consensus
termcloudprogramming
The problem of getting several machines to agree on one value when any of them may be slow, may fail, and cannot tell those apart.
It sounds administrative and it is the hardest thing in distributed systems, because a node that has not answered is indistinguishable from a node that is dead. Every replicated database, cluster leader election and configuration store rests on a consensus algorithm, and the practical consequence is a majority rule: an odd number of members, a quorum required to decide, and a minority that must stop rather than proceed alone. That is why a three-node cluster survives one failure and a two-node cluster survives none.
Also known as: distributed consensus