DeepFlow — eBPF Observability for Cloud & AI

Practical Notes

• GitHub: 4,074 stars · 453 forks; pushed 2026-05-12 (verified via GitHub API).
• Repo includes manifests/deepflow-docker-compose/docker-compose.yaml for all-in-one deployment.
• Topics include opentelemetry, kubernetes, and llm, signaling focus on modern cloud + AI workloads.

Main

Use DeepFlow as a “no-regrets baseline” before you add app-level tracing everywhere:

• Start with the all-in-one Compose deployment to validate data flow end-to-end.
• Connect one Kubernetes cluster or one VM pool first; confirm you can see service maps / flows and correlate spikes.
• Only then add targeted OpenTelemetry instrumentation for the top 1–3 critical services.

For production, plan capacity around your traffic volume (agents can help estimate storage/retention from your p95 throughput) and keep a clear retention policy for traces vs metrics vs logs.

FAQ

Q: Is DeepFlow only for Kubernetes? A: No—README and repo structure mention both Kubernetes and host/VM deployments.

Q: Do I still need OpenTelemetry? A: Often yes for deep app semantics, but DeepFlow can give you useful coverage with minimal instrumentation.

Q: What should I verify first? A: That the all-in-one deployment is stable and you can correlate a known incident spike across flows/metrics/traces.