A system for multi-staged in-memory checkpoint replication with relaxed
consistency includes a replication manager and a cluster with a plurality
of nodes. In accordance with a configurable replication policy, the
replication manager may be configured to identify a primary group of
nodes to store primary replicas of a specified source memory region of a
source node, and to identify a node of the primary group as an initiator
of replication to a secondary group of nodes. In response to a
replication request, the replication manager may be configured to copy
contents of one or more data blocks of the source memory region to a each
node of the primary group within a single atomic transaction, and
initiate an operation to copy the contents of the data blocks from the
initiator to an additional node of the secondary group.