Awesome AI for Infra › Predictive Analytics for Operations
aqstack/sentinel
Sentinel is an intelligent agent designed for Kubernetes edge environments, addressing challenges posed by unreliable network connectivity and limited resources. It proactively monitors critical node metrics such as CPU thermals, memory pressure, disk I/O, and network health. Using lightweight statistical models, Sentinel predicts potential failures before they occur, enabling preemptive interventions. A key innovation is its ability to survive network partitions; when edge nodes lose contact with the central control plane, they form a local consensus group using a Raft-lite mechanism to continue operating autonomously. This ensures that critical decisions like workload migrations or cordoning unhealthy nodes can still be made locally. Sentinel then logs these autonomous decisions for reconciliation once connectivity is restored. Unlike traditional cloud AIOps tools that assume abundant resources, Sentinel is engineered for efficiency, running its prediction models with minimal memory footprint (<64MB RAM) without requiring GPUs or cloud inference. It integrates with Kubernetes, Prometheus for metrics, and offers a robust API for health checks and decision logging, making it a powerful solution for maintaining High Availability and resilience in challenging edge deployments.