Monitoring containers: which tool to choose?

A team of researchers analyzed 69 tools to monitor containerized applications, microservices, and DevOps environments. The goal? Find the most effective way to control resources in complex contexts. The test was conducted on two infrastructures: on-premises (Docker Swarm, Kubernetes) and cloud (Google Kubernetes Engine). A test application was stressed with Apache JMeter and k6, simulating 100 to 1,000 virtual users. Metrics on CPU, memory, I/O, and network were collected using Prometheus, Grafana, cAdvisor, and Docker stats. Results show that both monitoring stacks accurately captured CPU usage peaks during load tests. Prometheus and cAdvisor recorded variations every 15-30 seconds, while Google Cloud Monitoring every 60 seconds. The research offers a practical guide to choosing the right tools based on specific needs. Containerizing applications is now essential for scalability and flexibility, but without proper monitoring, it becomes impossible to manage resources optimally. This study provides a clear framework to navigate a market full of solutions.

The research identified 69 unique tools divided into four non-mutually exclusive categories: container-based (18), cloud-based (25), microservices-oriented (13), and DevOps-focused (38). Some tools belong to multiple categories, totaling 94 assignments. The classification is based on visualization capabilities, supported metrics, and quality attributes such as performance, security, interoperability, and usability. Tools were tested in two deployment scenarios: on-premises with Docker Swarm and Kubernetes, and cloud with Google Kubernetes Engine. The test application was subjected to load and stress tests using Apache JMeter and k6 with a number of virtual users ranging from 100 to 1,000. Collected metrics included CPU usage, memory, I/O, and network. Experimental results demonstrate that both monitoring stacks captured variations in processor utilization under load at their default sampling resolutions. Prometheus and cAdvisor have a sampling interval of 15-30 seconds, while Google Cloud Monitoring records every 60 seconds. During tests with 1,000 virtual users, both stacks showed sharp CPU usage peaks. The main contribution of this work is a practical guide for selecting monitoring tools, developed on a reference experimental environment. This type of analysis is crucial for organizations adopting modern container-based and microservices architectures, where visibility across all system components is essential to ensure optimal performance and reliability.