适配环境JDK 21LTS、Spring Boot 3.5.11适用人群Java 后端开发、架构师、技术选型决策者特点基于 Spring Boot 3.5.x JDK21 实战验证代码可直接运行避免常见版本与虚拟线程误用问题一、技术背景1️⃣ JDK21JDK21 是当前长期支持版本LTS虚拟线程Project Loom正式 GA大幅降低高并发场景下的线程资源占用成本。2️⃣ Spring Boot 3.5.11Spring Boot 3.5.11 为 3.5.x 稳定维护版本基于 Spring Framework 6.x全面支持 JDK17对 JDK21 运行稳定。3️⃣ 三大主流 MQ:contentReference[oaicite:0]{index0}:contentReference[oaicite:1]{index1}:contentReference[oaicite:2]{index2}三者在架构模型、事务能力、吞吐目标和生态定位上差异明显。二、核心能力对比基于 Spring Boot 3.5.11 JDK21对比维度RabbitMQRocketMQKafka实现语言ErlangJavaJavaSpring 集成spring-boot-starter-amqprocketmq-spring-boot-starterspring-kafka架构模型Exchange-QueueNameServer BrokerBroker Topic Partition顺序消息单队列顺序支持全局/分区顺序分区内顺序事务支持无原生分布式事务原生事务消息半消息机制支持 Producer 事务延迟消息插件或 TTL原生支持需业务实现典型定位业务解耦 / 中小系统金融级核心业务大数据 / 流处理说明吞吐量与延迟强依赖硬件、磁盘、刷盘策略、消息大小与网络环境不给出固定数值。三、场景选型建议✅ 选择 RabbitMQ 的场景适合中小规模业务系统快速上线项目IoT / MQTT 协议场景运维团队规模较小特点易上手、生态成熟、可视化控制台友好。✅ 选择 RocketMQ 的场景适合电商订单金融支付库存扣减延迟关闭订单特点原生事务消息支持顺序消息延迟消息能力强Java 生态集成自然✅ 选择 Kafka 的场景适合日志采集用户行为分析实时数仓事件溯源高吞吐数据流特点分区模型天然支持扩展支持 Producer 事务流处理生态完善Flink / Spark四、Spring Boot 3.5.11 JDK21 实战整合1️⃣ 整合 RabbitMQ依赖dependencygroupIdorg.springframework.boot/groupIdartifactIdspring-boot-starter-amqp/artifactId/dependencyapplication.ymlspring:threads:virtual:enabled:true# 开启虚拟线程支持rabbitmq:host:localhostport:5672username:guestpassword:guest配置虚拟线程执行器BeanpublicExecutorrabbitVirtualExecutor(){returnneworg.springframework.core.task.VirtualThreadTaskExecutor(rabbit-virtual-);}BeanpublicSimpleRabbitListenerContainerFactoryrabbitListenerContainerFactory(ConnectionFactoryconnectionFactory,ExecutorrabbitVirtualExecutor){SimpleRabbitListenerContainerFactoryfactorynewSimpleRabbitListenerContainerFactory();factory.setConnectionFactory(connectionFactory);factory.setTaskExecutor(rabbitVirtualExecutor);factory.setConcurrentConsumers(5);factory.setMaxConcurrentConsumers(20);returnfactory;}生产者 消费者SpringBootApplicationpublicclassRabbitApp{publicstaticvoidmain(String[]args){SpringApplication.run(RabbitApp.class,args);}BeanCommandLineRunnerrunner(RabbitTemplatetemplate){returnargs-{template.convertAndSend(test-queue,Hello RabbitMQ);};}RabbitListener(queuesToDeclareQueue(test-queue),containerFactoryrabbitListenerContainerFactory)publicvoidconsume(Stringmsg){System.out.println(收到: msg);System.out.println(虚拟线程: Thread.currentThread().isVirtual());}}2️⃣ 整合 RocketMQ依赖dependencygroupIdorg.apache.rocketmq/groupIdartifactIdrocketmq-spring-boot-starter/artifactIdversion2.3.2/version/dependencyapplication.ymlspring:threads:virtual:enabled:truerocketmq:name-server:localhost:9876producer:group:test-producer-group生产者AutowiredprivateRocketMQTemplaterocketMQTemplate;BeanCommandLineRunnerrocketRunner(){returnargs-{rocketMQTemplate.convertAndSend(test-topic,Hello RocketMQ);};}消费者RocketMQMessageListener(topictest-topic,consumerGrouptest-consumer-group)ComponentpublicclassRocketConsumerimplementsRocketMQListenerString{OverridepublicvoidonMessage(Stringmessage){System.out.println(收到: message);System.out.println(虚拟线程: Thread.currentThread().isVirtual());}}RocketMQ 的线程模型由客户端内部管理如需定制线程池应通过对应配置项或扩展方式调整。3️⃣ 整合 Kafka依赖dependencygroupIdorg.springframework.kafka/groupIdartifactIdspring-kafka/artifactId/dependencyapplication.ymlspring:threads:virtual:enabled:truekafka:bootstrap-servers:localhost:9092consumer:group-id:test-groupauto-offset-reset:earliestKafka Listener 容器工厂BeanpublicConcurrentKafkaListenerContainerFactoryString,StringkafkaListenerContainerFactory(ConsumerFactoryString,Stringcf){ConcurrentKafkaListenerContainerFactoryString,StringfactorynewConcurrentKafkaListenerContainerFactory();factory.setConsumerFactory(cf);factory.setConcurrency(5);returnfactory;}生产者 消费者AutowiredprivateKafkaTemplateString,StringkafkaTemplate;BeanCommandLineRunnerkafkaRunner(){returnargs-kafkaTemplate.send(test-topic,Hello Kafka);}KafkaListener(topicstest-topic,containerFactorykafkaListenerContainerFactory)publicvoidlisten(Stringmessage){System.out.println(收到: message);System.out.println(虚拟线程: Thread.currentThread().isVirtual());}五、生产级优化建议RabbitMQ开启生产者确认机制合理设置 prefetch使用持久化队列 持久化消息监控积压与连接数RocketMQ合理设置刷盘策略SYNC_FLUSH / ASYNC_FLUSH核心业务使用事务消息Broker 主从部署Kafka分区数量与副本因子合理规划使用事务保证 Exactly-Once启用批量消费提升吞吐六、最终选型总结场景推荐快速开发 / 中小系统RabbitMQ金融级核心交易RocketMQ大数据 / 日志 / 流处理Kafka七、结语在Spring Boot 3.5.11 JDK21技术栈下虚拟线程降低并发成本三大 MQ 均可稳定集成选型关键在业务模型而非“谁吞吐更高”架构没有绝对优劣只有是否适合你的系统。