gIMI — gRPC Infrastructure Management Interface
Subtitle: Open telemetry and management interface for hypervisor, VM, and Kubernetes layers.
Problem
Today we stitch together exporters, agents, vendor APIs, SNMP, syslog, custom collectors, and separate monitoring stacks to get data out of infrastructure.
That sprawl is hard to operate and makes correlation painful. Understanding why a pod is slow often means hopping across Kubernetes, hypervisor, host, and storage tooling.
Goal
gIMI aims to provide an open, extensible, gRPC-based interface for infrastructure management and telemetry—so host, VM, Kubernetes node, workload, network, and storage can be observed in a common event chain.
Highlights
- Host inventory
- VM inventory
- Kubernetes node information
- Health and metric endpoints
- OpenTelemetry-compatible export
- eBPF-backed low-level visibility
- Remote operational endpoints (target)
Architecture approach
gIMI has two main parts:
- Host agent — Collects from libvirt, QEMU, KVM, OVS/OVN, container runtime, and hardware on the host.
- gRPC API — Exposes inventory, metrics, telemetry, and control endpoints as a clear API.
Planned additions:
- Kubernetes collector — Node, pod, workload telemetry and OpenSDDC topology graph.
- OpenTelemetry bridge — Integration with existing observability stacks.
Roadmap
- Phase 2: Host inventory, KVM/libvirt VM discovery, Kubernetes cluster discovery, basic telemetry export.
- Phase 5: Kubernetes collector for pod ↔ VM ↔ host ↔ storage chains, OpenDIMA integration.