RabbitMQ.ClientKit.ChannelPooling 1.1.0

RabbitMQ.ClientKit

RabbitMQ.ClientKit is a small RabbitMQ client toolkit aimed at the common cases:

• publishing strongly typed messages with JSON serialization
• declaring queue or exchange topology when needed
• starting async consumers with explicit ack, reject, or requeue behavior
• swapping channel-management strategies without changing producer or consumer code

The current packaged release line targets .NET 8 and uses independent SemVer.

Packages

Package	Purpose
RabbitMQ.ClientKit	Core connection, publishing, consuming, serialization, and transient channel creation
RabbitMQ.ClientKit.ChannelPooling	Reusable producer and consumer channel strategies built on the core abstractions
RabbitMQ.ClientKit.DynamicConfiguration	Optional reloadable configuration support for appsettings, databases, and other dynamic sources

Core usage

using RabbitMQ.ClientKit;
using RabbitMQ.ClientKit.Configuration;
using RabbitMQ.ClientKit.Models;

var connection = new RabbitMqConnectionOptions
{
    HostName = "localhost",
    UserName = "guest",
    Password = "guest",
    ClientProvidedName = "orders-api"
};

await using var client = new RabbitMqClient(connection);

await client.PublishAsync(
    new { OrderId = 42, Status = "Created" },
    new RabbitMqPublishOptions
    {
        QueueName = "orders.created",
        Topology = new RabbitMqTopologyOptions
        {
            Queue = new RabbitMqQueueOptions
            {
                Name = "orders.created",
                Durable = true
            }
        }
    });

await using var subscription = await client.SubscribeAsync<dynamic>(
    new RabbitMqConsumerOptions
    {
        QueueName = "orders.created",
        Topology = new RabbitMqTopologyOptions
        {
            Queue = new RabbitMqQueueOptions
            {
                Name = "orders.created",
                Durable = true
            }
        }
    },
    async (message, cancellationToken) =>
    {
        Console.WriteLine(message.Payload);
        await Task.CompletedTask;
        return RabbitMqConsumeResult.Ack;
    });

Batch publishing and publisher confirms

RabbitMQ.Client v7+ establishes publisher confirmations at channel creation (CreateChannelOptions). RabbitMQ.ClientKit enables confirm tracking on publisher channels with a configurable timeout (default 30s).

Batch publishing

Batch publishing reuses a single leased channel and publishes each message individually while tracking confirmations.

var result = await client.PublishBatchAsync(
    new[]
    {
        new { OrderId = 42, Status = "Created" },
        new { OrderId = 43, Status = "Created" }
    },
    new RabbitMqPublishOptions
    {
        QueueName = "orders.created",
        PublisherConfirms = true
    });

Batch result status (NotSent vs Unconfirmed)

PublishBatchAsync returns a RabbitMqPublishBatchResult indicating the exact publish outcome:

Status	Meaning	Transaction / Rollback Guidance
Confirmed	All messages published and confirmed by the broker (or empty batch).	Safe to commit.
NotSent	Failure occurred before the batch wrote its first publish frame (e.g. channel lease, topology declaration, or first-message serialization threw).	Safe to roll back database transactions.
Unconfirmed	Failure occurred after at least one frame was written (e.g. publish or confirm timeout threw mid-batch). Contains AttemptedCount and ConfirmedCount.	Do NOT roll back transactions; broker may have received/queued messages.

