Nginx理解
本文主要帮助理解 Linux操作系统下Nginx 的设计和实现，对源码理解也有一定的帮助。相关代码来自nginx GitHub源码仓上的release-1.16.0。个人观点，仅供参考～
1. Nginx是什么&有些什么功能、特点
基本、常用的http服务器功能
- 静态文件服务、首页设置
- 反向代理、负载均衡
- FastCGI协议支持、uwsgi协议支持
- https、http/2、虚拟主机支持
- 更多
邮件代理服务器功能
不常用，详见：mail_proxy_server_features
TCP/UDP代理服务器功能
不常用，详见：generic_proxy_server_features
架构和可扩展性、跨平台
- master/worker进程模型
- 模块化设计，功能文档基本上可与代码文件一一对应，阅读、理解源码更方便，开发新功能模块更容易
- 灵活的配置，可以说Nginx的所有功能都是配置驱动的
- 充分利用操作系统提供的高性能API：异步io（事件驱动）等
- 多操作系统支持
2. socket server编程&http协议(tcp/ip)
tcp协议为上层提供数据传输功能，不关心传输的数据的格式；http协议规定了数据的格式、通讯双方（客户端、web服务器程序）的交互，典型的“请求-应答”模式（websocket和http/2打破了这种模式）。这里的数据就是我们写的程序发送的或者接收到的数据。何为协议也需要我们好好思考的，平时定义的json接口也属于协议，同样的json数据可以使用tcp协议发送，也可以使用http协议发送，而http协议又是在tcp协议之上的。
浏览器发送http请求示例（为方便显示\r\n后面加了换行）：
GET / HTTP/1.1\r\n
Host: 127.0.0.1\r\n
Connection: keep-alive\r\n
Cache-Control: max-age=0\r\n
Upgrade-Insecure-Requests: 1\r\n
User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10_13_5) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/73.0.3683.103 Safari/537.36\r\n
Accept: text/html,application/xhtml+xml,application/xml;q=0.9,image/webp,image/apng,*/*;q=0.8,application/signed-exchange;v=b3\r\n
Accept-Encoding: gzip, deflate, br\r\n
Accept-Language: zh-CN,zh;q=0.9,en;q=0.8\r\n\r\n
http服务器应答示例（为方便显示\r\n后面加了换行）：
HTTP/1.1 301 Moved Permanently\r\n
Server: GitHub.com\r\n
Content-Type: text/html\r\n
Location: https://gobuild.tech/\r\n
X-GitHub-Request-Id: F368:05D3:205F88:230D33:5CC40FF1\r\n
Content-Length: 164\r\n
Accept-Ranges: bytes\r\n
Date: Sat, 27 Apr 2019 08:17:00 GMT\r\n
Via: 1.1 varnish\r\n
Age: 11\r\n
Connection: keep-alive\r\n
X-Served-By: cache-hnd18720-HND\r\n
X-Cache: HIT\r\n
X-Cache-Hits: 1\r\n
X-Timer: S1556353020.348737,VS0,VE1\r\n
Vary: Accept-Encoding\r\n
X-Fastly-Request-ID: 20026dc4a1797ecdb889bd0c6ab786f6ab351e65\r\n\r\n
<html><head><title>301 Moved Permanently</title></head><body bgcolor="white"><center><h1>301 Moved Permanently</h1></center><hr><center>nginx</center></body></html>
一个简单的http服务器
#coding:utf-8
#python3 simple-http-server.py
import socket
server_address = ('127.0.0.1', 80)
server_socket = socket.socket(family=socket.AF_INET, type=socket.SOCK_STREAM)
server_socket.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR , True) # 避免测试时发生 Address already in use 错误
server_socket.bind(server_address)
server_socket.listen(8)
print('serving http://%s:%d' % server_address)
while True:
client_socket, client_address = server_socket.accept()
print('client', client_address, 'connected')
data = client_socket.recv(1024)
if not data:
print('client', client_address, 'maybe close')
client_socket.close()
continue
print('client', client_address, 'send:')
print(data)
client_socket.send(b'HTTP/1.1 200 OK\r\n') # 状态行
client_socket.send(b'Connection: Close\r\n') # 响应头
client_socket.send(b'\r\n') # 响应头结束
client_socket.send(b'<h1>It works!</h1>') # 响应体
client_socket.close()
思考：
- 如何解析&存放请求头
- 如何根据请求头执行不同的处理逻辑
- 如何生成并发送响应（读取文件并发送文件内容等）
- 如何处理并发
一个简单的http客户端
