ccfos/huatuo

eBPF-based Linux kernel observability 🚀🚀

What it solves

HUATUO addresses the difficulty of gaining deep, real-time visibility into the Linux kernel and hardware performance without introducing significant overhead. It solves the problem of "performance jitters" (like CPU spikes or I/O surges) in cloud-native and AI infrastructure by providing automated tracing and continuous profiling that doesn't require modifying application code.

How it works

It uses eBPF and other Linux kernel dynamic tracing technologies (kprobe, tracepoint, ftrace) to implement zero-instrumentation observability. It monitors kernel subsystems—such as memory management, CPU scheduling, and networking—and uses an event-driven mechanism to automatically capture runtime context when "slow paths" (like page faults or scheduling delays) are triggered.

Who it’s for

It is designed for operators of large-scale cloud-native computing environments, AI computing clusters, and bare-metal infrastructure who need to troubleshoot system failures and optimize high availability and performance.

Highlights

  • Low Overhead: Maintains performance impact below 1% using BPF.
  • AutoTracing: Automatically retains snapshots to diagnose transient performance issues like CPU sys spikes and Loadavg spikes.
  • Hardware Support: Observes not only CPUs and memory but also PCIe interconnects, network adapters, and AI accelerators (GPUs and NPUs).
  • Ecosystem Ready: Integrates with Prometheus, Grafana, Pyroscope, and Elasticsearch, and automatically maps data to Kubernetes container labels.

Related

  • Project
  • Project
  • Project
  • Project
  • Project