Ever since we announced Windows Azure there has been speculation about how reliable it will be and how we will insure that your application will continue to operate reliably in the face of changes to your application or the underlying software. Although we did say that we use multiple role instances and update domains to insure that your allocation has no single point of failure it was not until recently that we exposed more of the details of how this will work.
Now we have published the documentation on how this will work here and there is an excellent blog post here that explains the process in more detail.
There is also a new Deployment and Management API documented here which can be used as an alternative to using the portal for many deployment and management operations. But that will have to wait for another blog post.