#-*- coding:utf-8 -*-
# python3 simple-http-client.py
import socket
client = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
server_address = ('185.199.111.153', 80) # 可域名直连，但python解析到的ip可能连不上
client.connect(server_address)
client.send(b'GET / HTTP/1.1\r\n')
client.send(b'Host: gobuild.tech\r\n')
client.send(b'\r\n')
print('recv:')
print(client.recv(1024))
可以结合两个示例实现简单的代理功能。
相关链接：
一个简单的异步（事件驱动）http服务器示例
#-*- coding:utf-8 -*-
# python3 simple-epoll-http-server.py
import socket
import select
server_address = ('127.0.0.1', 80)
server_socket = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
server_socket.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, True)
server_socket.bind(server_address)
server_socket.listen(8)
server_socket.setblocking(False)
#创建epoll事件对象，后续要监控的事件添加到其中
epoll = select.epoll()
#注册服务器监听fd到等待读事件集合，server_socket是用来accept新连接的，不能读/写数据
epoll.register(server_socket.fileno(), select.EPOLLIN)
#用于关联事件和socket对象；和C语言不同，注册事件时不能关联上下文
fd_to_socket = {server_socket.fileno(): server_socket, }
while True:
print("http://%s:%d waiting connection active......" % server_address)
#轮询注册的事件集合，返回值为[(文件句柄，对应的事件)，(...),....]
events = epoll.poll(-1) # 永不超时
if not events:
print("epoll.poll error")
break
for fd, event in events:
#如果活动socket为当前服务器socket，表示有新连接
if fd == server_socket.fileno():
connection, client_address = server_socket.accept()
print(client_address, 'connected')
connection.setblocking(False)
client_fd = connection.fileno()
fd_to_socket[client_fd] = connection
epoll.register(client_fd, select.EPOLLIN) # 只添加读事件
continue
connection = fd_to_socket[fd]
close_reason = 'close because of network event'
if event & select.EPOLLIN: # 可读事件
print(connection.getpeername(), 'can read')
data = connection.recv(1024)
if data:
print(connection.getpeername(), 'send:')
print(data)
epoll.modify(fd, select.EPOLLOUT)
continue
#else: # 关闭
close_reason = 'close because of recv error'
if event & select.EPOLLOUT: # 可写事件
print(connection.getpeername(), 'can write')
#发送响应数据
connection.send(b'HTTP/1.1 200 OK\r\n')
connection.send(b'Connection: close\r\n')
connection.send(b'\r\n')
connection.send(b'<h1>It Works!</h1>')
# 数据可能只写到操作系统的缓冲区，后面的close调用会尝试把数据发送到网络
close_reason = 'close after send response data'
#if event & select.EPOLLHUP: # 关闭事件
print(connection.getpeername(), close_reason)
epoll.unregister(fd)
connection.close()
del fd_to_socket[fd]
epoll.unregister(server_socket.fileno())
epoll.close()
server_socket.close()
思考：
- Nginx会在哪个进程调用listen、在哪个进程调用accept、如何均衡每个进程上的连接数（Nginx的进程模型）
- 事件驱动给代码编写（组织）带来哪些挑战（可对比前面两个示例，发送或接收的数据过大可能需要多次调用send/recv）
- 事件处理逻辑代码执行时间过长会有什么影响
相关链接：
3. 开发指南
内容主要摘自官方Development guide，参考淘宝团队的Nginx开发从入门到精通-nginx平台初探
内置数据结构
模块开发
- 理解请求处理的11个阶段
- 模块编译配置、模块类型（–add-module参数所需）
- 配置指令定义
- 模块功能实现
代码行数最短的Nginx模块：empty_gif，注意指令定义和如何关联模块功能。
// 摘自 https://github.com/nginx/nginx/blob/release-1.16.0/src/http/modules/ngx_http_empty_gif_module.c
/*
* Copyright (C) Igor Sysoev
* Copyright (C) Nginx, Inc.
