v6.2.1: CallbackException is not invoked for unhandled exceptions from AsyncEventingBasicConsumer.Received

[qrli]

rabbitmq-dotnet-client v6.2.1
rabbitmq server v3.8.3

When an unhandled exception thrown from within AsyncEventingBasicConsumer.Received, IModel.CallbackException will not be triggered in v6.2.1. It had always been working up to v6.1.0.

I see you have unit tests for this scenario for the sync BasicConsumer, but not for the Async consumer. So the issue may be only for the async API.

[michaelklishin]

A contribution would be welcome. How unhandled exceptions are treated by async APIs in general can be very different from "regular" code.

[qrli]

I dug in the code base for sometime, searching for clues. It is my first time reading it, so still quite unfamilar.
I kind of believe I found the cause in AsyncConsumerWorkService.cs:

                        try
                        {
                            Task task = work.Execute();
                            if (!task.IsCompleted)
                            {
                                await task.ConfigureAwait(false);
                            }
                        }
                        catch (Exception e)

This code tries to optimize in case Task has already finished. However, IsCompleted does not mean success. It can be failure with an exception. In case the handler runs fast and throws, the above logic just eats the exception inside Task.Exception silently.

In addition, if the task is not awaited, the exception will not be unwrapped, going back to the early days issues of AggregatedException parsing.

To my knowledge, await has some builtin optimization in case the task has already completed. I'm not exactly sure how it compares with the manual optimization above, but from correctness point of view, I think it is easier and more robust to simply await the task than trying to micro-optimize here.

[michaelklishin]

Addressed in #946 contributed by @BarShavit 👏