Topology Aware Consistency Policies
I am increasingly fascinated by consistency options, in a distributed
storage system, made available by topology awareness on the
client. For example, if you consider a write committed iff the
write has been made to a majority of all storage nodes and a
majority of the local nodes, where local would typically be "same
datacenter," it allows you to achieve repeatable read
read what you wrote
consistency locally when a majority of local nodes have responded
to a read request with a matching response, while still providing
overall consistency across the entire system.