Introduction to Rate Limiting, Circuit Breaking, and Degradation (with Sentinel Practice)

Based on the key video points from Chapter 3 (3-1 ~ 3-9) of the courseware, this guide organizes a set of service protection guidelines for beginners, helping to understand "why rate limiting, circuit breaking, and degradation are needed," and how to quickly get started with Sentinel.

Quick Overview of Learning Path

• 3-1 Understanding the background of service avalanche, rate limiting, circuit breaking, and degradation
• 3-2 Sentinel vs. Hystrix comparison, clarifying technology selection
• 3-3 Sentinel QPS Rate Limiting Basics
• 3-4 Predictive Rate Limiting and Cold Start
• 3-5 Throttling Configuration Demo
• 3-6 Sentinel Circuit Breaking and Degradation Strategies
• 3-7 Circuit Breaking Strategy – Based on Error Ratio
• 3-8 Degradation Strategy – Based on Response Time
• 3-9 Gin + Sentinel Practical Rate Limiting

Causes of Service Avalanche

When sudden traffic exceeds the system's capacity (e.g., system capacity is 1k QPS but suddenly 2k QPS floods in), it can lead to slower responses, soaring error rates, and subsequently drag down more dependencies, forming an "avalanche." Common causes:

• Sudden Traffic Spikes: Caused by major promotions, trending news, web crawlers, etc., leading to a surge in access volume.
• Dependency Failures: Downstream services experience delays or crashes, leading to accumulated calls.
• Resource Exhaustion: Threads/connections/CPU are fully occupied, causing more requests to queue up.

The three-piece strategy for handling: Rate Limiting to block the flood, Circuit Breaking to pull the circuit breaker, and Degradation to provide a fallback experience.

graph LR
    A[突发流量] --> B(限流：削峰填谷)
    B --> C{依赖正常?}
    C -->|是| D[服务稳定运行]
    C -->|否| E(熔断：暂时切断调用)
    E --> F(降级：返回备用方案)
    F --> D

Rate Limiting: Adding a 'Sluice Gate' to the Entrance

Rate Limiting controls the number of requests or concurrency within a unit of time, allowing the system to operate within a controllable range. Common strategies:

Rate limiting messages should be friendly: "Currently, there are many visitors, please try again later or pay attention to notifications."

graph TD
    subgraph 限流闸门
        Tokens[令牌桶] -->|取令牌| Request[请求放行]
        Request --> Service[核心服务]
    end
    Tokens -.超限.-> Fallback[返回 429 或排队提示]

Circuit Breaking: Making the System 'Rejuvenated'

"Circuit breaking" is similar to an electrical circuit breaker – when abnormal indicators such as error rate or response time are detected, calls are temporarily cut off to prevent fault propagation. Because the system can catch its breath and recover after the 'breaker is pulled,' it's jokingly referred to in class as making the system "rejuvenated."

Sentinel supports three types of circuit breaking rules:

• Slow Call Ratio: Triggered when the proportion of requests exceeding the configured response time threshold is too high.
• Exception Ratio: Circuit breaking occurs when the proportion of exceptions within the statistical window exceeds the limit.
• Exception Count: Circuit breaking occurs when the total number of exceptions within a unit of time reaches the threshold.

Circuit breaking typically goes through three stages:
Open（全阻断） -> Half-Open（少量探测） -> Closed（恢复正常）。

Degradation: Providing an Alternative Experience

When the core process is under too much pressure or a dependency is circuit-broken, degradation strategies provide users with "simplified but usable" results:

• Return cached or fallback data (e.g., display yesterday's inventory).
• Prompt for later processing, asynchronous callback (order queuing, ticket handling).
• Temporarily disable high-cost features (e.g., turn off recommendation lists, only retain basic search).

The key to degradation is preparation in advance: copy, fallback interfaces, and frontend placeholders must all be ready.

Sentinel Quick Overview

Sentinel is an open-source high-availability protection framework from Alibaba, with core values:

• Unified console for managing rules such as rate limiting, circuit breaking, degradation, and system protection.
• Supports Java and Go, providing rich ecosystem adaptations (Dubbo, Spring Cloud, Gin, etc.).
• Possesses capabilities such as real-time monitoring, link aggregation, and rule pushing.