Raw publishing

You can publish raw byte payloads directly without invoking serializer logic:

ReadOnlyMemory<byte> rawBytes = new byte[] { 0x01, 0x02, 0x03 };

await client.PublishAsync(
    rawBytes,
    new RabbitMqPublishOptions
    {
        QueueName = "events.raw",
        Properties = new RabbitMqMessageProperties
        {
            ContentType = "application/octet-stream"
        }
    });

Replace-then-ack and delivery handles

Consumers receive an IRabbitMqDeliveryHandle on RabbitMqReceivedMessage<T>.Delivery that enables in-place message replacement and manual acknowledgment. When AutoAck = true, Delivery is null because the broker settles the delivery before the handler runs.

await using var subscription = await client.SubscribeAsync<OrderMessage>(
    new RabbitMqConsumerOptions
    {
        QueueName = "orders.input",
        ConsumerName = "orders-processor"
    },
    async (message, cancellationToken) =>
    {
        var replacement = new ProcessedOrder { Id = message.Payload.Id, Status = "Processed" };

        await message.Delivery!.ReplaceAndAckAsync(
            replacement,
            new RabbitMqPublishOptions { QueueName = "orders.output" },
            cancellationToken);

        return RabbitMqConsumeResult.Handled;
    });

Replace-then-ack semantics

ReplaceAndAckAsync implements a confirm-then-ack workflow on the same consumer channel:

1. Publishes the replacement message and waits for the broker confirmation.
2. If the replacement confirm fails or throws: negatively acknowledges (nack) the original message with requeue.
3. If the replacement confirm succeeds: acknowledges (ack) the original message. If the subsequent ack throws, the exception is swallowed and the original is not nacked (preventing duplicate replacements).

When using ReplaceAndAckAsync or manual delivery settlement, return RabbitMqConsumeResult.Handled so the consumer wrapper does not double-ack or double-nack.

Replacement publishes honor the supplied RabbitMqPublishOptions, including Topology, so a handler can declare the destination exchange and queues before republishing.

Confirm-then-ack duplicate window

If a consumer crashes or network connectivity is lost in the brief window between a successful replacement confirm and the broker processing the original message's ack, RabbitMQ will redeliver the original message upon reconnect. Message handlers must be designed to be idempotent.

Topology and multi-queue declarations

Topology declarations can be initialized automatically on publish/consume or executed standalone via DeclareAsync:

var topology = new RabbitMqTopologyOptions
{
    Queue = new RabbitMqQueueOptions { Name = "orders.work", Durable = true },
    Queues =
    [
        new RabbitMqQueueOptions { Name = "orders.failed", Durable = true },
        new RabbitMqQueueOptions { Name = "orders.invalid", Durable = true }
    ]
};

// Declare topology without publishing
await client.DeclareAsync(topology);

Advanced topology

Topology declarations are optional, but when you supply them the toolkit will declare the configured exchange, queue, and bindings before publishing or consuming.

That makes the topology model the place to express broker-native queue and binding features such as:

• durable vs transient exchanges and queues
• auto-delete and exclusive queues
• queue bindings with explicit routing keys
• queue arguments such as TTL, max length, quorum queue settings, and dead-letter routing

Dead-letter queues are configured through normal RabbitMQ queue arguments rather than a special wrapper-specific API. For example:

var topology = new RabbitMqTopologyOptions
{
    Exchange = new RabbitMqExchangeOptions
    {
        Name = "orders",
        Type = "direct",
        Durable = true
    },
    Queue = new RabbitMqQueueOptions
    {
        Name = "orders.primary",
        Durable = true,
        Arguments = new Dictionary<string, object?>
        {
            ["x-dead-letter-exchange"] = "orders.dlx",
            ["x-dead-letter-routing-key"] = "orders.dead",
            ["x-message-ttl"] = 30_000
        }
    },
    Bindings =
    [
        new RabbitMqQueueBindingOptions
        {
            ExchangeName = "orders",
            QueueName = "orders.primary",
            RoutingKey = "orders.created"
        }
    ]
};

Once a queue is configured with dead-lettering, returning RabbitMqConsumeResult.Reject or disabling requeue-on-failure will let RabbitMQ route the message to the broker-configured DLX/DLQ.

Consumer behavior

RabbitMqConsumerOptions gives you the common RabbitMQ consumption controls without forcing you down a single acknowledgment model.

