负载均衡(Load Balancing)是一种计算机网络和服务器管理技术,旨在分配网络流量、请求或工作负载到多个服务器或资源,以确保这些服务器能够高效、均匀地处理负载,并且能够提供更高的性能、可用性和可扩展性。
这篇文章,我们聊聊六种通用的负载均衡算法。
图片
1.轮询 (Round Robin)
轮询是指将请求按顺序轮流地分配到后端服务器上,它均衡地对待后端的每一台服务器,而不关心服务器实际的连接数和当前的系统负载。
图片
示例代码:
import java.util.List;
import java.util.concurrent.atomic.AtomicInteger;
publicclass RoundRobin {
privatefinal List<String> servers;
privatefinal AtomicInteger index = new AtomicInteger(0);
public RoundRobin(List<String> servers) {
this.servers = servers;
}
public String getServer() {
int currentIndex = index.getAndIncrement() % servers.size();
return servers.get(currentIndex);
}
}
2.粘性轮询 (Sticky Round-Robin)
粘性轮询是标准轮询算法的一个变种,它通过记住客户端与服务实例的映射关系,确保来自同一客户端的连续请求会被路由到同一个服务实例上。
它的特点是:
- 会话保持:一旦客户端首次请求被分配到某个服务实例,后续请求会"粘"在这个实例上
- 客户端识别:通常基于客户端IP、会话ID或特定HTTP头来识别客户端
- 故障转移:当目标服务实例不可用时,系统会重新分配客户端到其他可用实例
图片
示例代码:
import java.util.List;
import java.util.Map;
import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.atomic.AtomicInteger;
publicclass StickyRoundRobin {
privatefinal List<String> servers;
privatefinal AtomicInteger index = new AtomicInteger(0);
privatefinal Map<String, String> clientToServer = new ConcurrentHashMap<>();
public StickyRoundRobin(List<String> servers) {
this.servers = servers;
}
public String getServer(String clientId) {
return clientToServer.computeIfAbsent(clientId,
k -> servers.get(index.getAndIncrement() % servers.size()));
}
}
3.加权轮询 (Weighted Round-Robin)
加权轮询是标准轮询算法的增强版本,它允许管理员为每个服务实例分配不同的权重值。权重越高的实例处理越多的请求,从而实现更精细的负载分配。
图片
它的特点是:
- 权重分配:每个服务实例都有对应的权重值
- 比例分配:请求按权重比例分配到不同实例
- 动态调整:权重可以动态修改以适应不同场景
示例代码:
private static Map<String, Integer> serverMap = new ConcurrentHashMap<>();
//记录服务器权重总和
privatestaticint totalWeight = 0;
public static String weightRandom() {
//获取服务器数量
int serverCount = serverMap.size();
//如果没有可用的服务器返回null
if (serverCount == 0) {
returnnull;
}
//在此处为避免多线程并发操作造成错误,在方法内部进行锁操作
synchronized (serverMap) {
//计算服务器权重总和
for (Map.Entry<String, Integer> entry : serverMap.entrySet()) {
totalWeight += entry.getValue();
}
//生成一个随机数
int randomWeight = new Random().nextInt(totalWeight);
//遍历服务器列表,根据服务器权重值选择对应地址
for (Map.Entry<String, Integer> entry : serverMap.entrySet()) {
String serverAddress = entry.getKey();
Integer weight = entry.getValue();
randomWeight -= weight;
if (randomWeight < 0) {
return serverAddress;
}
}
}
//默认返回null
returnnull;
}
publicclass WeightRandomLoadBalancer implements LoadBalancer {
private List<String> servers = new ArrayList<>();
private Map<String, Integer> weightMap = new HashMap<>();
public WeightRandomLoadBalancer(Map<String, Integer> servers) {
this.servers.addAll(servers.keySet());
for (String server : servers.keySet()) {
int weight = servers.get(server);
weightMap.put(server, weight);
}
}
@Override
public String chooseServer() {
int weightSum = weightMap.values().stream().reduce(Integer::sum).orElse(0);
int randomWeight = ThreadLocalRandom.current().nextInt(weightSum) + 1;
for (String server : servers) {
int weight = weightMap.get(server);
if (randomWeight <= weight) {
return server;
}
randomWeight -= weight;
}
returnnull;
}
}
4.源地址哈希法 (Hash)
源地址哈希法是一种基于客户端 IP 地址的负载均衡算法,通过哈希函数将客户端IP映射到特定的服务器,确保来自同一IP的请求总是被转发到同一台服务器。
图片
示例代码:
import java.util.List;
import java.util.zip.CRC32;
publicclass SourceIPHashLoadBalancer {
privatefinal List<String> servers;
public SourceIPHashLoadBalancer(List<String> servers) {
this.servers = servers;
}
public String getServer(String clientIP) {
