Fine-grained distributed consistency guarantees with effect orchestration

Highly-available distributed applications typically require data to be replicated over geo-distributed stores that offer weak consistency guarantees by default. Unfortunately, undesirable behaviors may arise under weak consistency that can violate application correctness, forcing designers to either implement complex ad-hoc mechanisms to avoid these anomalies or by sacrificing performance, choose to run applications using stronger levels of consistency. In this paper, we describe a lightweight runtime system that relieves developers from having to make such tradeoffs. Instead, our approach leverages declarative axiomatic specifications that reflect the necessary constraints any correct implementation must satisfy to guide a runtime consistency enforcement and monitoring mechanism. Experimental results show that the performance of our (provably optimal and safe) automatically derived fine-grained consistency enforcement mechanisms is better than common store-offered consistency guarantees.