消息队列消息推送引擎流量控制设计  

Flow control design for messa push engine in message queuing

作　　者：董纯铿 耿煜[2] 刘志军[1] DONG Chunkeng;GENG Yu;LIU Zhijun(School of Software Engineering,Shenzhen Institute of Information Technology,Shenzhen,Guangdong,China 518172;School of International Exchange and Cooperation,Shenzhen Institute of Information and Technology,Shenzhen,Guangdong,China 518172)

机构地区：[1]深圳信息职业技术学院软件学院,广东深圳518172 [2]深圳信息职业技术学院国际交流与合作学院,广东深圳518172

出　　处：《深圳信息职业技术学院学报》2023年第6期16-24,共9页Journal of Shenzhen Institute of Information Technology

基　　金：广东省教育科学规划课题(高等教育专项)(项目编号:2021GXJK548)。

摘　　要：消息队列在分布式微服务架构中常用于服务解耦,在高吞吐应用中将任务异步化从而达到削峰平谷的作用。消息推送引擎存在于消息队列和消息处理服务之间,承担消息推送协议转换及推送消息到目标服务的任务。当前消息推送引擎存在两大问题:1、推送速率不能匹配下游服务处理消息速率,2、消息积压造成拥塞。当推送引擎推送消息速率大于下游服务处理消息的最大能力时,一方面会直接压垮下游应用,另一方面会造成消息积压导致推送引擎本身内存溢出而崩溃;当推送引擎推送消息小于下游服务处理消息的能力时,会加大消息流转的时延,不能满足低时延应用要求。本文提出一种基于背压策略,结合推送时延、带宽、内存三方指标动态调整推送速率的流量控制设计,可以做到通过下游服务处理消息速率反向调节推送消息的速率,在带宽、内存允许的情况下以最大能力将消息流转出,从而实现可靠的高吞吐低时延消息流转。Message queuing is often used for service decoupling in distributed microservice architecture,which asynchronizes tasks in high-throughput applications to achieve peak shaving and valley leveling.The message push engine exists between the message queue and the message processing service,responsible for transforming the message push protocol and pushing messages to the target service.There are two major issues with the current message push engine:the push rate cannot match the downstream service's message processing rate and message backlog causes congestion.When the push engine pushes messages at a rate greater than the maximum capacity of downstream services to process messages,on the one hand,it will directly crush downstream applications;On the other hand,it can cause message backlog and cause the push engine to overflow memory and crash.When the push engine pushes messages that are smaller than the downstream service's ability to process messages,it increases the latency of message flow and cannot meet the requirements of low latency applications.This article proposes a flow control design based on backpressure strategy,which dynamically adjusts the push rate by combining the three indicators of push delay,bandwidth,and memory.It can achieve reverse adjustment of the push message rate through downstream service processing,and flow out messages with maximum capacity under the conditions of bandwidth and memory,thus achieving reliable high throughput and low latency message flow.

关 键 词：消息队列 推送引擎 流量控制 拥塞控制 

分 类 号：TP393.06[自动化与计算机技术—计算机应用技术]