State Checkpointing — References

The algorithms and surveys that define what a consistent snapshot is, and the delivery semantics that determine whether resuming repeats an effect.

What a consistent snapshot is

• Distributed Snapshots: Determining Global States of Distributed Systems — K. Mani Chandy and Leslie Lamport, ACM Transactions on Computer Systems, 1985. The algorithm for capturing a globally consistent state without stopping the system, and the definition of consistency that makes a snapshot restorable at all. Cited by name.
• A Survey of Rollback-Recovery Protocols in Message-Passing Systems — E. N. Elnozahy, Lorenzo Alvisi, Yi-Min Wang and David B. Johnson, ACM Computing Surveys, 2002. Coordinated versus uncoordinated checkpointing, message logging, and the domino effect that uncoordinated checkpoints can produce. The reference treatment of the trade-offs in this capability. Cited by name.
• Lightweight Asynchronous Snapshots for Distributed Dataflow — Paris Carbone, Gyula Fóra, Stephan Ewen, Seif Haridi and Kostas Tzoumas, arXiv preprint, 2015 (arXiv:1506.08603). Barrier- based snapshots taken at declared points in a running dataflow without halting it. Cited by name.

Why resuming can repeat an effect

• Virtual Time and Global States of Distributed Systems — Friedemann Mattern, in Parallel and Distributed Algorithms, 1989. Consistent cuts and the formal basis for deciding whether a captured state could have occurred. Cited by name.
• Life Beyond Distributed Transactions: An Apostate's Opinion — Pat Helland, CIDR, 2007. Why operations crossing a system boundary must be made idempotent and addressed by a stable key rather than wrapped in a transaction, which is the design behind a side-effect ledger. Cited by name.

Practice

• Temporal documentation — a working treatment of deterministic replay, workflow versioning for in-flight executions, and the separation between workflow state and activity side effects. Linked at its root; section paths move between releases.

No verified reference is recorded for snapshot retention policy in agent runtimes, or for reaching snapshots during subject erasure. Both are engineering practice without published study.