MODULE 07 · SYSTEM ARCHITECTURE
A load balancer distributes incoming traffic across healthy servers or application instances.
A load balancer receives traffic and chooses a healthy server to handle each request.
It sits in front of multiple servers or application instances. By spreading requests instead of sending everything to one machine, it helps a product stay responsive and available as demand changes.
Think of a load balancer like a restaurant host who sends arriving guests to open tables.
The host does not cook the food. The host watches which tables are available, avoids sending guests to a closed table, and keeps one table from receiving every group. The servers do the actual application work.
A load balancer turns a group of servers into one dependable destination for callers.
Spread traffic across available servers so work is shared.
Use health checks to stop routing traffic to an instance that cannot serve correctly.
Choose a method such as round robin, least connections, or weighted routing.
Give a service room to handle more concurrent requests than one server could handle.
Keep requests moving when one instance becomes unhealthy or unavailable.
Let clients call one stable address instead of knowing every backend instance.
Route browser traffic across several copies of the web or application tier.
Keep a stable API address while backend capacity grows behind it.
Send requests to healthy instances of a service such as checkout or search.
Spread a campaign or sale surge across a prepared pool of instances.
Route around an unhealthy availability zone when the design supports it.
Balance traffic between workers, inference servers, or other internal services.
Imagine an online retailer launching a limited-time sale.
The team runs several checkout instances and configures health checks. The load balancer has one public checkout address.
Requests are spread across healthy instances. If Checkout 2 fails its health check, new requests go to Checkout 1 or 3.
The load balancer improves the request path, but it does not make a database unlimited. The checkout service and its data layer still need capacity and protection.
The load balancer is a traffic-distribution layer. It can work behind a DNS name or API gateway, and it can sit in front of services that use databases, caches, or queues.
Load balancer: mainly distributes traffic across healthy instances. API Gateway: manages an API entry boundary with policies such as authentication, rate limiting, and request routing.
Load balancer: chooses where requests go. Autoscaling: adds or removes instances. They often work together but solve different problems.
Load balancer: sends a request to an application server. CDN: serves cached content from edge locations closer to users.
Load balancer: makes the routing decision. Health check: supplies a signal about whether an instance should receive traffic.
| ENGLISH TERM | 中文 | SIMPLE ENGLISH | 中文小白理解 |
|---|---|---|---|
| Load Balancer | 负载均衡器 | A component that distributes incoming traffic across healthy servers. | 把进入系统的请求分配给健康服务器的组件。 |
| Traffic | 流量 | Requests and data moving through a system. | 在系统中流动的请求和数据。 |
| Server Instance | 服务器实例 | One running copy of an application or service. | 正在运行的一份应用或服务副本。 |
| Health Check | 健康检查 | A test that indicates whether an instance can serve traffic. | 判断某个实例能不能继续接收请求的测试。 |
| Round Robin | 轮询 | A rule that sends requests to instances in repeating order. | 按顺序轮流把请求发给不同实例。 |
| Least Connections | 最少连接 | A rule that prefers the instance with fewer active connections. | 优先把请求给当前连接数较少的实例。 |
| Failover | 故障切换 | Moving work to a healthy alternative after a failure. | 出故障后把工作转移到可用的备用路径。 |
| Horizontal Scaling | 水平扩展 | Adding more instances to handle more work. | 增加更多实例来处理更多工作。 |
| Availability | 可用性 | How consistently a service is reachable and usable. | 服务能否持续被访问和使用。 |
| Backend | 后端 | The server-side software that handles application work. | 在服务器端处理应用业务的软件部分。 |
A load balancer gives one stable entry point a way to distribute requests across healthy application instances.
It is a traffic and availability tool—not a replacement for capacity planning, autoscaling, caching, or a resilient data layer.