container monitoring

Practical guide to GPU-aware monitoring for AI containers: establish quick CLI baselines, then run cAdvisor, Node Exporter, and NVIDIA DCGM, scraped by Prometheus and visualized in Grafana, with Bash PID-to-container mapping and a lightweight logger; add PromQL alerts to catch OOMs, GPU saturation, and flapping containers so you ship faster, cut costs, and stop fire drills.