Option	Purpose
QueueName	The queue to consume from
ConsumerName	Stable logical consumer identity used by reusable consumer channel providers
ConsumerTag	Explicit RabbitMQ consumer tag
AutoAck	Let RabbitMQ acknowledge deliveries automatically
PrefetchCount / GlobalPrefetch	Control QoS and delivery batching on the channel
Exclusive	Request an exclusive consumer
NoLocal	Ask the broker not to deliver publications from the same connection
RequeueOnFailure	On unhandled exceptions, choose whether the toolkit nacks with requeue or without requeue
Arguments	Pass broker-specific consumer arguments
Topology	Declare the required exchange/queue/bindings before consuming

The handler result controls normal acknowledgment flow:

• RabbitMqConsumeResult.Ack acknowledges the delivery
• RabbitMqConsumeResult.Reject rejects it without requeue
• RabbitMqConsumeResult.Requeue nacks it and requeues it

Unhandled exceptions are also surfaced to the caller. When AutoAck is false, the toolkit will BasicNack using RequeueOnFailure.

Example with explicit prefetch and dead-letter-friendly failure behavior:

await using var subscription = await client.SubscribeAsync<OrderCreated>(
    new RabbitMqConsumerOptions
    {
        QueueName = "orders.primary",
        ConsumerName = "orders-worker",
        PrefetchCount = 32,
        RequeueOnFailure = false,
        Topology = topology
    },
    async (message, cancellationToken) =>
    {
        if (!await TryProcessAsync(message.Payload, cancellationToken))
        {
            return RabbitMqConsumeResult.Reject;
        }

        return RabbitMqConsumeResult.Ack;
    });

Message properties and message context

Publishing supports the most commonly used AMQP properties through RabbitMqMessageProperties, including:

• Headers
• CorrelationId
• MessageId
• ReplyTo
• Expiration
• Priority
• Type
• UserId
• AppId
• TimestampUtc
• ContentType / ContentEncoding
• Persistent

Example:

await client.PublishAsync(
    new { OrderId = 42, Status = "Created" },
    new RabbitMqPublishOptions
    {
        ExchangeName = "orders",
        RoutingKey = "orders.created",
        Properties = new RabbitMqMessageProperties
        {
            CorrelationId = "order-42",
            MessageId = Guid.NewGuid().ToString("N"),
            ReplyTo = "orders.reply",
            Type = "orders.created",
            Headers = new Dictionary<string, object?>
            {
                ["tenant"] = "acme",
                ["source"] = "orders-api"
            },
            TimestampUtc = DateTimeOffset.UtcNow
        }
    });

On the consume side, each RabbitMqReceivedMessage<T> includes:

• Payload for the deserialized body
• Body for the raw bytes
• Context for broker metadata such as DeliveryTag, Redelivered, Exchange, RoutingKey, CorrelationId, MessageId, ReplyTo, TimestampUtc, and Headers

That lets you keep strongly typed handlers while still inspecting the transport-level metadata when you need it.

Connection tuning

RabbitMqConnectionOptions intentionally exposes the connection settings that most applications end up tuning in production:

Option	Purpose
ConnectionUri	Use a single URI instead of individual host settings
HostName / Port / UserName / Password / VirtualHost	Standard connection parameters
ClientProvidedName	Connection name shown in RabbitMQ management tooling
AutomaticRecoveryEnabled	Reconnect automatically after network interruption
TopologyRecoveryEnabled	Re-declare broker topology after reconnect
ConsumerDispatchConcurrency	Number of concurrent async consumer dispatches per connection
RequestedHeartbeat	Heartbeat interval
RequestedConnectionTimeout	Initial connection timeout
NetworkRecoveryInterval	Retry interval used by automatic recovery

Example:

var connection = new RabbitMqConnectionOptions
{
    ConnectionUri = "amqp://guest:guest@localhost:5672/",
    ClientProvidedName = "orders-api",
    AutomaticRecoveryEnabled = true,
    TopologyRecoveryEnabled = true,
    ConsumerDispatchConcurrency = 8,
    RequestedHeartbeat = TimeSpan.FromSeconds(30),
    RequestedConnectionTimeout = TimeSpan.FromSeconds(15),
    NetworkRecoveryInterval = TimeSpan.FromSeconds(5)
};

