Publishing message batches w/ confirmation using .NET & RabbitMQ.Client 7.x

[jehrenzweig-captionhealth]

I'm attempting to publish messages in batches using the .NET RabbitMq.Client 7.0.0 package, and I'm running into some problems. More specifically, exceptions are being thrown when I await a ValueTask that calls IChannel.BasicPublishAsync(), for one of two reasons:

1. The connection is already closed (AlreadyClosedException)
2. basic.return is returned (PublishException where IsReturn == true)

I'm using a set of channel options similar to the ones described in the tutorial:

private static readonly CreateChannelOptions _createChannelOptions = new(
    publisherConfirmationsEnabled: true,
    publisherConfirmationTrackingEnabled: true,
    outstandingPublisherConfirmationsRateLimiter: new ThrottlingRateLimiter(25));

The ValueTask objects that call IChannel.BasicPublishAsync() all have mandatory: true set.

What's strange is that each of the messages I attempt to publish does get routed to a queue, after an exception is thrown. So it's working... but it's also throwing exceptions I don't know how to explain or avoid.

Any suggestions on where to start looking? Thanks in advance!

[michaelklishin]

Moved to a much more visible repository.

Server logs and a traffic capture will help narrow down the problem quicker.

[lukebakken]

@jehrenzweig-captionhealth - if you can provide code that reliably reproduces this issue, I will be able to quickly assist you.

Questions:

• Are the exceptions random?
• Does RabbitMQ log anything at the time of the exceptions?

[jehrenzweig-captionhealth]

Yep, I know a standalone repro case will help things along. I'll see if I can cobble something together.

To answer your questions:

• Nope, these two types of exceptions are thrown every time. Though I'm not sure if the same messages sent to the same exchanges are producing the same exceptions, though... I'll keep digging.
• For each message in the batch, I see the following message appear in the RMQ logs: Channel is stopping with 1 pending publisher confirms. So if I have a batch with 3 messages, I'll see that message 3 times in the logs.

[lukebakken]

Nope, these two types of exceptions are thrown every time

OK, this sounds like a bug in your topology setup. When you provide code to reproduce the issue, ensure that it sets up exchanges and queues in the same way as in your environment, or you can export your RabbitMQ definitions and include that. Thanks.

[jehrenzweig-captionhealth]

I'm pleased to report that I believe I've solved the problem & AFAIK it was entirely on my end. Between refactoring our code to implement the new async API & the removal of batch processing, I believe I'd introduced an errant async/await method call that was causing channels and/or connections to be incorrectly & prematurely closed/disposed. A few fixes later, I'm seeing RMQ batch-enqueuing messages with publish confirms enabled sans exceptions.

Thanks for the quick replies yesterday!

[lukebakken]

I appreciate the follow-up, and how you have contributed to this project. Please feel free to continue to ask for assistance, if needed, and contribute, when you can 😸