Along with including a readable copy of the original queue message along with the stack trace in your application’s diagnostics, it’s absolutely imperative that you implement a poison queue.
Poison or dead letter queues are essential in pull based systems, because they allow us to relieve the system from having to keep processing the same message over and over again.
A typical pull based system will use queues to absorb and protect services from peak loads. Allowing them to run at their own pace. Furthermore, it allows us to distribute the queue processing load over many compute nodes. Adding and removing compute nodes can be achieved by using the Auto Scaling features which can be found in the Windows Azure Management Portal.
Alexandre Brisebois ☁ 