Browsing by Author "Haapaniemi, Antti"

Today’s applications are largely deployed in the cloud and are often implemented using the microservice architecture, which divides a software system into distributed services and provides a solution to multiple issues in monolithic software systems, such as maintainability, scalability, and technology lock. However, industry experts find monitoring microservices a challenge due to the added complexity and distributed nature of microservices. Microservices are typically monitored by intrusive approaches, which incur an added cost of development by requiring instrumentation of the source code or adding monitoring agents. In contrast, the non-intrusive approaches do not. Microservices often communicate using the HTTP protocol via a centralized API Gateway, which could provide a convenient way to monitor microservices without disruption to the microservices. In this thesis, we study non-intrusive approaches to monitor microservices and present our non-intrusive approach to monitor microservices for faults and performance issues by examining anomalies in HTTP requests transferring through a centralized API Gateway. Our approach requires minimal disruption and is easy to implement as it utilizes the API Gateway for monitoring. We implemented our approach in the Amazon Web Services (AWS) cloud environment to an existing software system to find real-world issues and challenges of applying our approach. Alarms were created by using anomaly detection capabilities provided by the AWS CloudWatch service. Our non-intrusive approach can monitor latency, traffic, and errors, but not saturation. Multiple incidents of interest were detected during our evaluation period. We present the challenges and issues we were faced with. In addition, we introduce guidelines and a library to further simplify the deployment of our approach.