Lock-free, allocation-conscious, thread-safe time-series primitives for building fast counters, rolling analytics, and metric pipelines in .NET 10 / C# 14.

Package links open the latest stable release on NuGet; badges update automatically.

Package	Latest version	Downloads	Description
ManagedCode.TimeSeries			Thread-safe time-series accumulators, summers, and extensions.
ManagedCode.TimeSeries.Orleans			Orleans serializers and surrogates for the core time-series types.

ManagedCode.TimeSeries is a high-performance .NET time series metrics library. It provides optimized, time-bucketed collections to aggregate events into time series:

• Accumulators store raw events per time bucket (queue per bucket).
• Summers store aggregated values per time bucket (sum/min/max/replace).
• Groups manage many series by key (per endpoint, per customer, per shard, etc.).

It's designed for high-throughput analytics: clicks, tokens, CPU usage, money, events, and any numeric telemetry.

• Lock-free ingestion for high write rates.
• Thread-safe reads/writes for multi-threaded pipelines.
• UTC-normalized timestamps for deterministic ordering.
• Grouped series for multi-tenant metrics.
• Orleans-friendly converters and System.Text.Json helpers.

• Web analytics (page views, clicks, conversion funnels).
• Infrastructure metrics (CPU, memory, latency, tokens).
• Billing counters (money, usage, events per window).

• SampleInterval: bucket size (e.g., 1s, 10s, 1m).
• Samples/Buckets: ordered view of buckets keyed by UTC timestamps (Buckets is an alias of Samples).
• MaxSamplesCount: maximum bucket count per series; 0 means unbounded (this is not the event count).
• DataCount: total events processed (not the number of buckets).
• UTC normalized: you can pass any DateTimeOffset, but values are normalized to UTC internally; offsets are not stored.
• Thread-safe: concurrent reads/writes are supported across all public types.

dotnet add package ManagedCode.TimeSeries

using ManagedCode.TimeSeries.Accumulators;

var requests = new IntTimeSeriesAccumulator(
    sampleInterval: TimeSpan.FromSeconds(5),
    maxSamplesCount: 60); // 60 buckets -> 5 minutes at 5s interval

// Use current time (UTC internally)
requests.Record(1);
requests.Record(1);

Console.WriteLine($"Buckets: {requests.Samples.Count}");
Console.WriteLine($"Events: {requests.DataCount}");

using ManagedCode.TimeSeries.Summers;
using ManagedCode.TimeSeries.Extensions;

var latency = new NumberTimeSeriesSummer<decimal>(TimeSpan.FromMilliseconds(500));

latency.Record(12.4m);
latency.Record(9.6m);

Console.WriteLine($"Sum: {latency.Sum()}");
Console.WriteLine($"Avg: {latency.Average():F2}");
Console.WriteLine($"Min/Max: {latency.Min()} / {latency.Max()}");

using ManagedCode.TimeSeries.Accumulators;
using ManagedCode.TimeSeries.Extensions;

var perEndpoint = new IntGroupTimeSeriesAccumulator(
    sampleInterval: TimeSpan.FromSeconds(1),
    maxSamplesCount: 300,
    deleteOverdueSamples: true);

perEndpoint.AddNewData("/home", 1);
perEndpoint.AddNewData("/checkout", 1);

foreach (var (endpoint, accumulator) in perEndpoint.Snapshot())
{
    Console.WriteLine($"{endpoint}: {accumulator.DataCount} hits");
}

using ManagedCode.TimeSeries.Accumulators;
using ManagedCode.TimeSeries.Serialization;

var series = new IntTimeSeriesAccumulator(TimeSpan.FromSeconds(1), maxSamplesCount: 10);
series.AddNewData(1);
series.AddNewData(2);

var json = TimeSeriesJsonSerializer.SerializeAccumulator<int, IntTimeSeriesAccumulator>(series);
var restored = TimeSeriesJsonSerializer.DeserializeAccumulator<int, IntTimeSeriesAccumulator>(json);

Console.WriteLine(restored.DataCount); // same as original

Converters in ManagedCode.TimeSeries.Orleans are marked with [RegisterConverter] and are auto-registered when the assembly is loaded. Reference the package from your silo and client projects so the converters are available. Converters are provided for int/float/double accumulators, summers, and grouped series, plus generic numeric summers.

• Public APIs accept DateTimeOffset so callers can pass any offset they have.
• Internally, timestamps are normalized to UTC (offset zero) for speed and determinism.
• If you need to preserve offsets, store them separately alongside your metric data.
• If you do not need custom timestamps, use Record(value) or AddNewData(value) and let the library use DateTime.UtcNow.

• Keep MaxSamplesCount reasonable to limit memory.
• Use Record(value) (or AddNewData(value)) unless you must pass custom timestamps.
• Use summers for high-volume metrics (they avoid storing every event).
• Avoid locks in code that touches these structures.

• Accumulators use ConcurrentQueue<T> to remain lock-free and thread-safe under concurrent writes.
• Even in single-threaded runtimes, timers and background cleanup can introduce concurrency.
• If you need single-thread-only storage, consider using summers or ask for a dedicated single-thread accumulator.

• See docs/Architecture/Overview.md for detailed workflows, data model, and module boundaries.

dotnet restore ManagedCode.TimeSeries.slnx
dotnet build ManagedCode.TimeSeries.slnx --configuration Release
dotnet format ManagedCode.TimeSeries.slnx
dotnet build ManagedCode.TimeSeries.slnx --configuration Release
dotnet test ManagedCode.TimeSeries.Tests/ManagedCode.TimeSeries.Tests.csproj --configuration Release

1. Restore/build/test using the commands above.
2. Keep new APIs covered with tests (see existing samples in ManagedCode.TimeSeries.Tests).
3. Keep hot paths lock-free; only introduce locking when unavoidable and document the trade-off.
4. Update this README if you change behavior or public APIs.

MIT (c) ManagedCode SAS.