/grpcdemo

Primary LanguageJava

目的

GRPC作为国外的主流RPC开发框架对于**的大多数小伙伴而言还比较陌生。由于GRPC本身在GO语言更为流行，因此想通过Java来 学习并使用GRPC框架可能在初期会遇到一些问题。希望通过该项目展示一个更纯粹的GRPC Java使用方式，同时对于自己， 也希望记录如何从熟悉的HTTP1.1线性编程**转向为HTTP2双工异步编程**过程中的学习总结。

项目模块

根据连接发起的角色分为两个模块，分别放在以下两个不同的目录

  • server: GRPC服务端代码
  • client: GRPC客户端代码

    • 命名规范

    包命名

  • org.example.grpc.unary: GRPC请求响应模式包
  • org.example.grpc.stream: GRPC流模型包
  • org.example.sdk: 客户端SDK工具包，降低客户端使用GRPC的难度

    • GRPC服务端

  • GRPCServer : GRPC服务端启动类，默认端口9999

    • GRPC客户端

  • XXXClient: GRPC客户端访问启动类，默认连接localhost:9999

    • IDL/接口定义文件

    位置 : src/main/proto

    学习笔记

    双向流API

    1. By default, each channel being created map to one TCP connection(could be 0 or many), you can use netstat to verify
    2. Channel can be reused by multiple stream, each stream on the same channel has different stream ID, you can use wireshark capturing HTTP2 package to verify
    3. If channel is not shutdown, on linux system with default tcp keepalive setting, the TCP keepalive probe should be sent after 7200 seconds, you can use wireshark to see there is no TCP probe when no RPC call
    4. Refer to doc via link client side keepalive to know why should use PING frame to keepalive
    5. Both GRPC server and GRPC client has GRPC ping setting and the setting of the ping on both side should consider setting on another side. Below are the basic configuration on both side
      Detailed info please ref to https://grpc.github.io/grpc-java/javadoc/
      keepAliveTime
      keepAliveTimeout
      permitKeepAliveWithoutCalls
      permitKeepAliveTime
    6. From GRPC best practice, client and server should use PING rather than native TPC keepalive to keep the GRPC connection alive
    7. Refer to GRPC official doc via link GRPC Performance, get inspiration on how to improve performance
    8. Consider pooling stub and channel to improve GRPC client side performance and avoid max concurrent stream issue. Ref to Stream Concurrency
    9. ManagedChannel has getState API to get its state