Network Observability with eBPF and OpenTelemetry
summary
Demonstration of the OpenTelemetry eBPF Instrumentation (OBI) network subsystem shows eBPF programs on hosts tracking per-flow bytes and direction, exporting metrics via Prometheus or OTLP; a 3-node demo runs as a daemonset per node, naive collection produces ~1,000 metric series in a minute but grouping by subnet/zone collapses cardinality while preserving protocol and namespace labels. Implementation includes GeoIP enrichment (added without touching eBPF bytecode), Helm chart for Kubernetes, and ~60 contributors; Grafana Labs reports large inter-zone cost savings using this approach.
AI-written summary · about
chapters (16)
- 0:00Network observability intro
- 1:03Why network observability matters
- 4:10SNMP limitations
- 5:54NetFlow and sFlow monitoring
- 7:38Host-based eBPF observability
- 10:24BPF for packet capture
- 11:48From BPF to eBPF analysis
- 13:33OpenTelemetry eBPF intro
- 14:16OBI basic metrics config
- 16:41Enhancing metrics with attributes
- 18:47Efficient flow grouping
- 20:52Geo IP enrichment
- 23:02Kubernetes zone-aware metrics
- 27:55Dashboards with enriched data
- 29:38Contribute to OBI
- 31:22Q&A on topology/storage
related talks
30mAgentless instrumentation with eBPF, Prometheus and OpenTelemetry
31mZero instrumentation observability with eBPF
27mStep By Step Kubernetes Observability with eBPF
26mOpenTelemetry or eBPF? That is the Question
23mOpenTelemetry Auto Instrumentation for Effortless Observability
33mEffortless, Standardised Homelab Observability With eBPF
22mHarnessing Observability for 5G Performance: eBPF and OpenTelemetry I...
35mAn Exemplary Path: Leveraging EBPFs and OpenTelemetry To Auto-instrument for Exemplars