Fault-tolerant Distributed Database for Public Key Infrastructure

Abstract

This thesis presents a practical study focusing on improving service level availability (SLA) and reducing maintenance costs of distributed SQL database clusters by migrating from a traditional manually-managed virtual machine (VM) architecture to a Kubernetes-based one. The original architecture suffers from complex manual recovery processes during node malfunctions, leading to increased downtime. Three failure modes inherent to the original architecture are investigated: 1. primary node failure; 2. replica node failure; 3. split-brain mode.

The proposed architecture leverages Kubernetes to automate cluster recovery from primary and replica node failures, significantly reducing downtime compared to manual processes. Furthermore, Kubernetes’ architectural features eliminate split-brain mode entirely.