Log-based model to enforce data consistency on agnostic fault-tolerant systems

Abstract

Agnostic fault-tolerant systems cannot recover to a consistent state if a failure/crash occurs during a transaction. By their nature, inconsistent states are very difficult to be treated and recovered into the previous consistent state. One of the most common fault tolerance mechanisms consists in logging the system state whenever a modification takes place, and recovering the system to the system previous consistent state in the event of a failure. This principle was used to design a general recovering log-based model capable of providing data consistency on agnostic fault-tolerant systems. Our proposal describes how a logging mechanism can recover a system to a consistent state, even if a set of actions of a transaction were interrupted mid-way, due to a server crash. Two approaches of implementing the logging system are presented: on local files and on memory in a remote fault-tolerant cluster. The implementation of a proof of concept resorted to a previous proposed framework, which provides common relational features to NoSQL database management systems. Among the missing features, the previous proposed framework used in the proof of concept, was not fault-tolerant.