Kubernetes High-Availability Software Architecture Options for Two-Node Clusters in IoT Applications

Distributed Control Systems used in industrial automation comprise numerous real-time embedded controllers and server-side supervisory software, enabling continuous operation of product plants (power, chemicals, metals, etc.). Future control systems are envisioned to rely on microservice architectures and benefit from cloud-native technologies including container-based virtualization. Kubernetes, being the defacto standard for container orchestration, provides a number of options for high-availability, which is vital for industrial processes. The built-in options for high-availability in Kubernetes utilize consensus protocols requiring at least three nodes to achieve a quorum. However, many small-scale industrial sites are typically equipped with only two supervisory server nodes to save costs. Customers are reluctant to add additional resources for the use of Ku-bernetes in such smaller facilities, thus requiring alternative mechanisms for achieving high-availability. In this paper, we provide an overview of high-availability mechanisms for two-node Kubernetes clusters and report on our experiences with a custom solution. We find that a variety of viable solutions exist, each with its own benefits and drawbacks, providing software architects for various IoT applications knowledge to make informed tradeoff decisions.