Identifying waste resources in modern-scale cloud infrastructures is a critical
sustainability issue since it helps free up additional capacities for extra
tasks and improves the performance of the systems while optimizing their costs.
It is already well-recognized as a challenging task for human operators
regarding manual and massive action efforts. At the same time, the problem is
quite complicated -- a complete and satisfactory solution is yet to be
achieved. The paper proposes a novel and AI-driven approach to the problem.
Applying rule induction learning across the history of service deployment
instances to the log event data of the underlying entities, we extract
conditions that lead to specific patterns, such as Resource Termination, thus
providing a predictive mechanism for detecting objects subject to such actions
in a real-time fashion. This explainable recommender system (called Cloud
Sweeper) serves as an AI operations assistant for cloud users and Site
Reliability Engineers (SRE) in their administrative duties.