Client caching

RabbitMqClientCache provides thread-safe caching and reuse of RabbitMqClient instances keyed by the effective connection configuration (broker identity, credentials, and connection settings), avoiding duplicate connection management overhead when connecting dynamically across multiple services:

await using var cache = new RabbitMqClientCache();

// Resolve or create a client from connection options
var client1 = cache.GetOrCreate(new RabbitMqConnectionOptions
{
    HostName = "rabbit-cluster",
    Port = 5672,
    VirtualHost = "orders"
});

// Or resolve directly by URI
var client2 = cache.GetOrCreate(new Uri("amqp://guest:guest@rabbit-cluster:5672/orders"));

Core DI usage

RabbitMQ.ClientKit now includes IServiceCollection extensions for the built-in transient channel strategy.

using Microsoft.Extensions.DependencyInjection;
using RabbitMQ.ClientKit;
using RabbitMQ.ClientKit.Configuration;

var services = new ServiceCollection();

services.AddRabbitMqClient(new RabbitMqConnectionOptions
{
    HostName = "localhost",
    UserName = "guest",
    Password = "guest",
    ClientProvidedName = "orders-api"
});

You can then inject:

• RabbitMqClient for the convenience facade
• RabbitMqPublisher if you only publish
• RabbitMqConsumer if you only subscribe
• IRabbitMqSerializer, IRabbitMqConnectionManager, and the channel-provider abstractions if you need lower-level control

Example consumer-facing service:

using RabbitMQ.ClientKit;
using RabbitMQ.ClientKit.Configuration;

public sealed class OrderPublisher(RabbitMqClient rabbitMqClient)
{
    public Task PublishCreatedAsync(int orderId, CancellationToken cancellationToken = default) =>
        rabbitMqClient.PublishAsync(
            new { OrderId = orderId, Status = "Created" },
            new RabbitMqPublishOptions
            {
                QueueName = "orders.created",
                Topology = new RabbitMqTopologyOptions
                {
                    Queue = new RabbitMqQueueOptions
                    {
                        Name = "orders.created",
                        Durable = true
                    }
                }
            },
            cancellationToken);
}

Pooling extension usage

using RabbitMQ.ClientKit.ChannelPooling;
using RabbitMQ.ClientKit.Configuration;

var pooledClient = PooledRabbitMqClientFactory.Create(
    new RabbitMqConnectionOptions
    {
        HostName = "localhost",
        ClientProvidedName = "payments-api"
    },
    new RabbitMqChannelPoolingOptions
    {
        ProducerPoolSize = 16
    });

Pooling DI usage

RabbitMQ.ClientKit.ChannelPooling also includes IServiceCollection extensions that swap in the pooled producer and reusable consumer channel providers.

using Microsoft.Extensions.DependencyInjection;
using RabbitMQ.ClientKit.Configuration;
using RabbitMQ.ClientKit.ChannelPooling;

var services = new ServiceCollection();

services.AddPooledRabbitMqClient(
    new RabbitMqConnectionOptions
    {
        HostName = "localhost",
        ClientProvidedName = "payments-api"
    },
    new RabbitMqChannelPoolingOptions
    {
        ProducerPoolSize = 16
    });

That registration still resolves RabbitMqClient, RabbitMqPublisher, and RabbitMqConsumer, but the underlying channel strategy changes to:

• a bounded producer-channel pool for publish-heavy workloads
• a reusable consumer-channel store keyed by consumer name, so long-lived consumer channels can be retained and reused between subscriptions

Named DI registrations

You can also register multiple RabbitMQ clients side-by-side by name with .NET 8 keyed services.

