Choice of Disciplines for Recovery of a Fault-Tolerant Cluster with Continuous Availability of an Analytical DBMS

This paper examines options for building a highly reliable cluster of duplicated computing systems designed to support the fault-tolerant operation of an analytical DBMS. The effectiveness of these critical-to-data-loss cluster systems largely depends on the process of recovery in different failure scenarios. Markov models are proposed to justify the selection of recovery options for these cluster systems. The reliability of these Markov models is unique in that it considers the implementation of three levels of data protection. The first level implements data replication within the duplicated system, the second level restores data based on the latest backup, and the third level restores data accumulated after the latest backup until the failure of two memory nodes that store data replicas. The third stage of data recovery can be combined with the execution of functional queries. The research aims to identify factors that affect the cluster’s reliability and fault tolerance and improve its readiness to ensure continuous availability of the DBMS without losing accumulated data during system operation.