[larsmak]

When the cloud was first being introduced one of the core principles was that your instances should be considered ephemeral - and that they can terminate at any time, without any notice or further explanation. I think some people sometimes forgets that this still holds today.

Don't expect your instance to run forever. Design your architecture with this in mind. There is a reason e.g. aws provides multiple availability zones.

[devicenull]

It's interesting, because these days there are people that define cloud as "high availability VMs". They expect that in the case of a host failure that everything can be transparently migrated to another host.

[larsmak]

I'd say that depends on what kind of service the provider is offering. Later generation cloud providers might specializing in providing such VMs - Heroku and GAE comes to mind. But on providers that offer more "bare metal" VMs one should not expect such migration. Indeed it would be almost impossible to do, since it would require a lot of information about the application you are running. Heroku and GAE solves this by limiting the capabilities of the VMs so that they have more control/information about the running applications.