hesela.devCausal consistency
Causal consistency requires reads to respect dependencies between operations while allowing concurrent writes to be ordered differently by different observers.
Program order, a read taking its value from a write, and transitivity establish potential causality. For read/write registers, each process must have a legal view of its own operations and all writes that preserves this order. A read returns the value of the latest preceding write to that key in that view, or the initial value if none exists. [1]
Concurrent means unordered by causality, not simultaneous on a clock. Plain causal consistency does not require one shared order of concurrent writes. An overwritten intermediate value can remain unread. [1]
Causal+ additionally requires convergent conflict handling. The 2011 COPS design checks dependencies before exposing replicated writes, and its COPS-GT extension adds consistent multi-key reads. The model assumes communication between threads passes through the store. [2]
The COPS client tracks key/version pairs rather than fetching each value's full dependency list. After a put, its context retains only that put, which transitively represents its earlier dependencies. COPS-GT keeps additional dependency information for get transactions. [2]
An ordering guarantee alone is not a durability promise. A documented causal-session configuration permits acknowledged writes to roll back. [3]
Relations
Citations
- Ahamad et al. Causal memory: definitions, implementation, and programming. Distributed Computing 9, 37-49 (1995), sections 1, 2 and 4.
- Lloyd et al. Don't Settle for Eventual: Scalable Causal Consistency for Wide-Area Storage with COPS. SOSP 2011, sections 3, 4.3, 4.4 and 4.6.
- MongoDB Manual: Causal Consistency and Read and Write Concerns; durability and rollback.
Further reading
Sources & caveats
- Ahamad et al. (1995), sections 1, 2 and 4, and COPS (2011), sections 3-4.6 and Appendix A, establish the scoped definitions. The PDF filename containing 1996 does not change the printed publication year.
- The MongoDB documentation supplies a configuration-specific rollback example, not a claim that every causal implementation rolls back writes.
- No system execution or benchmark reproduction was performed. Status is unknown because this is a semantic guarantee, not a product lifecycle.