数据结构

butil::atomic bvar

线程模型(Thread Module)

bthread是brpc实现的一套“协程”。传统意义上的协程是一个线程中开辟多个协程，也就是通常意义的N:1协程。而bthread是M:N的“协程”，每个bthread之间是平等的，所谓的M:N是指协程可以在线程间迁移。

使用bthread 用户态协程，结合N:M的调度模型，以及work-steal 抢占式调度实现；为bthread打造的butex同步组件，为实现高性能的RPC框架提供了较为高效和简洁的并发编程基础。

要实现M:N其中关键就是：工作窃取（Work Stealing）算法

bthread_concurrency

bthread-local

https://github.com/apache/incubator-brpc/blob/master/docs/cn/bthread.md

内存管理(Buffer & Memory Management)

buffer-ManangerMent通过IOBuf（非连续零拷贝缓存）方式减少数据传递处理过程中的拷贝。各种ThreadLocal的ResoucePool，Object Pool等来优化定长的短生命周期的申请和释放。

执行队列(exec queue):多生产者单消费者无锁队列，实现了多生产者单消费者之间的的高效通信，在实现单TCP连接复用，高效发送-接收数据上发挥了很重要的作用。

超时定时器(Timer Keeping): 使用高效的定时器管理超时的RPC，并进行回调。

避免过多编码：将数据Body作为Attachment的方式，而不需要编解码 baidu_std and hulu_pbrpc supports attachments which are sent along with messages and set by users to bypass serialization of protobuf.

使用说明

https://github.com/apache/incubator-brpc/blob/master/docs/cn/server.md

https://github.com/apache/incubator-brpc/blob/master/docs/cn/client.md

超时配置

timeout_ms connect_timeout_ms

ENOSERVICE              = 1001;  // Service not found
ENOMETHOD               = 1002;  // Method not found
EREQUEST                = 1003;  // Bad Request
ERPCAUTH                = 1004;  // Unauthorized, can't be called EAUTH
                                 // directly which is defined in MACOSX
ETOOMANYFAILS           = 1005;  // Too many sub calls failed
EPCHANFINISH            = 1006;  // [Internal] ParallelChannel finished
EBACKUPREQUEST          = 1007;  // Sending backup request
ERPCTIMEDOUT            = 1008;  // RPC call is timed out
EFAILEDSOCKET           = 1009;  // Broken socket
EHTTP                   = 1010;  // Bad http call
EOVERCROWDED            = 1011;  // The server is overcrowded
ERTMPPUBLISHABLE        = 1012;  // RtmpRetryingClientStream is publishable
ERTMPCREATESTREAM       = 1013;  // createStream was rejected by the RTMP server
EEOF                    = 1014;  // Got EOF
EUNUSED                 = 1015;  // The socket was not needed
ESSL                    = 1016;  // SSL related error
EH2RUNOUTSTREAMS        = 1017;  // The H2 socket was run out of streams
EREJECT                 = 1018;  // The Request is rejected

// Errno caused by server
EINTERNAL               = 2001;  // Internal Server Error
ERESPONSE               = 2002;  // Bad Response
ELOGOFF                 = 2003;  // Server is stopping
ELIMIT                  = 2004;  // Reached server's limit on resources
ECLOSE                  = 2005;  // Close socket initiatively
EITP                    = 2006;  // Failed Itp response

Backup request

先向其中一个server发送请求，如果在ChannelOptions.backup_request_ms 后还没回来，再向另一个server发送.之后哪个先回来就取哪个

流式RPC

https://brpc.apache.org/zh/docs/client/streaming-rpc/

熔断策略

https://github.com/apache/brpc/blob/master/docs/cn/circuit_breaker.md

熔断触发
熔断恢复

健康检查策略

https://github.com/apache/brpc/blob/master/docs/cn/load_balancing.md#%E5%81%A5%E5%BA%B7%E6%A3%80%E6%9F%A5 /health 健康健康接口到底是让谁调用的？在brpc源码里并没有它的调用之处。

请求录制

rpc_dump_dir，默认放置在./rpc_data/rpc_dump/

负载均衡策略

rr
wrr
random
wr
la locality-aware，优先选择延时低的下游
- 原理 https://brpc.apache.org/zh/docs/rpc-in-depth/locality-aware/
c_murmurhash

源码学习

::google::protobuf::Closure* done brpc::ClosureGuard done_guard(done);

server.AddService server.Start

BRPC