Abstract: Methods and apparatus for instance migration to support rapid recovery from correlated failures are described. A failure event affecting one or more compute instances of a provider network, including a particular compute instance hosted at a first instance host, is detected based on an analysis of health status information. A determination is made as to whether a particular compute instance meets an acceptance criterion for a failure-induced migration. The acceptance criterion may be based on storage-related requests from the particular compute instance. If the particular compute instance meets the acceptance criterion, one or more configuration operations are initiated to re-launch the particular compute instance at a different instance host.

Abstract: A provider network may implement compute instance migrations across availability zones. Compute instances may be located in a particular availability zone of provider network that is implemented across multiple availability zones. A request may be received, from a client of the provider network or other component of the provider network, to migrate a compute instance that is currently operating for a client and located in one availability zone to another availability zone. A destination compute instance may be provisioned in the other availability zone based on a configuration of the currently operating compute instance. In some embodiments, other computing resources utilized by the currently operating compute instance, such as data storage resources, may be moved to the other availability zone. Migration may be completed such that the destination compute instance is currently operating for the client and the compute instance is not.

Abstract: A method for migrating a virtual machine instance within a service provider environment may include determining to migrate an instance running on a source server computer, the source server computer associated with a plurality of hardware resource settings. A target server computer may be identified based at least in part on the plurality of hardware resource settings. A local data volume of the target server computer may be synchronized with a local data volume of the source server computer. A notification may be sent to the user, indicating migration of the instance can be initiated. Upon receiving a reboot request, disconnecting at least one network interface attached to the instance. The instance may be terminated from running on the source server computer, and may be launched on the target server computer using the synchronized local data volume and instance state data.