*/
#include <ngx_config.h>
#include <ngx_core.h>
#include <ngx_http.h>
static char *ngx_http_empty_gif(ngx_conf_t *cf, ngx_command_t *cmd,
void *conf);
static ngx_command_t ngx_http_empty_gif_commands[] = {
{ ngx_string("empty_gif"),
NGX_HTTP_LOC_CONF|NGX_CONF_NOARGS,
ngx_http_empty_gif,
0,
0,
NULL },
ngx_null_command
};
/* the minimal single pixel transparent GIF, 43 bytes */
static u_char ngx_empty_gif[] = {
'G', 'I', 'F', '8', '9', 'a', /* header */
/* logical screen descriptor */
0x01, 0x00, /* logical screen width */
0x01, 0x00, /* logical screen height */
0x80, /* global 1-bit color table */
0x01, /* background color #1 */
0x00, /* no aspect ratio */
/* global color table */
0x00, 0x00, 0x00, /* #0: black */
0xff, 0xff, 0xff, /* #1: white */
/* graphic control extension */
0x21, /* extension introducer */
0xf9, /* graphic control label */
0x04, /* block size */
0x01, /* transparent color is given, */
/* no disposal specified, */
/* user input is not expected */
0x00, 0x00, /* delay time */
0x01, /* transparent color #1 */
0x00, /* block terminator */
/* image descriptor */
0x2c, /* image separator */
0x00, 0x00, /* image left position */
0x00, 0x00, /* image top position */
0x01, 0x00, /* image width */
0x01, 0x00, /* image height */
0x00, /* no local color table, no interlaced */
/* table based image data */
0x02, /* LZW minimum code size, */
/* must be at least 2-bit */
0x02, /* block size */
0x4c, 0x01, /* compressed bytes 01_001_100, 0000000_1 */
/* 100: clear code */
/* 001: 1 */
/* 101: end of information code */
0x00, /* block terminator */
0x3B /* trailer */
};
static ngx_http_module_t ngx_http_empty_gif_module_ctx = {
NULL, /* preconfiguration */
NULL, /* postconfiguration */
NULL, /* create main configuration */
NULL, /* init main configuration */
NULL, /* create server configuration */
NULL, /* merge server configuration */
NULL, /* create location configuration */
NULL /* merge location configuration */
};
ngx_module_t ngx_http_empty_gif_module = {
NGX_MODULE_V1,
&ngx_http_empty_gif_module_ctx, /* module context */
ngx_http_empty_gif_commands, /* module directives */
NGX_HTTP_MODULE, /* module type */
NULL, /* init master */
NULL, /* init module */
NULL, /* init process */
NULL, /* init thread */
NULL, /* exit thread */
NULL, /* exit process */
NULL, /* exit master */
NGX_MODULE_V1_PADDING
};
static ngx_str_t ngx_http_gif_type = ngx_string("image/gif");
static ngx_int_t
ngx_http_empty_gif_handler(ngx_http_request_t *r)
{
ngx_http_complex_value_t cv;
if (!(r->method & (NGX_HTTP_GET|NGX_HTTP_HEAD))) {
return NGX_HTTP_NOT_ALLOWED;
}
ngx_memzero(&cv, sizeof(ngx_http_complex_value_t));
cv.value.len = sizeof(ngx_empty_gif);
cv.value.data = ngx_empty_gif;
r->headers_out.last_modified_time = 23349600;
return ngx_http_send_response(r, NGX_HTTP_OK, &ngx_http_gif_type, &cv);
}
static char *
ngx_http_empty_gif(ngx_conf_t *cf, ngx_command_t *cmd, void *conf)
{
ngx_http_core_loc_conf_t *clcf;
clcf = ngx_http_conf_get_module_loc_conf(cf, ngx_http_core_module);
clcf->handler = ngx_http_empty_gif_handler;
return NGX_CONF_OK;
}