using Microsoft.Extensions.DependencyInjection;
using RabbitMQ.ClientKit;
using RabbitMQ.ClientKit.Configuration;
using RabbitMQ.ClientKit.ChannelPooling;

var services = new ServiceCollection();

services.AddNamedRabbitMqClient(
    "orders",
    new RabbitMqConnectionOptions
    {
        HostName = "orders-rabbit",
        ClientProvidedName = "orders-api"
    });

services.AddNamedPooledRabbitMqClient(
    "payments",
    new RabbitMqConnectionOptions
    {
        HostName = "payments-rabbit",
        ClientProvidedName = "payments-api"
    },
    new RabbitMqChannelPoolingOptions
    {
        ProducerPoolSize = 16
    });

Resolve them either directly from the service provider:

var provider = services.BuildServiceProvider();

var ordersClient = provider.GetRequiredKeyedService<RabbitMqClient>("orders");
var paymentsPublisher = provider.GetRequiredKeyedService<RabbitMqPublisher>("payments");
var paymentsConsumer = provider.GetRequiredKeyedService<RabbitMqConsumer>("payments");

Or inject them into application services with keyed DI:

using Microsoft.Extensions.DependencyInjection;
using RabbitMQ.ClientKit;

public sealed class BillingPublisher([FromKeyedServices("payments")] RabbitMqPublisher publisher)
{
    public RabbitMqPublisher Publisher { get; } = publisher;
}

appsettings.json registration

You can also register a default connection, named connections, and arrays of named producer/consumer definitions from configuration.

{
  "RabbitMq": {
    "DefaultConnection": {
      "HostName": "default-rabbit",
      "UserName": "guest",
      "Password": "guest",
      "ClientProvidedName": "default-api"
    },
    "NamedConnections": [
      {
        "Name": "billing",
        "HostName": "billing-rabbit",
        "UserName": "guest",
        "Password": "guest",
        "ClientProvidedName": "billing-api"
      }
    ],
    "Producers": [
      {
        "Name": "orders-created",
        "Publish": {
          "QueueName": "orders.created"
        }
      },
      {
        "Name": "billing-charged",
        "ConnectionName": "billing",
        "Publish": {
          "ExchangeName": "billing",
          "RoutingKey": "charged"
        }
      }
    ],
    "Consumers": [
      {
        "Name": "orders-worker",
        "Subscribe": {
          "QueueName": "orders.created"
        }
      },
      {
        "Name": "billing-worker",
        "ConnectionName": "billing",
        "Subscribe": {
          "QueueName": "billing.charged",
          "ConsumerName": "billing-worker-channel"
        }
      }
    ]
  }
}

Register that section like this:

using RabbitMQ.ClientKit;

builder.Services.AddRabbitMqClientKit(builder.Configuration.GetSection("RabbitMq"));

Then resolve configured endpoints through IRabbitMqEndpointResolver:

using RabbitMQ.ClientKit;

public sealed class OrderPublisherService(IRabbitMqEndpointResolver rabbitMq)
{
    public Task PublishCreatedAsync(object message, CancellationToken cancellationToken = default)
    {
        var producer = rabbitMq.GetRequiredProducer("orders-created");
        return producer.Publisher.PublishAsync(message, producer.Options, cancellationToken);
    }
}

ConnectionName is optional on producers and consumers. When omitted, the definition uses DefaultConnection; when supplied, it resolves against one of the NamedConnections.

Configured endpoints resolve to richer handles than just raw options:

• IRabbitMqEndpointResolver.GetRequiredProducer(name) returns a RabbitMqConfiguredProducer with the producer registration, resolved publisher, resolved client, and bound RabbitMqPublishOptions
• IRabbitMqEndpointResolver.GetRequiredConsumer(name) returns a RabbitMqConfiguredConsumer with the consumer registration, resolved consumer, resolved client, and bound RabbitMqConsumerOptions

Dynamic configuration package

If you want settings to come from a database or another reloadable source instead of being fixed at startup, use the optional RabbitMQ.ClientKit.DynamicConfiguration package.

