this is not a issue, it's my thought about im

[blueeyezhy]

大佬您好，im架构上利用kafka配置2个topic，一个用来处理客户端进来的msg，一个用来处理服务端出去的msg，平台架构会不会更简单，实现起来更容易？

ws微服务包含 3个模块：websocket manager （管理客户端 websocket）；kafka “in” topic的生产者（将所有进来的销售，发给kafka > broker）；kafka “out”topic的消费者（从kafka out topic消费消息，发给websocket manager）

chat微服务包含 3 个模块：kafka “in” topic的消费者（从 “in”topic 消费销售）； 消息处理（将消息按照收件人（单人或者group）进行> 拆分重新打包成 msgs，然后grpc给 db微服务保存，同时转给 kafka “out” topic的生产者）；kafka “out” topic的生产者将msg 发给 > kafka broker

每个消息 由客户端 指定一个 uuid，跟踪该消息的状态（sending，sent，arrived，read），不用mongo，只用pg和消息状态，跟踪送 > 达，已读，未读，消息同步

[blueeyezhy]

sequenceDiagram
autonumber
    box lightyellow frontend <br/> ios/androd <br/> win/web
    participant clients
    end
    box lightyellow backend <br/> grpc micro services
    participant ws_svc
    participant kafka_br
    participant chat_svc
    participant db_svc
    participant postgres
    participant service_center 

    end

    note over clients,postgres: message loop

    rect rgb(191,239,255)
    note over clients,ws_svc: web socket (incoming msg)
    clients ->>+ ws_svc: msg
    note over kafka_br,ws_svc: kafka in_topic produce 
    ws_svc ->>+ kafka_br: msg
    ws_svc ->>- clients: ack("Sent")
    end

    rect rgb(191,239,255)
    note over kafka_br,chat_svc: kafka in_topic consume
    kafka_br ->>+ chat_svc: msg
    chat_svc ->>+ db_svc: msgs parsed
    db_svc ->>+ postgres: msgs
    postgres ->>- db_svc: ack
    db_svc ->>- chat_svc : ack
    note over kafka_br,chat_svc: kafka out_topic produce
    chat_svc ->>- kafka_br: msgs parsed
    end

    rect rgb(191,239,255)
    note over kafka_br,ws_svc: kafka out_topic consume
    kafka_br ->>+ ws_svc: msg
    note over clients,ws_svc: web socket (outgoing msg)
    ws_svc ->>+ clients: msg
    clients ->>- ws_svc: ack("Arrived")
    end

    note over clients,service_center : grpc request & response

[blueeyezhy]

[Xu-Mj]

sequenceDiagram
autonumber
  box lightyellow frontend <br/> ios/androd <br/> win/web
  participant clients
  end
  box lightyellow backend <br/> grpc micro services
  participant ws_svc
  participant kafka_br
  participant chat_svc
  participant db_svc
  participant postgres
  participant service_center 

  end

  note over clients,postgres: message loop

  rect rgb(191,239,255)
  note over clients,ws_svc: web socket (incoming msg)
  clients ->>+ ws_svc: msg
  note over kafka_br,ws_svc: kafka in_topic produce 
  ws_svc ->>+ kafka_br: msg
  ws_svc ->>- clients: ack("Sent")
  end

  rect rgb(191,239,255)
  note over kafka_br,chat_svc: kafka in_topic consume
  kafka_br ->>+ chat_svc: msg
  chat_svc ->>+ db_svc: msgs parsed
  db_svc ->>+ postgres: msgs
  postgres ->>- db_svc: ack
  db_svc ->>- chat_svc : ack
  note over kafka_br,chat_svc: kafka out_topic produce
  chat_svc ->>- kafka_br: msgs parsed
  end

  rect rgb(191,239,255)
  note over kafka_br,ws_svc: kafka out_topic consume
  kafka_br ->>+ ws_svc: msg
  note over clients,ws_svc: web socket (outgoing msg)
  ws_svc ->>+ clients: msg
  clients ->>- ws_svc: ack("Arrived")
  end

    note over clients,service_center : grpc request & response

这里从服务器出来的消息为什么还要再次经过kafka？这不是增加了复杂度了嘛~

[blueeyezhy]

我想着消息的进出，无论是单对单还是group chat都通过kafka解耦，功能上更模块化；kafka的broker更充分利用，整体上性能会更好吧

[Xu-Mj]

没有必要，增加不了吞吐，还增加了延迟，增加了复杂度，增加了出错的概率，功能上更模块化这块怎么说？

[zoujiaqing]

kafka 就是来保障消息能触达，稳定为先，性能优化次之，当然有更好的方案可以拿出来探讨。