A Study on the Aging and Fault Tolerance of Microservices in Kubernetes

Microservice-based applications are increasingly being adopted along with cloud service models, and nowadays serve millions of customers daily. They are supported by container-based architectures which are managed by orchestration platforms, such as Kubernetes, that monitor, manage, and automate most of the tasks. Although these tools provide failover capabilities, it is not yet studied how effective they are in dealing with diverse types of faults. Fault injection is an effective methodology for validating components that are supposed to detect the malfunctions and report/correct them. This paper studies the effectiveness of Kubernetes in dealing with faults and aging in microservices, and on the possibility of using faults to accelerate aging effects for testing purposes. For this, we conducted an analysis of the implementation and tuning of Kubernetes probes, followed by experiments with varying load and fault injection into two distinct and representative microservice testbeds to analyze the capacity of probes in detecting issues in applications. The goal is to improve the knowledge of researchers and developers on whether Kubernetes can detect different faults and aging issues. Also, even though some services tend to accumulate aging effects, with increasing resource consumption, Kubernetes does not detect them nor acts on them, indicating that probes may be insufficient for aging scenarios. Results also showed that fault injection is useful to accelerate aging effects for the testing and evaluation purposes.