Data Systems
FLP
no async consensus can always terminate with 1 crash
Source board
This definition comes from the board below. Open it to see the term in context.
Consistency and ConsensusData SystemsMore terms from this board
- 2PCtwo-phase commit (prepare → commit); BLOCKING if coordinator dies
- causal consistencypreserve cause→effect order, cheaply
- consensusnodes agree on one value (agreement/integrity/validity/termination)
- coordination serviceZooKeeper/etcd, consensus-as-a-service
- coordinator/participantthe 2PC roles
- Lamport timestampa counter giving causal total order
- linearizabilityact like one copy, reads always see the latest write (recency)
- partial vs total ordersome-events-unordered vs one-timeline
- Paxos/Raft/Zabconsensus algorithms
- strict serializabilitylinearizable + serializable
- total order broadcastsame-order delivery to all = a replicated log = consensus