Mastering High Availability in Microservices with Sentinel Go

Background

Microservice stability is a critical concern when moving from monolithic to distributed architectures. Service dependencies become more complex, and high‑availability challenges increase.

Sentinel Overview

Sentinel is an open‑source flow‑control component from Alibaba that protects distributed services through rate limiting, traffic shaping, circuit breaking, and system‑adaptive protection. It supports Java, Go, C++ and integrates with Service Meshes such as Istio and Envoy. Sentinel Go, released early this year, adds native support for Go microservices and provides modules for go‑micro, gRPC, Dubbo, Gin, etc.

Core Concepts

Resource : a logical unit such as an API endpoint, RPC method, or code block that is instrumented with a name.

Rule : a condition that triggers a control action for one or more resources. Rules can be updated at runtime via dynamic data sources like etcd, Consul, or Nacos.

High‑Availability Scenarios

1. Flow Control

Sudden traffic spikes can overwhelm a service, causing CPU/load spikes and crashes. Sentinel lets you define QPS‑based rules to reject excess requests once a threshold is exceeded.

_, err = flow.LoadRules([]*flow.FlowRule{{
    Resource:          "some-service", // resource name
    MetricType:        flow.QPS,        // QPS mode
    Count:             10,              // max 10 QPS per instance
    ControlBehavior:   flow.Reject,    // reject excess requests
}})

2. Warm‑Up Throttling

When a service starts with low load, a sudden surge can crash it. Warm‑up mode gradually raises the allowed traffic to the configured limit, giving the system time to initialize resources such as DB connections or caches. Sentinel Go 0.6.0 introduced this mode.

3. Concurrency Control & Circuit Breaking

Unstable downstream services can cause thread‑pool exhaustion. Sentinel provides lightweight concurrency limits and circuit‑breaker strategies based on response time or error ratio to isolate failing calls. Two circuit‑breaker strategies are supported: slow‑call (response‑time) and error‑ratio.

4. Hotspot Protection

Hot parameters (e.g., popular product IDs during a promotion) can bypass caches and overload databases. Sentinel can identify hot parameter values and limit their QPS or concurrency, even supporting per‑user rate limits via argument‑based rules. Hotspot rules can be dynamically updated through the same data‑source adapters.

5. System‑Adaptive Protection

When static rules are insufficient, Sentinel’s adaptive protection monitors CPU, load, QPS, latency, and concurrency, adjusting traffic limits dynamically similar to TCP BBR. This keeps the system at optimal throughput without overload.

Getting Started

Clone the Sentinel Go repository and explore the examples:

Repository: https://github.com/alibaba/sentinel-golang