What are the exceptions worth retrying (the one representing transient errors) when IModel.BasicPublish fails ?

[EnricoMassone]

Hello,

I've already tried to ask this same question on both stackoverflow and the Google user group. An interesting discussion has started in stackoverflow, but I still have no clear plan to proceed.

To summarize, my scenario is the following:

• we are using the RabbitMQ .NET client from .NET core 3.1 applications (both ASP.NET core web applications and worker services)
• we are using the 5.1.0 version of the .NET client (migration to 6.X is in plan, but still to be started)
• we want to improve the resiliency of our applications, so we want to add a retry policy to the code sending messages to RabbitMQ. We want to retry over transient failures only.
• we are already using the connection auto recovery feature, but as documented here attempts to call IModel.BasicPublish when the connection is down will result in errors and the driver leaves the handling of this scenario to the developer (re delivery of messages is not automatically attempted by the driver itself)

We are struggling with identifying the exceptions thrown by IModel.BasicPublish which represent transient errors and so are worth retrying. We know that all the exceptions thrown by the driver derive from RabbitMQClientException and are defined in the RabbitMQ.Client.Exceptions namespace.

It seems to us that, among all the client defined exceptions, only BrokerUnreachableException, ConnectFailureException and OperationInterruptedException are worth retrying. All the others seem to be related with errors which are not transient in nature.

As an interesting sidenote the Microsoft microservice architecture sample application has a retry policy similar to the one we would like to define. They retry on both BrokerUnreachableException and SocketException. See here.

Can you please provide some guidance on this topic ?

Many thanks for helping.

[lukebakken]

https://github.com/dotnet/eShop