Sentinel Architecture Diagram:

graph LR
    subgraph 控制台
        Dashboard[Dashboard 控制台]
    end
    Dashboard -- 推送规则 --> DataSource[数据源/配置中心]
    DataSource -- 动态规则 --> Sentinel[Sentinel 客户端]
    Sentinel -- 上报指标 --> Dashboard
    Sentinel --> App[业务应用]

Getting Started Quickly with Go Language (Taking Gin Project as an Example)

1. Install Dependencies

bash
go get github.com/alibaba/sentinel-golang@latest
go get github.com/alibaba/sentinel-golang/pkg/adapters/gin

1. Initialize Sentinel

```go
package main

import (
"log"
"github.com/alibaba/sentinel-golang/api"
"github.com/alibaba/sentinel-golang/core/flow"
)

func initSentinel() {
if err := api.InitDefault(); err != nil {
log.Fatalf("init sentinel: %v", err)
}
_, err := flow.LoadRules([]*flow.Rule{
{
Resource: "GET:/api/orders",
MetricType: flow.QPS,
Count: 200,
ControlBehavior: flow.WarmUp,
WarmUpPeriodSec: 10,
WarmUpColdFactor: 3,
},
})
if err != nil {
log.Fatalf("load flow rules: %v", err)
}
}
<code>

<ol>
<li><strong>Integrate Gin Middleware</strong></li>
</ol>

</code>go
r := gin.Default()
r.Use(sgin.SentinelMiddleware())

r.GET("/api/orders", func(c *gin.Context) {
if entry, blockErr := api.Entry("GET:/api/orders"); blockErr != nil {
c.JSON(429, gin.H{"code": 429, "msg": "拥挤，请稍后重试"})
return
} else {
defer entry.Exit()
}
c.JSON(200, gin.H{"orders": []string{"#1201", "#1202"}})
})
```

1. Local Debugging Suggestions
2. Use ab/wrk for stress testing to verify if rate limiting takes effect.
3. Observe real-time QPS and Block metrics in logs and Dashboard.

Key Points for Java Environment Integration

<!-- Maven -->
<dependency>
  <groupId>com.alibaba.csp</groupId>
  <artifactId>sentinel-core</artifactId>
  <version>1.8.6</version>
</dependency>

@Configuration
public class SentinelConfig {
    @PostConstruct
    public void init() throws Exception {
        List<FlowRule> rules = new ArrayList<>();
        FlowRule rule = new FlowRule("product_list");
        rule.setGrade(RuleConstant.FLOW_GRADE_QPS);
        rule.setCount(1000);
        rule.setControlBehavior(RuleConstant.CONTROL_BEHAVIOR_RATE_LIMITER);
        rule.setMaxQueueingTimeMs(500);
        rules.add(rule);
        FlowRuleManager.loadRules(rules);
    }
}

Combined with Spring Cloud Alibaba, degradation callback methods can be annotated with @SentinelResource.

Dashboard Startup Guide

1. Download the console:

bash
wget https://github.com/alibaba/Sentinel/releases/download/1.8.6/sentinel-dashboard-1.8.6.jar

1. Start:

bash
java -Dserver.port=8080 -Dcsp.sentinel.dashboard.server=localhost:8080 \
-Dproject.name=sentinel-dashboard -jar sentinel-dashboard-1.8.6.jar

1. Client parameters (Java example):

bash
-Dcsp.sentinel.dashboard.server=localhost:8080 \
-Dcsp.sentinel.api.port=8719 \
-Dproject.name=demo-service

1. Log in to the console (default username and password are both sentinel), create flow control/circuit breaking/system protection rules in the interface, and push them in real-time.

Formulating Your Own Protection 'Battle Plan'

• Observation Metrics: QPS, average response time, error rate, downstream dependency health.
• Threshold Setting: Based on capacity assessment and stress test data; set differentiated rules for different interfaces.
• Circuit Breaking Recovery: Configure Half-Open probing logic to ensure the service can automatically close the circuit after recovery.
• Degradation Copy: Confirm fallback pages and prompts with product/frontend teams in advance.
• Post-mortem Review: After each rate limiting / circuit breaking trigger, record the cause, adjust thresholds, or scale up.

sequenceDiagram
    participant User
    participant Gateway as API 网关
    participant Service as 核心服务
    participant Fallback as 降级/缓存

    User-->>Gateway: 请求下单
    Gateway-->>Service: 令牌校验
    alt QPS 超限
        Gateway-->>User: 429 拥堵提示
    else 调用异常
        Service-->>Gateway: 熔断触发
        Gateway-->>Fallback: 查询兜底方案
        Fallback-->>User: 排队中，请稍后
    end

Quick Review Checklist

• Rate limiting controls inbound traffic within the system's capacity, prioritizing the protection of core capabilities.
• Circuit breaking actively cuts off calls when dependencies are abnormal, allowing the system to 'recover' quickly.
• Degradation is a fallback solution, allowing users to perceive the service as 'still alive' even during failures.
• Sentinel covers rate limiting, circuit breaking, degradation, and system protection, with visual configuration and real-time monitoring via the console.
• Preparing thresholds, copy, monitoring, and drills in advance is essential to 'rejuvenate' and cope before a real avalanche strikes.