It adds:

• IRabbitMqDynamicConfigurationSource for loading the latest snapshot
• IRabbitMqDynamicEndpointResolver for resolving the current client/producer/consumer
• RabbitMqDynamicProducer and RabbitMqDynamicConsumer handles that read the latest configuration on each use

Example registration with a custom source:

using RabbitMQ.ClientKit.DynamicConfiguration;

builder.Services.AddSingleton<IRabbitMqDynamicConfigurationSource, DatabaseRabbitMqConfigurationSource>();
builder.Services.AddDynamicRabbitMqClientKit<DatabaseRabbitMqConfigurationSource>();

Or from a reloadable configuration section:

using RabbitMQ.ClientKit.DynamicConfiguration;

builder.Services.AddDynamicRabbitMqClientKit(builder.Configuration.GetSection("RabbitMq"));

Example source shape:

using RabbitMQ.ClientKit.Configuration;
using RabbitMQ.ClientKit.DynamicConfiguration;

public sealed class DatabaseRabbitMqConfigurationSource : IRabbitMqDynamicConfigurationSource
{
    public ValueTask<RabbitMqDynamicConfigurationSnapshot> GetConfigurationAsync(CancellationToken cancellationToken = default)
    {
        var defaultConnection = new RabbitMqConnectionOptions
        {
            HostName = "rabbit-a",
            ClientProvidedName = "orders-api"
        };

        var producers = new Dictionary<string, RabbitMqProducerRegistration>
        {
            ["orders-created"] = new(
                "orders-created",
                null,
                new RabbitMqPublishOptions
                {
                    QueueName = "orders.created"
                })
        };

        return ValueTask.FromResult(new RabbitMqDynamicConfigurationSnapshot(defaultConnection, producers: producers));
    }
}

Using the dynamic resolver:

using RabbitMQ.ClientKit.DynamicConfiguration;

public sealed class OrderPublisherService(IRabbitMqDynamicEndpointResolver rabbitMq)
{
    public async Task PublishCreatedAsync(object message, CancellationToken cancellationToken = default)
    {
        var producer = rabbitMq.GetProducer("orders-created");
        await producer.PublishAsync(message, cancellationToken);
    }
}

Refreshing configuration:

await rabbitMq.RefreshAsync(cancellationToken);

Future producer calls will use the latest loaded options and connection mapping. Existing active consumer subscriptions are not hot-swapped automatically; refresh affects future SubscribeAsync(...) calls.

For configuration binding, Subscribe is the preferred consumer section name. The older Consume section name is still accepted for compatibility.

The pooling package uses:

• a bounded producer-channel pool for publish-heavy workloads
• a reusable consumer-channel store keyed by consumer name, so long-lived consumer channels can be retained and reused between subscriptions

Customization and extensibility

The README examples use the default JSON serializer and the built-in connection/channel implementations, but the public abstractions are intentionally open for customization.

Useful extension points include:

• IRabbitMqSerializer when you want a serializer other than the built-in JsonRabbitMqSerializer
• IRabbitMqConnectionManager when connection creation/lifetime needs to be controlled externally
• IRabbitMqProducerChannelProvider and IRabbitMqConsumerChannelProvider when you want a different channel-leasing strategy
• IRabbitMqEndpointResolver for resolving configured named endpoints from DI
• IRabbitMqDynamicConfigurationSource for loading dynamic configuration snapshots from a database, API, or custom provider
• IRabbitMqDynamicClientActivator when dynamically resolved clients need a custom construction strategy

For example, a custom serializer can be registered in DI before the higher-level services are consumed:

public sealed class CustomSerializer : IRabbitMqSerializer
{
    public string ContentType => "application/json";

    public ReadOnlyMemory<byte> Serialize<T>(T message)
    {
        throw new NotImplementedException();
    }

    public T Deserialize<T>(ReadOnlyMemory<byte> body)
    {
        throw new NotImplementedException();
    }
}

var services = new ServiceCollection();
services.AddRabbitMqClient(new RabbitMqConnectionOptions
{
    HostName = "localhost"
});
services.AddSingleton<IRabbitMqSerializer, CustomSerializer>();

If you replace default services after calling AddRabbitMqClient(...), use the usual .NET DI override patterns so the final registration order matches the behavior you want.

