📌 Typical scenario

Two (or more) gateways behind a load balancer need to record operations to the same stream.
Lockfree.EventStore ensures order and integrity even under high parallelism, without relying on locks, keeping all state in memory.

📚 Complete documentation

Below is the full technical documentation of the API, advanced features, benchmarks, and usage examples.

Key Features

• Lock-free MPMC writes with FIFO discard
• Key-based partitioning for high concurrency
• Consistent snapshots without blocking producers
• Functional aggregations and time-window queries
• Zero external dependencies, ready for AOT/Trimming
• Fluent API for advanced configuration
• Built-in metrics and observability
• Specialized aggregations (Sum, Average, Min, Max)

Basic Usage Example

var store = new EventStore<Pedido>();
store.TryAppend(new Pedido { Id = 1, Valor = 10m, Timestamp = DateTime.UtcNow });

var total = store.Aggregate(() => 0m, (acc, e) => acc + e.Valor,
    from: DateTime.UtcNow.AddMinutes(-10));

New Constructors

// Explicit capacity
var store = new EventStore<Pedido>(capacidade: 100_000);

// Capacity and partitions
var store = new EventStore<Pedido>(capacidade: 50_000, particoes: 8);

// Advanced configuration
var store = new EventStore<Pedido>(new EventStoreOptions<Pedido>
{
    Capacidade = 100_000,
    Particoes = 16,
    OnEventDiscarded = evt => Logger.LogTrace("Event discarded: {Event}", evt),
    OnCapacityReached = () => Metrics.IncrementCounter("eventstore.capacity_reached"),
    TimestampSelector = new PedidoTimestampSelector()
});

// Fluent API
var store = EventStore.For<Pedido>()
    .WithCapacity(100_000)
    .WithPartitions(8)
    .OnDiscarded(evt => Log(evt))
    .OnCapacityReached(() => NotifyAdmin())
    .WithTimestampSelector(new PedidoTimestampSelector())
    .Create();

State Properties

store.Count          // Current number of events
store.Capacity       // Maximum configured capacity
store.IsEmpty        // Whether it's empty
store.IsFull         // Whether it reached maximum capacity
store.Partitions     // Number of partitions

Specialized Aggregations

// Count by time window
var count = store.Count(from: inicio, to: fim);

// Sum of values
var sum = store.Sum(evt => evt.Valor, from: inicio, to: fim);

// Average
var avg = store.Average(evt => evt.Valor, from: inicio, to: fim);

// Min and max
var min = store.Min(evt => evt.Pontuacao, from: inicio, to: fim);
var max = store.Max(evt => evt.Pontuacao, from: inicio, to: fim);

// With filters
var filteredSum = store.Sum(
    evt => evt.Valor, 
    filter: evt => evt.Tipo == "Pagamento",
    from: inicio, 
    to: fim
);

Snapshots with Filters

// Filtered snapshot
var eventosRecentes = store.Snapshot(
    filter: evt => evt.Timestamp > DateTime.UtcNow.AddMinutes(-5)
);

// Snapshot by time window
var snapshot = store.Snapshot(from: inicio, to: fim);

// Snapshot with filter and time window
var filtrado = store.Snapshot(
    filter: evt => evt.Valor > 100,
    from: inicio,
    to: fim
);

Cleanup and Maintenance

// Clear all events
store.Clear();
store.Reset(); // Alias for Clear()

// Purge old events (requires TimestampSelector)
store.Purge(olderThan: DateTime.UtcNow.AddHours(-1));

Metrics and Observability

// Detailed statistics
store.Statistics.TotalAppended        // Total appended events
store.Statistics.TotalDiscarded       // Total discarded events
store.Statistics.AppendsPerSecond     // Current append rate
store.Statistics.LastAppendTime       // Timestamp of the last append

Examples

MetricsDashboard

Fully featured web API for collecting and querying real-time metrics:

cd .\samples\MetricsDashboard
 dotnet run

Available endpoints:

• POST /metrics - Add metric
• GET /metrics/sum?label=cpu_usage - Sum values by label
• GET /metrics/top?k=5 - Top K metrics

See samples/MetricsDashboard/TESTING.md for a complete testing guide.

Full API

• TryAppend(event) — Adds an event, lock-free
• Aggregate — Aggregates values by time window
• Snapshot() — Returns an immutable copy of events
• Count/Sum/Average/Min/Max — Specialized aggregations
• Clear/Reset/Purge — Cleanup methods
• Query — Flexible queries with filters
• Statistics — Monitoring metrics

Partitions

The default number of partitions is Environment.ProcessorCount. You can force the partition using TryAppend(e, partition).

Snapshots

Snapshot() returns an approximate immutable copy of the current state of all partitions, ordered from the oldest to the newest event per partition.

Performance

Designed for high concurrency and low latency. Global order across partitions is approximate.

Performance Benchmarks

Value Types vs Reference Types

Structure of Arrays (SoA) vs Array of Structures (AoS)

Conclusions:

1. Value types are significantly faster than reference types for reads and writes.
2. SoA improves cache locality and reduces memory pressure.
3. For high throughput, the EventStoreV2 implementation is recommended.

// Using EventStoreV2 with value types
var store = new EventStoreV2(capacidade: 1_000_000, particoes: 16);
store.Add("sensor1", 25.5, DateTime.UtcNow.Ticks);
double media = store.Average("sensor1");

Limitations

• Global order is only approximate across partitions
• Fixed capacity; old events are discarded when exceeded

License

MIT