An AP system preserves availability and partition tolerance when network partitions prevent nodes from communicating. It may temporarily return inconsistent results rather than stop serving requests. section 1.2.1 section 8.1
ScyllaDB is typically classified as AP. Its quorums and eventual consistency help keep the cluster running when nodes are lost or divided, while tunable consistency levels let developers choose stronger or weaker guarantees. section 1.2.1 section 8.1
With three replicas, consistency levels make the availability trade-off concrete:
- ALL requires all three replicas and tolerates no replica failure. section 1.2.1
- QUORUM requires two replicas and tolerates one failure. section 1.2.1
- ONE requires a single replica and tolerates two failures, accepting weaker consistency. section 1.2.1
Table 1.1 Sample of consistency level options, assuming a cluster with three replicas
|
Consistency level
|
Description
|
Number required to succeed
|
Failures tolerated
|
|---|---|---|---|
ALL
|
Requires all replicas to succeed
|
3
|
0
|
QUORUM
|
Requires a majority of replicas to succeed
|
2
|
1
|
ONE
|
Requires a single replica to succeed
|
1
|
2
|
Compare replica requirements and tolerated failures for common consistency levels. section 1.2.1
A network partition
When part of a cluster cannot communicate with another part, an AP database continues accepting work where possible. Replicas may temporarily disagree, with consistency restored later rather than making the database unavailable immediately. section 1.2.1 section 8.1
