-value pairs
logger.LogStructured(LogLevel.Error, "Payment failed",
    properties: [new("OrderId", orderId), new("Amount", amount)],
    exception: ex);

// EventId — tagged for monitoring and alerting
logger.Info(new EventId(1001, "OrderPlaced"), $"Order {orderId} placed");

Source-Generated Messages

The [PicoLogMessage] generator is embedded in PicoLog.Abs (no extra package). Reference PicoLog.Abs and the analyzer activates automatically:

public static partial class AppLogs
{
    [PicoLogMessage(LogLevel.Info, EventId = 1001, Message = "Order {OrderId} placed")]
    public static partial void OrderPlaced(this ILogger logger, string orderId);

    [PicoLogMessage(LogLevel.Error, EventId = 2001,
        EventName = "PaymentFailed", Message = "Payment of {Amount} failed")]
    public static partial void PaymentFailed(this ILogger logger, decimal amount);
}

logger.OrderPlaced("ORD-12345");
logger.PaymentFailed(99.99m);

The generator emits Log calls with FormattableString at compile time — zero runtime reflection.

Sinks

Built-in

// Plain console
new ConsoleSink(new ConsoleFormatter())

// Color-coded console (Trace=Gray, Debug=Cyan, Info=Green, Error=Red, ...)
new ColoredConsoleSink(new ConsoleFormatter())

// File with batching
new FileSink(new ConsoleFormatter(), new FileSinkOptions
{
    FilePath = "logs/app.log",
    BatchSize = 32,
    QueueCapacity = 4096,
    FlushInterval = TimeSpan.FromMilliseconds(100)
})

Custom

public sealed class CustomSink : ILogSink
{
    public Task WriteAsync(LogEntry entry, CancellationToken ct = default)
    {
        // Write to your backend
        return Task.CompletedTask;
    }
    public ValueTask DisposeAsync() => ValueTask.CompletedTask;
}

Implement IFlushableLogSink to support explicit flush:

public sealed class BufferedSink : IFlushableLogSink
{
    public Task WriteAsync(LogEntry entry, CancellationToken ct = default) { /* buffer */ }
    public ValueTask FlushAsync(CancellationToken ct = default) { /* flush buffer */ }
    public ValueTask DisposeAsync() => ValueTask.CompletedTask;
}

Custom Formatter

public sealed class JsonFormatter : ILogFormatter
{
    public string Format(LogEntry entry)
    {
        return JsonSerializer.Serialize(new
        {
            entry.Timestamp,
            Level = entry.Level.ToString(),
            entry.Category,
            entry.Message,
            entry.Exception?.Message
        });
    }
}

Logging Scopes

using (logger.BeginScope(new { RequestId = "abc-123" }))
{
    logger.Info("Processing request");
    // Logs include scope context
}

Configuration

Queue Behavior

Three backpressure strategies when the queue is full:

var options = new LoggerFactoryOptions
{
    QueueCapacity = 10000,
    QueueFullMode = LogQueueFullMode.Wait,
    SyncWriteTimeout = TimeSpan.FromMilliseconds(500)
};

Category Filtering

var options = new LoggerFactoryOptions
{
    MinLevel = LogLevel.Info,  // global default
    FilterRules =
    {
        new LogFilterRule("Microsoft", LogLevel.Warning),
        new LogFilterRule("MyApp.Data", LogLevel.Debug)
    }
};

Rules are applied in reverse order — the last matching rule wins.

Dropped Message Notification

var options = new LoggerFactoryOptions
{
    QueueCapacity = 1000,
    OnMessagesDropped = (category, count) =>
        Console.Error.WriteLine($"Dropped {count} messages for {category}")
};

Timestamp Control

var options = new LoggerFactoryOptions
{
    TimestampProvider = TimeProvider.System  // default; inject fake for testing
};

Flush & Shutdown

using var factory = new LoggerFactory(sinks, options);

// Flush without shutdown — drains pending entries, factory remains usable
await factory.FlushAsync();

// Dispose drains all pipelines and sinks, aggregates errors
await factory.DisposeAsync();

Set ShutdownTimeout to bound the drain:

var options = new LoggerFactoryOptions
{
    ShutdownTimeout = TimeSpan.FromSeconds(5)  // TimeSpan.Zero = wait indefinitely
};

On shutdown timeout, sinks receive a drain CancellationToken. The factory waits for the processing pipeline to finish draining before disposing resources.

DI Integration (PicoLog.DI)

dotnet add package PicoLog.DI

using PicoLog.DI;

container.AddPicoLog(o =>
{
    o.MinLevel = LogLevel.Info;
    o.WriteTo.ColoredConsole();
    o.WriteTo.File("logs/app.log");
    o.WriteTo.File(opts =>
    {
        opts.FilePath = "logs/errors.log";
        opts.BatchSize = 1;  // write immediately
    });
    o.Factory.QueueFullMode = LogQueueFullMode.DropOldest;
});

// Resolve typed logger
var logger = container.CreateScope().GetService<ILogger<Program>>();

Consume Registered Sinks

// Register a custom sink in DI, then tell PicoLog to use it
container.RegisterSingleton<ILogSink>(_ => new CustomSink());

container.AddPicoLog(o =>
{
    o.ReadFrom.RegisteredSinks();  // discover ILogSink from container
    o.WriteTo.ColoredConsole();    // additional owned sinks
});

Custom Sink Registration

container.AddPicoLog(o =>
{
    o.WriteTo.Sink(new CustomSink());
    o.WriteTo.Sink(() => new LazySink());         // factory
    o.WriteTo.Sink(fmt => new FormattedSink(fmt)); // formatter-aware factory
});

OpenTelemetry Metrics

PicoLogMetrics exposes standard counters via System.Diagnostics.Metrics:

All metrics are AOT-compatible and integrate with OpenTelemetry collectors.

Packages

← Back to PicoInfra