The SberTech team, specifically the R4C.Support unit, conducted an R&D project to automate database maintenance tasks. The research focused on managing a fleet of over 800 PostgreSQL database instances using autonomous agents.
The proposed architecture integrates Grafana and Prometheus for monitoring, utilizing customized exporters to gather metrics. The team explored using AI agents to handle routine tickets, such as disk space cleanup and scheduled database shutdowns.
A core part of the study involved developing an autonomous agent capable of executing VACUUM and ANALYZE commands. The system incorporates LLM-based anomaly detection to monitor database performance and trigger maintenance actions automatically.
The project also includes a dedicated event feed to track agent operations and provides a framework for integrating LLMs into database administration workflows. The team documented the implementation process, including code examples for the agent's logic.
