Alibaba Sentinel — Flow Control and Circuit Breaking for Distributed Systems

Introduction

Sentinel is a flow control and circuit-breaking library developed by Alibaba for protecting distributed service architectures. It provides real-time monitoring, traffic shaping, circuit breaking, and adaptive system protection to keep services stable under high load or partial failures.

What Sentinel Does

• Enforces flow control rules with QPS limiting, concurrency limiting, and traffic shaping
• Provides circuit breaking based on slow request ratio, error ratio, or error count thresholds
• Offers adaptive system protection that adjusts limits based on CPU, load, and inbound QPS
• Delivers a real-time monitoring dashboard showing per-resource metrics and rule configuration
• Supports hot parameter flow control to limit traffic on specific argument values

Architecture Overview

Sentinel operates as an in-process library that wraps protected resources (methods, URLs, RPC calls) with slot chains. Each slot chain includes a statistics slot for metrics collection, a flow control slot for rate limiting, a degrade slot for circuit breaking, and a system protection slot. The Dashboard is a separate Spring Boot application that collects metrics from Sentinel clients via HTTP and allows dynamic rule configuration pushed to clients in real time.

Self-Hosting & Configuration

• Add the Sentinel core dependency or Spring Cloud Alibaba Sentinel starter to your project
• Deploy the Sentinel Dashboard JAR for visual rule management and monitoring
• Configure transport settings so your application registers with the dashboard
• Define flow rules, degrade rules, and system rules via the dashboard UI or programmatically
• Persist rules to Nacos, ZooKeeper, or Apollo for durability across restarts

Key Features

• Millisecond-level flow control with QPS, thread count, and warm-up rate-limiting strategies
• Circuit breaker with half-open recovery and configurable slow-call and error-ratio thresholds
• Real-time dashboard with per-second metrics, topology graphs, and one-click rule creation
• Seamless integration with Spring Cloud, Dubbo, gRPC, Servlet, and reactive frameworks
• Cluster flow control mode for global rate limiting across multiple service instances

Comparison with Similar Tools

• Resilience4j — Lightweight Java circuit breaker; Sentinel adds flow control, system protection, and a monitoring dashboard
• Hystrix — Netflix circuit breaker (archived); Sentinel is actively maintained with more protection strategies
• Envoy rate limiting — Sidecar-based; Sentinel operates in-process with lower latency and richer Java integration
• Spring Cloud Circuit Breaker — Abstraction layer; Sentinel is a concrete implementation with its own dashboard
• Polly (.NET) — .NET resilience library; Sentinel is JVM-native with cluster-wide flow control capabilities

FAQ

Q: How does Sentinel differ from Hystrix? A: Sentinel provides flow control, system protection, and a real-time dashboard in addition to circuit breaking. Hystrix is archived and limited to circuit breaking and bulkheading.

Q: Can Sentinel persist rules across restarts? A: Yes. Configure a dynamic rule source like Nacos, ZooKeeper, or Apollo. Without persistence, rules are lost on restart.

Q: Does Sentinel work with non-Spring applications? A: Yes. Sentinel core is framework-agnostic. You can use it with raw Java, Dubbo, Servlet filters, or any JVM application.

Q: What is the performance overhead of Sentinel? A: Sentinel uses efficient sliding window statistics. Overhead is typically under 1ms per resource invocation in benchmarks.

Sources

• https://github.com/alibaba/Sentinel
• https://sentinelguard.io/en-us/