deepflowio/deepflow
Deepflow
DeepFlow 是一个 eBPF 可观测性平台，提供了一套用于持续性能分析、分布式追踪、服务依赖映射和统一遥测存储的套件。它作为一个监控系统，无需手动进行应用程序插桩或修改源代码即可收集指标、追踪和性能分析数据。
该平台通过使用协议感知的数据包解析来重构请求链，并利用自动服务依赖映射来可视化应用程序与基础设施之间的交互，从而脱颖而出。它利用专为高基数信号优化而设计的数据存储，允许用户通过 SQL 和 PromQL 接口查询统一数据。
该系统涵盖了广泛的可观测性领域，包括带有 CPU 和非 CPU 火焰图的应用程序性能分析、网络性能收集和云基础设施监控。它集成了内核级遥测收集，以跨服务和线程收集系统指标和应用程序层元数据。
Features
- eBPF-Based Collection - Implements an observability platform that uses eBPF for automatic kernel-level telemetry collection without application instrumentation.
- Unified Observability Data Models - Unifies metrics, logs, and traces into a single data model queryable via SQL and PromQL.
- Continuous Profilers - Continuously samples on-CPU and off-CPU call stacks to generate flame graphs for production performance analysis.
- eBPF-Based Application Profilers - Continuously samples stack traces via eBPF to identify on-CPU and off-CPU performance bottlenecks in compiled-language applications.
- eBPF Tooling - Leverages eBPF tooling to gather system and network performance metrics without modifying application source code.
- Observability Platforms - Provides a full observability platform using eBPF to collect metrics, logs, and traces without manual instrumentation.
- Network Protocol Parsing - Parses binary network protocol streams to extract application-layer metadata and reconstruct distributed request chains.
- Architecture Dependency Mapping - Provides automated visualization of connections and relationships between infrastructure components via network flow data.
- Distributed Request Tracking - Tracks the full lifecycle of requests across gateways, databases, and network interfaces to remove observability blind spots.
- Application Layer Protocol Dissectors - Reconstructs and decodes application-layer protocols from network traffic using eBPF to extract deep performance insights.
- Distributed Tracing - Maps request chains across services and threads using eBPF and protocol extraction to identify bottlenecks.
- Application Performance Profiling - Collects CPU and memory flame graphs in production to identify processing bottlenecks with minimal overhead.
- eBPF Profilers - Leverages eBPF for continuous kernel-level analysis and performance profiling of production processes with minimal overhead.
- Service Dependency Mapping - Automatically discovers and visualizes communication paths and interactions between application services and infrastructure.
- Telemetry Data Stores - Implements a unified backend for storing and querying industry-standard observability data using SQL and PromQL.
- High-Cardinality Metric Metadata - Applies encoding techniques to high-cardinality metric metadata to maintain query performance and reduce storage costs.
- High-Cardinality Optimizations - Reduces storage overhead for high-cardinality telemetry data by compressing and indexing common resource tags.
- Data Storage Optimizers - Optimizes the formatting of metadata tags to minimize storage overhead and support large telemetry datasets.
- Unified Telemetry Backends - Stores diverse telemetry signals in a single repository queryable via both SQL and PromQL interfaces.
- Observability Signal Unifications - Standardizes tags across metrics, logs, and traces to ensure consistent visibility across monitoring stacks.
- Topology Visualizations - Maps service interactions and infrastructure components through automated discovery to visualize and resolve performance bottlenecks.
- Cloud Resource Metadata Detection - Automatically discovers and attaches cloud infrastructure attributes to telemetry to provide operational context.
- Cloud Resource Monitoring - Provides comprehensive monitoring of cloud infrastructure by enriching telemetry with resource metadata.
- Flame Graphs - Generates call stack flame graphs for CPU, GPU, and memory usage to locate bottlenecks across business and kernel functions.
- Observability and Profiling - Cloud-native observability for distributed applications.
