A Kubernetes Controller for Managing the Availability of Elastic Microservice Based Stateful Applications

TL;DR

This paper evaluates Kubernetes' ability to ensure high availability for stateful microservices and proposes an HA State Controller to improve recovery times and guarantee service availability.

Contribution

It introduces an HA State Controller that enhances Kubernetes by enabling application state replication and automatic redirection for better availability.

Findings

Kubernetes' repair actions alone cannot meet 99.999% availability.

The proposed HA State Controller improves recovery time by 50%.

Different deployment architectures impact availability and scaling overhead.

Abstract

The architectural style of microservices has been gaining popularity in recent years. In this architectural style, small and loosely coupled modules are deployed and scaled independently to compose cloud-native applications. Carrier-grade service providers are migrating their legacy applications to a microservice based architecture running on Kubernetes which is an open source platform for orchestrating containerized microservice based applications. However, in this migration, service availability remains a concern. Service availability is measured as the percentage of time the service is provisioned. High Availability (HA) is achieved when the service is available at least 99.999% of the time. In this paper, we identify possible architectures for deploying stateful microservice based applications with Kubernetes and evaluate Kubernetes from the perspective of availability it provides for its managed applications. The results of our experiments show that the repair actions of Kubernetes cannot satisfy HA requirements, and in some cases cannot guarantee service recovery. Therefore, we propose an HA State Controller which integrates with Kubernetes and allows for application state replication and automatic service redirection to the healthy microservice instances by enabling service recovery in addition to the repair actions of Kubernetes. Based on experiments we evaluate our solution and compare the different architectures from the perspective of availability and scaling overhead. The results of our investigations show that our solution can improve the recovery time of stateful microservice based applications by 50%.