if (servers.isEmpty()) {
returnnull;
}
// 计算IP的哈希值
long hash = calculateHash(clientIP);
// 取模确定服务器索引
int index = (int) (hash % servers.size());
return servers.get(Math.abs(index));
}
private long calculateHash(String ip) {
CRC32 crc32 = new CRC32();
crc32.update(ip.getBytes());
return crc32.getValue();
}
}
5.最少连接 (Least Connections)
最少连接算法是一种动态负载均衡策略,它会将新请求分配给当前连接数最少的服务器,以实现更均衡的服务器负载分配。
图片
它的特点是:
- 实时监控:跟踪每台服务器的活跃连接数
- 动态决策:新请求总是分配给当前连接数最少的服务器
- 自适应:自动适应不同请求处理能力的服务器
示例代码:
import java.util.List;
import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.atomic.AtomicInteger;
publicclass LeastConnectionsLoadBalancer {
privatefinal List<String> servers;
privatefinal ConcurrentHashMap<String, AtomicInteger> connectionCounts;
public LeastConnectionsLoadBalancer(List<String> servers) {
this.servers = servers;
this.connectionCounts = new ConcurrentHashMap<>();
servers.forEach(server -> connectionCounts.put(server, new AtomicInteger(0)));
}
public String getServer() {
if (servers.isEmpty()) {
returnnull;
}
// 找出连接数最少的服务器
String selectedServer = servers.get(0);
int minConnections = connectionCounts.get(selectedServer).get();
for (String server : servers) {
int currentConnections = connectionCounts.get(server).get();
if (currentConnections < minConnections) {
minConnections = currentConnections;
selectedServer = server;
}
}
// 增加选中服务器的连接数
connectionCounts.get(selectedServer).incrementAndGet();
return selectedServer;
}
public void releaseConnection(String server) {
connectionCounts.get(server).decrementAndGet();
}
}
6.最快响应时间 (Least Response Time)
最快响应时间(Least Response Time,LRT)通过选择当前响应时间最短的服务器来处理新请求,从而优化整体系统性能。
图片
LRT 算法基于以下核心判断标准:
- 实时性能监控:持续跟踪每台服务器的历史响应时间
- 动态路由决策:新请求总是分配给响应最快的可用服务器
- 自适应学习:根据服务器性能变化自动调整流量分配
示例代码:
import java.util.*;
import java.util.concurrent.*;
import java.util.concurrent.atomic.*;
publicclass LeastResponseTimeLoadBalancer {
privatefinal List<String> servers;
privatefinal ConcurrentHashMap<String, ResponseTimeStats> serverStats;
// 响应时间统计结构
staticclass ResponseTimeStats {
privatefinal AtomicInteger totalRequests = new AtomicInteger(0);
privatefinal AtomicLong totalResponseTime = new AtomicLong(0);
privatevolatileboolean isHealthy = true;
public void recordResponseTime(long responseTimeMs) {
totalRequests.incrementAndGet();
totalResponseTime.addAndGet(responseTimeMs);
}
public double getAverageResponseTime() {
int requests = totalRequests.get();
return requests == 0 ? 0 : (double)totalResponseTime.get() / requests;
}
}
public LeastResponseTimeLoadBalancer(List<String> servers) {
this.servers = new CopyOnWriteArrayList<>(servers);
this.serverStats = new ConcurrentHashMap<>();
servers.forEach(server -> serverStats.put(server, new ResponseTimeStats()));
}
public String getServer() {
if (servers.isEmpty()) returnnull;
return servers.stream()
.filter(server -> serverStats.get(server).isHealthy)
.min(Comparator.comparingDouble(server ->
serverStats.get(server).getAverageResponseTime()))
.orElse(null);
}
public void updateResponseTime(String server, long responseTimeMs) {
ResponseTimeStats stats = serverStats.get(server);
if (stats != null) {
stats.recordResponseTime(responseTimeMs);
}
}
public void markServerDown(String server) {
ResponseTimeStats stats = serverStats.get(server);
if (stats != null) stats.isHealthy = false;
}
public void markServerUp(String server) {
ResponseTimeStats stats = serverStats.get(server);
if (stats != null) stats.isHealthy = true;
}
}