safety-property · also safety (distributed systems)

Safety property

A safety property is an execution requirement whose every violation has a finite prefix that no continuation can make compliant.

Related Observatory article: Read on hesela.com

In one consensus instance, choosing two different values violates agreement; later agreement cannot erase those conflicting decisions. Safety does not require a decision to occur. Protocol guarantees still depend on the specified fault model and retained state. [2]

A finite run without a violation does not prove safety for every allowed execution. The definition classifies a property, not the completeness of a test suite. [1]

A specified completion deadline is also a safety requirement in a timed model: later completion cannot undo a missed deadline. This is different from unbounded eventual completion. [3]

Relations

related
Liveness property, Consensus, Linearizability

Citations

  1. Alpern and Schneider. Defining Liveness. Information Processing Letters 21(4), 181-185 (1985), sections 1-3. (accessed 2026-11-20)
  2. Lamport. Paxos Made Simple (2001), sections 2.1, 2.4 and 2.5. (accessed 2026-11-20)
  3. Schneider and Lamport. Another Position Paper on Fairness. Software Engineering Notes 13(3), 1988, pp. 1-2. (accessed 2026-11-20)
Sources & caveats
  • Primary-source synthesis with constructed reasoning examples, not a protocol execution, implementation certification or new proof.

JSON: /data/terms/safety-property.json

Published by Hesela team · Updated

Report a correction · Editorial standards · Reuse guidance

Cite this page
Hesela team. Safety property. Hesela. 2026-11-20. https://hesela.dev/terms/safety-property/