Abstract: A storage account is replicated across multiple data centers so as to withstand an outage of the storage account. A front end and applications use data stored in the storage accounts to manage resources of a cloud computing system. A client includes an interface used by the front end to access the storage accounts, in addition to an interface that is used by the applications to access the storage accounts. Other features, which ensure that a read operation survives even if a head replica or a tail replica is down, include reading from a head replica instead of from a tail replica, a two phase prepare-commit operation to propagate changes from head to tail replicas, and provisioning a back end repair service if a failure occurs in the middle of the prepare-commit operation.

Abstract: Methods and devices for data migration may include initially processing requests from a plurality of geographic regions for a cloud service using a global back-end service with a global storage account storing data. The methods and devices may include establishing a region back-end service with a region storage account in at least one geographic region of the plurality of geographic regions to support the cloud service for users in the at least one geographic region, wherein the region back-end service includes a region RTable. The methods and devices may include receiving, by the region back-end service, user requests for the cloud service from one or more users in the at least one geographic region and accessing, via the region RTable, one or more rows of data associated with the at least one geographic region from the global storage account in response to the user requests.

Abstract: A storage account is replicated across multiple data centers so as to withstand an outage of the storage account. A front end and applications use data stored in the storage accounts to manage resources of a cloud computing system. A client includes an interface used by the front end to access the storage accounts, in addition to an interface that is used by the applications to access the storage accounts. Other features, which ensure that a read operation survives even if a head replica or a tail replica is down, include reading from a head replica instead of from a tail replica, a two phase prepare-commit operation to propagate changes from head to tail replicas, and provisioning a back end repair service if a failure occurs in the middle of the prepare-commit operation.