Consistent cut
A consistent cut is a selection of initial prefixes of process histories that contains every causal predecessor of each included event.
A cut selects how far each process history is included. It is consistent when an included event never depends on an excluded event. Process-local order, message sending before receipt, and transitivity define the happened-before relation used here. [1] [2]
A message receipt cannot be included if its send is excluded. The reverse is allowed: an included send with an excluded receipt represents a message in transit in the associated channel state. Including neither event or both events also satisfies this single-message condition. Other events and dependencies still require checking. [1]
This is a logical boundary, not a requirement that all process states were captured at one wall-clock instant. The original Chandy-Lamport snapshot algorithm records a compatible global state while computation continues, assuming reliable FIFO channels and finite delivery delays. Its progress and reachability conditions belong to the algorithm, not to the definition of a consistent cut. [3]
Recovery inference: causal closure alone does not establish transaction completion, durable storage or a successful restore. A cut can include an unfinished transaction; a recovery protocol must preserve or reconstruct the state it needs. The single-message example is illustrative, not a captured execution or a restore experiment. [1] [3]
Relations
Citations
- Babaoğlu and Marzullo: Consistent Global States of Distributed Systems: Fundamental Concepts and Mechanisms, sections 4.5-4.6.
- Lamport (1978): Time, Clocks, and the Ordering of Events in a Distributed System, Communications of the ACM 21(7), 558-565; happened-before relation.
- Chandy and Lamport (1985): Distributed Snapshots: Determining Global States of Distributed Systems, ACM TOCS 3(1), 63-75, sections 2-4.
Sources & caveats
- The author-hosted global-state chapter supplies the prefix-cut and causal-closure definitions; Lamport supplies the event relation. Chandy-Lamport supplies the snapshot and channel-state interpretation.
- This record describes a mathematical property, not a product lifecycle, snapshot implementation or backup certification. Recovery implications are engineering inferences; no native distributed-system experiment was performed.