You might want to observe how different software versions behave in terms of throughput or resource usage. This might be easier to analyze by graphing the patterns before and after the upgrade in clear terms. In this specific example, we are going to look at container CPU usage for node_exporter before and after an upgrade.

Keep in mind a couple of considerations: for the sake of this example, node_exporter is running as a container, which is ill-advised in a real-world scenario. Furthermore, we’ll be using container_cpu_usage_seconds_total from cAdvisor instead of process_cpu_seconds_total, which is collected directly from natively instrumented applications, so that this method can be applied to any kind of containerized process, consolidating the use of cAdvisor metrics along the way.