DeepFlow

Active
GitHub Go Apache-2.0

Description

DeepFlow is an eBPF-based zero-code observability platform covering metrics, distributed tracing, logs, and continuous profiling, with SmartEncoding for full-stack correlation and fast queries for cloud-native and AI workloads.

Key Features

  • Zero-code collection - eBPF-based auto-capture of metrics, traces, and logs from any-language services
  • Distributed tracing for any request - covers gateways, service meshes, databases, queues, DNS, and NICs
  • Continuous profiling - function-call stacks for CPU, GPU, memory, and network at under 1% overhead
  • Full-stack correlation - correlates application, infrastructure, and AI service observability signals
  • Wide protocol support - serves as a backend for Prometheus, OpenTelemetry, SkyWalking, and Pyroscope
  • SmartEncoding storage - separates tags from data, reducing storage cost by 10x vs ClickHouse String

Use Cases

💡 Debug performance bottlenecks across language boundaries in Kubernetes-hosted AI services
💡 Auto-trace LLM inference paths to pinpoint slow-response root causes
💡 Continuously profile production workloads with minimal overhead
💡 Feed eBPF trace data into existing Prometheus and OpenTelemetry stacks
💡 Build a unified observability platform for cloud-native and GPU workloads

Tags

observability ebpf distributed-tracing profiling kubernetes

Categories

📊 Observability

Quick Start

# Install DeepFlow Community into an existing Kubernetes cluster via Helm
helm repo add deepflow https://deepflowio.github.io/deepflow
helm install deepflow deepflow/deepflow -n deepflow --create-namespace

# Once the pods are ready, access the deepflow-server web UI
kubectl -n deepflow port-forward svc/deepflow-server 20417:20417

# After deploying deepflow-agent to your app nodes, data is auto-collected
Visit GitHub Visit Website View Docs

Related Projects