BareWire

High-performance async messaging library for .NET 10 / C# 14.

An alternative to MassTransit with a fundamentally different architecture: raw-first (no envelope by default), zero-copy pipeline, manual topology, and deterministic memory usage.

Key Design Principles

• Raw-first — default serializer produces raw JSON, no envelope. Envelope format is opt-in.
• Manual topology — ConfigureConsumeTopology = false by default. Auto-topology is opt-in.
• Zero-copy pipeline — IBufferWriter<byte> / ReadOnlySequence<byte> with ArrayPool. No byte[] allocation per-message in hot paths.
• Credit-based flow control — bounded channels, atomic inflight tracking, health alerts at 90% capacity.
• Familiar API — uses MassTransit naming conventions (IBus, IConsumer<T>, ConsumeContext<T>) for easy migration.

Packages

Package	Description
BareWire.Abstractions	Public interfaces, zero dependencies
BareWire	Core engine, pipeline, bus implementation
BareWire.Serialization.Json	JSON serializer (System.Text.Json)
BareWire.Serialization.MsgPack	MessagePack serializer (compact binary)
BareWire.CloudEvents	CloudEvents 1.0 envelope (binary + structured)
BareWire.Transport.RabbitMQ	RabbitMQ transport
BareWire.Transport.Kafka	Kafka transport
BareWire.Transport.AzureServiceBus	Azure Service Bus transport
BareWire.Transport.AWS.SQS	AWS SQS transport
BareWire.Transport.Google.PubSub	Google Pub/Sub transport
BareWire.Saga	SAGA state machine
BareWire.Saga.Redis	Redis SAGA persistence
BareWire.Outbox	Outbox/Inbox pattern
BareWire.Observability	OpenTelemetry integration
BareWire.Testing	In-memory test harness

Each transport also ships a single-call bundle package — BareWire.RabbitMQ, BareWire.Kafka, BareWire.AzureServiceBus, BareWire.AWS.SQS, BareWire.Google.PubSub — that registers the core and the transport in one AddBareWireWith{Transport} call.

Build & Test

# Build
dotnet build BareWire.slnx

# All tests
dotnet test BareWire.slnx

# Unit tests only
dotnet test tests/BareWire.UnitTests/

# Benchmarks
dotnet run --project tests/BareWire.Benchmarks/ -c Release -- --filter '*Publish*'

Performance Targets

• < 768 B/msg publish allocation, < 256 B/msg consume allocation
• > 500K msgs/s publish throughput
• > 300K msgs/s consume throughput (in-memory transport)

Allocation characteristics

• PublishRaw — constant 136 B regardless of payload size (100 B → 10 KB, no difference). Pre-serialized ReadOnlyMemory<byte> is passed through without copying.
• PublishTyped — ~544 B fixed overhead + serialized payload size. The serialization boundary copy (.ToArray() in MessagePipeline.ProcessOutboundAsync) is architecturally required — OutboundMessage must outlive the pooled writer scope.
• Serialization (raw) — constant 448 B regardless of payload size. PooledBufferWriter rents from ArrayPool<byte>.Shared, confirming ADR-003 zero-copy pipeline.

Full report: docs/articles/benchmark-report.md

License

MIT. See LICENSE.