The crucial part of any cluster, especially those that are aiming towards self-adaptation, is data. Hardly anyone will dispute the need to have the past and present metrics. Without them, we'd run like a headless chicken when things go wrong. The central question is not whether they are required but what we do with them. Traditionally, operators would spend endless hours watching dashboards. That is far from efficient. Watch Netflix instead. It is, at least, more entertaining. The system should use metrics. The system generates them, it collects them, and it should decide what actions to perform when they reach some thresholds. Only then, the system can be self-adaptive. Only when it acts without human intervention can it be self-sufficient.
A system that implements self-adaptation needs to collect data, store them, and act upon them. I will skip...
