Skip to content

kpod-metrics

eBPF-based pod-level kernel metrics collector for Kubernetes.

kpod-metrics runs as a DaemonSet on each node, attaches eBPF programs to kernel tracepoints, and exports per-pod CPU, network, memory, syscall, disk I/O, and filesystem metrics to Prometheus.

Demo

kpod-metrics demo

  • ⚡ Zero Instrumentation

    eBPF attaches to kernel tracepoints automatically. No code changes, no sidecars, no agents inside your pods.

  • 📊 Per-Pod Granularity

    Every metric is labeled with namespace, pod, container, and node. Drill down from cluster to individual container.

  • 🌐 L7 Protocol Detection

    Auto-detect HTTP, DNS, Redis, MySQL, Kafka, and MongoDB traffic with per-request latency and error tracking.

  • 📈 Prometheus-Native

    Metrics exported directly to Prometheus. Works with Grafana dashboards (included), ServiceMonitor, and OTLP export.

Quick Install

helm repo add kpod-metrics https://pjs7678.github.io/kpod-metrics
helm repo update
helm install kpod-metrics kpod-metrics/kpod-metrics \
  --namespace kpod-metrics --create-namespace

See the Quick Start guide for detailed setup instructions.

Key Features

  • 50+ metrics across CPU, network, memory, syscalls, disk I/O, filesystem, and interrupts
  • Auto-discovered service topology from TCP peer data — no configuration needed
  • L7 protocol detection for HTTP, DNS, Redis, MySQL, Kafka, MongoDB
  • Grafana dashboards included with 9 metric rows
  • Prometheus Operator support with ServiceMonitor and 18 alerting rules
  • OTLP export for OpenTelemetry-compatible backends
  • Three profiles (minimal, standard, comprehensive) to control overhead
  • Kotlin eBPF DSL — BPF programs defined in type-safe Kotlin, not raw C
  • Broad kernel support — 4.18+ (legacy) and 5.2+ (CO-RE)
  • Tested at scale — 1,000 nodes / 100,000 pods

Architecture

Node (DaemonSet pod)
┌─────────────────────────────────────────────────┐
│  Spring Boot (JDK 21 + Virtual Threads)         │
│                                                  │
│  MetricsCollectorService (every 30s default)    │
│  ├── eBPF Collectors ──► JNI ──► BPF Maps      │
│  └── Cgroup Collectors ──► /sys/fs/cgroup       │
│                                                  │
│  Prometheus exporter (:9090/actuator/prometheus) │
└─────────────────────────────────────────────────┘
         │ JNI (libkpod_bpf.so)
    ┌────▼────────────────────────┐
    │ Linux Kernel                │
    │ eBPF programs on            │
    │ tracepoints: sched_switch,  │
    │ tcp_sendmsg, oom_kill, ...  │
    └─────────────────────────────┘

Learn more in the Architecture overview.