Advancements in Zero-Downtime Rolling Major-Version Upgrades for Distributed NoSQL Clusters: Methodologies and Fault-Tolerance Analysis

Authors

  • Evans Ighodalo EviCore Healthcare, United States Author
  • Ayinoluwa Kolawole CVS Health, USA Author
  • Emmanuel Chukwuma Okeke Arcesium, USA Author
  • Toussida Fatah Tanguy Minoungou Independent Researcher, USA Author
  • Oluranti x Oluranti Ogundapo Google Inc., USA Author

DOI:

https://doi.org/10.66592/mcnd-048d

Keywords:

Rolling upgrade, zero downtime, NoSQL, Apache Cassandra, Mon-goDB, ScyllaDB, fault tolerance, quorum, mixed-version compati-bility, database automation, fore-casting, literature synthesis

Abstract

Major-version upgrades remain one of the most dangerous routine operations performed on distributed NoSQL databases, because a major version may change the inter-node messaging protocol, the on-disk storage format, the metadata subsystem, and client-facing semantics while the cluster must keep serving traffic. This paper examines how such upgrades can be performed as rolling operations with zero planned downtime and how automation and forecasting can make them safer. Methodologically, it combines a structured literature synthesis of 239 sources, comprising 15 foundational distributed-systems and vendor sources and 224 applied studies screened from a consolidated corpus of 1,056 records, with an analytical fault-tolerance model of quorum loss during rollout. The synthesis organizes the evidence into nine themes spanning compatibility mechanisms, cloud and service operations, security, forecasting, predictive maintenance, anomaly detection, release engineering, and governance. From it, the paper develops a conceptual model consisting of a five-phase upgrade state machine (preflight and forecasting, mixed-version operation, full binary rollout, feature finalization, and cleanup), a Poisson-based risk model, and a forecast-driven automation pipeline. Under representative assumptions for a 60-node cluster, the model shows that virtual-node token placement raises the risk of quorum loss during a naive one-node-at-a-time upgrade from about 0.3% to about 8.5% at replication factor three, while rack-aligned parallel rollout reduces it to about 0.6% and a replication factor of five reduces it below 0.01%. The discussion shows that topology decisions dominate upgrade risk, that separating binary rollout from feature finalization is the decisive safety property, and that automation mainly removes procedural and correlated-defect risk. Implications are drawn for practitioners, researchers, and governance functions, and the paper closes with its limitations and an agenda for future research, including formal verification of mixed-version behavior and learned risk models for upgrade gating.

Published

2025-12-25

How to Cite

Advancements in Zero-Downtime Rolling Major-Version Upgrades for Distributed NoSQL Clusters: Methodologies and Fault-Tolerance Analysis. (2025). Journal of Integrated Science, Technology and Management, 1(2), 114-133. https://doi.org/10.66592/mcnd-048d