Testing

Unit tests can run entirely in-process because the client toolkit surface is mostly channel-management and serialization logic.

For integration tests, you will want:

• Docker Desktop, Podman, or another way to run a disposable RabbitMQ broker
• a RabbitMQ image such as rabbitmq:management
• ports 5672 for AMQP and optionally 15672 for the management UI
• deterministic test credentials and vhost setup
• a test fixture that creates unique queue and exchange names per test run

A simple starting point is:

docker run --rm -d --name rabbitmq-test -p 5672:5672 -p 15672:15672 rabbitmq:management

Once that is available, integration tests should verify publish/consume round trips, topology declaration, ack/reject/requeue flows, and pooled channel behavior against the real broker.

The test project now includes container-backed coverage for those scenarios:

• integration tests are marked with Trait("Category", "Integration") and auto-skip when neither Docker nor Podman is available
• load tests are marked with Trait("Category", "Load") and auto-skip when neither Docker nor Podman is available
• the load suite defaults to 1000 messages and a 60 second completion window, configurable via RABBITMQ_CLIENTKIT_LOAD_MESSAGE_COUNT and RABBITMQ_CLIENTKIT_LOAD_TIMEOUT_SECONDS
• the load suite now covers both the default transient channel strategy and the pooled channel strategy with matched publish/consume scenarios
• RabbitMqLoadTests.TransientAndPooledClients_ReportComparativeMetrics runs both strategies back-to-back and writes comparable publish and end-to-end throughput ratios to the xUnit test output

Examples:

dotnet test --filter "Category=Integration"
dotnet test --filter "Category=Load"

The load scenarios are:

Test	Strategy	Output
RabbitMqLoadTests.TransientClient_PublishesAndConsumesConfiguredBurst	Core AddRabbitMqClient(...) transient producer/consumer channels	Publish duration and end-to-end throughput
RabbitMqLoadTests.PooledClient_PublishesAndConsumesConfiguredBurst	AddPooledRabbitMqClient(...) pooled producer/reusable consumer channels	Publish duration and end-to-end throughput
RabbitMqLoadTests.TransientAndPooledClients_ReportComparativeMetrics	Runs both strategies under the same settings	Per-scenario metrics plus pooled/transient throughput ratios

Performance snapshot

Current local baseline from RabbitMqLoadTests.TransientAndPooledClients_ReportComparativeMetrics with RABBITMQ_CLIENTKIT_LOAD_MESSAGE_COUNT=100000:

Strategy	Publish duration	Publish throughput	End-to-end duration	End-to-end throughput	Notes
Transient/default	1.928s	51,876 msg/s	5.349s	18,695 msg/s	Windows 11, Ryzen 7 5800X, 32 GB RAM, Docker 29.8.2, rabbitmq:3.13-management, dotnet test --no-build --filter "Category=Load" --logger "console;verbosity=detailed"
Pooled	1.556s	64,264 msg/s	5.567s	17,963 msg/s	Same host and broker settings as transient/default

In this comparative run, pooled throughput was about 1.24x faster for publish throughput and 0.96x of the transient/default end-to-end throughput.

The isolated single-scenario stress runs from the same 100k-message suite were still favorable to pooling end-to-end:

Scenario run	Publish throughput	End-to-end throughput
TransientClient_PublishesAndConsumesConfiguredBurst	43,379 msg/s	17,632 msg/s
PooledClient_PublishesAndConsumesConfiguredBurst	64,419 msg/s	18,600 msg/s

Treat these as reproducible local baselines rather than formal benchmarks: the numbers still include test-host, test-runner, and broker-container overhead, so absolute throughput and even the relative end-to-end winner can vary by machine and run order.

License

Licensed under the Apache License, Version 2.0. See LICENSE for details.

Maintained by Colin Campbell.