Possible double delivery/processing of a single message with multiple consumers (C# RabbitMQ.Client 7.1.2, prefetch=1, no redelivery, only one publish)

[sintesfabien]

Hello,

I am facing a critical and very strange issue with the C# RabbitMQ.Client library (v7.1.2, .NET Framework 4.8, Windows 11, RabbitMQ server 4.0.3 in a container, Erlang 26.2.5.5, Debian 11.11).

Problem summary

• When running a load test with a high number of consumers (e.g. 25+), the same message (same GUID in the JSON payload) is processed by two different consumers (different consumer numbers, different threads, different channels, different delivery tags) almost simultaneously.
• Each consumer is created with its own channel (no channel sharing, no consumer sharing, all objects are independent and thread-safe).
• Prefetch is set to 1 and autoAck is false.
• The message is published only once (confirmed by RabbitMQ tracing).
• The message is received only once in the RabbitMQ trace (on channel 20, for example), but my application logs show it being processed by two consumers (e.g. channel 20 and channel 17).
• No redelivery flag, no crash, no warning in RabbitMQ logs.
• Both consumers process and ack the message (with different delivery tags, as expected per channel).

Code architecture

• Each consumer is an instance of a class inheriting from an abstract base, with its own channel and AsyncEventingBasicConsumer.
• The event handler for message reception (ReceivedAsync) launches the processing in a Task.Run for asynchronism, but the message reception is sequential per consumer.
• All queue and exchange declarations are standard (durable, non-exclusive, non-autoDelete).
• The message is a simple JSON with a unique GUID.

What has been checked

• No channel sharing between consumers.
• No consumer sharing between threads.
• No double publication.
• No redelivery (no redelivered flag, no crash, no Nack).
• No buffer or object sharing.
• RabbitMQ trace shows only one "Message received" event for the message, but two consumers process it in the app logs.

Sample trace and logs
RabbitMQ trace:

2025-06-11T09:33:42.039+00:00: Message published
... (payload with UniqueMsgId: ff1dbbed-8331-4d52-8177-5476b31d8d93)
2025-06-11T09:35:15.224+00:00: Message received
Channel: 20
... (same payload)

App logs:

[09:35:15.225] Consumer 20 received message ff1dbbed-8331-4d52-8177-5476b31d8d93 (DeliveryTag: 123)
[09:35:15.227] Consumer 17 received message ff1dbbed-8331-4d52-8177-5476b31d8d93 (DeliveryTag: 456)

Questions

• Is there any known issue in RabbitMQ.Client (v7.1.2 or earlier) that could cause a single message to be delivered to two consumers at the same time, without redelivery, and with only one publish and one "Message received" in the trace?
• Could this be a race condition or bug in the client library under high load?
• Is there any way the same message could be delivered to two consumers, with different delivery tags and channels, but only one "Message received" event in the RabbitMQ trace?
• Any advice on further debugging or a minimal repro?
• Additional notes
• The problem only appears under high load (many consumers).
• The code is strictly 1 consumer = 1 channel = 1 thread.
• Upgrading the RabbitMQ.Client library is possible, but I would like to understand if this is a known issue or a misuse.

Thank you for your help and insights!

[michaelklishin]

@sintesfabien no, there are no such known issues.

We will not guess as to what might be going on without an executable way to reproduce against a community-supported RabbitMQ version (4.1.x). We do not guess in this community.

Take a traffic capture.

Most likely you assume that a certain exchange type, e.g. the direct exchange, can only route a message to one queue even though that's not the case (bind N queues with the same routing key to a direct exchange and it will route the message to N queues).

[sintesfabien]

Thank you for your feedback.
I am aware that a direct exchange can route a message to multiple queues if they are all bound with the same routing key.
However, this is not the case in my setup: each message is published only once to a single queue, and there is only one binding between the exchange and the queue.
The issue I am experiencing is that, very rarely, the same message in a single queue is delivered to two different consumers (on different channels), resulting in the message being processed twice.
This happens even though the message is only present once in the queue and is not being redelivered (no redelivery flag, no crash, no Nack).
So, the problem is not related to multiple queues or bindings, but to a rare double delivery to two consumers on the same queue.

[lukebakken]

Is there any known issue in RabbitMQ.Client (v7.1.2 or earlier) that could cause a single message to be delivered to two consumers at the same time, without redelivery, and with only one publish and one "Message received" in the trace?

No, you are the only user to report this, which makes it very likely the issue is due to how you're using this library.

Let me link to my previous comment: #1846 (comment)

We are willing to assist you, but only if you make it as easy as possible for us to help you. Right now, you are asking us to guess how to reproduce this issue, which is not something we will do for you.

We will need the following:

Answers to:

• Is this an application that operated normally for some time, and then started to experience this issue?
• If so, how long did it operate normally?
• When this issue started to arise, what changed?

We need a git repository with code to reproduce the issue, as well as the necessary RabbitMQ definitions. We should not have to spend any time getting your code to compile, run, or demonstrate the issue. Clear instructions are required.

Again, the simplest and most likely explanation is either your use of this library, or something particular to your environment.

[sintesfabien]

Hello,
I will try to provide a Git repository with a minimal reproducible example as soon as possible.
In the meantime, I have upgraded RabbitMQ server to version 4.1.1 with Erlang 27.3.4 (using the Docker image rabbitmq:4.1.1-management).
Unfortunately, the problem remains exactly the same.

[lukebakken]

Please answer my three questions as well.

[sintesfabien]

Hello,

This is a new application that has not yet been deployed in production.
The issues started to appear during load/integration testing using MSTest (near 50 consumers for 5000 messages).
There was no period where the application operated normally in a production environment.

Best regards

[sintesfabien]

Thank you for your help.
After further investigation, I found that the issue was caused by passing the BasicDeliverEventArgs object (ea) directly to an asynchronous task.
Since this object can be reused or modified by the RabbitMQ client, its content may change between the event handler and the asynchronous processing, leading to message mismatches and unexpected behavior.
This behavior is indeed documented in the RabbitMQ and AMQP client documentation.
Copying the message body and relevant properties immediately in the event handler before starting any asynchronous processing solves the problem.
Thank you again for your guidance and support.

[lukebakken]

Thank you for following up!