Incorrect values in the header of Dead Letters

[mahfuz840]

Describe the bug

Hi,

I am experiencing two issues in RabbitMQ 4.1.0 related to dead-letter headers:

1. When a message is dead-lettered for the first time, the x-first-death-exchange header is missing.

2. When a message is dead-lettered more than once, the x-first-death-queue header is overwritten.
   For example, if a message is rejected in queue A and routed to A.dlq, then later consumed and forwarded to queue B, and subsequently rejected again in queue B, the x-first-death-queue header is set to "B" instead of preserving the original value "A".

Please let me know if additional details or reproduction steps are required.

Reproduction steps

1. Set up queue A with a dead-letter exchange (DLX) that routes messages to A.dlq.
2. Publish a message to queue A.
3. Reject the message in queue A (e.g., basic.reject / basic.nack with requeue = false).
4. Observe the headers on the message when it arrives in A.dlq:
   • The x-first-death-exchange header is missing.
5. Consume the message from A.dlq and republish it to queue B (which is also configured with its own DLX).
6. Reject the message in queue B.
7. Observe the dead-letter headers on the message after it is dead-lettered from queue B:
   • The x-first-death-queue header is set to "B" instead of the original "A".

Expected behavior

• When a message is dead-lettered for the first time, all x-first-death-* headers (x-first-death-exchange, x-first-death-queue, x-first-death-reason) should be present and populated correctly.

• The x-first-death-queue and x-first-death-exchange headers should reflect the original queue and exchange where the message was first dead-lettered and must not be overwritten on subsequent dead-lettering events.

Additional context

No response

[ansd]

@mahfuz840 RabbitMQ 4.1 reached End of Community support on 31 Jan 2026. Can you please test the same steps with the latest version RabbitMQ 4.2.3?

[mahfuz840]

Yes, I tried with 4.2.3 as well and the result is same.

[lukebakken]

@mahfuz840 what would be extremely helpful is for you to provide code to reproduce this issue. Anything you can do to make it easier for Team RabbitMQ to investigate means we can focus on a fix instead of getting an environment up-and-running.

[mahfuz840]

It will take some time but I will try to provide code to reproduce the issue.

[lukebakken]

Thank you. That saves Team RabbitMQ a lot of time and guessing.

[ansd]

In 5.) you write that you republish a new message.

Starting in RabbitMQ 4, RabbitMQ won't interpret the x-death header anymore when receiving a message from AMQP 0.9.1 clients.
See PR description of #11339.
Hence at least your 2nd expectation is wrong.

[michaelklishin]

A super short justification for that: clients mess up x-deaths or republish, update them unintentionally, so this is a protection mechanism for the broker.

That header was never meant to be set or modified by applications. It's a read-only header maintained by RabbitMQ itself.

[mahfuz840]

Any observation on the missing x-first-death-exchange header value?

[ansd]

@mahfuz840 as @lukebakken pointed out, please provide code to repro this issue.