为什么选择RocketMQ

为什么 RocketMQ

在阿里孕育 RocketMQ 的雏形时期,我们将其用于异步通信、搜索、社交网络活动流、数据管道,贸易流程中。随着我们的贸易业务吞吐量的上升,源自我们的消息传递集群的压力也变得紧迫。

根据我们的研究,随着队列和虚拟主题使用的增加,ActiveMQ IO模块达到了一个瓶颈。我们尽力通过节流、断路器或降级来解决这个问题,但效果并不理想。于是我们尝试了流行的消息传递解决方案Kafka。不幸的是,Kafka不能满足我们的要求,其尤其表现在低延迟和高可靠性方面,详见这里。在这种情况下,我们决定发明一个新的消息传递引擎来处理更广泛的消息用例,覆盖从传统的pub/sub场景到高容量的实时零误差的交易系统。

Apache RocketMQ 自诞生以来,因其架构简单、业务功能丰富、具备极强可扩展性等特点被众多企业开发者以及云厂商广泛采用。历经十余年的大规模场景打磨,RocketMQ 已经成为业内共识的金融级可靠业务消息首选方案,被广泛应用于互联网、大数据、移动互联网、物联网等领域的业务场景。

为什么选择RocketMQ - 图1提示

下表显示了RocketMQ、ActiveMQ和Kafka之间的比较

RocketMQ vs. ActiveMQ vs. Kafka

Messaging ProductClient SDKProtocol and SpecificationOrdered MessageScheduled MessageBatched MessageBroadCast MessageMessage FilterServer Triggered RedeliveryMessage StorageMessage RetroactiveMessage PriorityHigh Availability and FailoverMessage TrackConfigurationManagement and Operation Tools
ActiveMQJava, .NET, C++ etc.Push model, support OpenWire, STOMP, AMQP, MQTT, JMSExclusive Consumer or Exclusive Queues can ensure orderingSupportedNot SupportedSupportedSupportedNot SupportedSupports very fast persistence using JDBC along with a high performance journal,such as levelDB, kahaDBSupportedSupportedSupported, depending on storage,if using levelDB it requires a ZooKeeper serverNot SupportedThe default configuration is low level, user need to optimize the configuration parametersSupported
KafkaJava, Scala etc.Pull model, support TCPEnsure ordering of messages within a partitionNot SupportedSupported, with async producerNot SupportedSupported, you can use Kafka Streams to filter messagesNot SupportedHigh performance file storageSupported offset indicateNot SupportedSupported, requires a ZooKeeper serverNot SupportedKafka uses key-value pairs format for configuration. These values can be supplied either from a file or programmatically.Supported, use terminal command to expose core metrics
RocketMQJava, C++, GoPull model, support TCP, JMS, OpenMessagingEnsure strict ordering of messages,and can scale out gracefullySupportedSupported, with sync mode to avoid message lossSupportedSupported, property filter expressions based on SQL92SupportedHigh performance and low latency file storageSupported timestamp and offset two indicatesNot SupportedSupported, Master-Slave model, without another kitSupportedWork out of box,user only need to pay attention to a few configurationsSupported, rich web and terminal command to expose core metrics