多進程結構是實現那些服務內容不相關的服務器的os層面的天然抽象。 html
1. 容錯:若系統在提供服務時,服務單元之間界限清晰沒有或不多交互(例如瀏覽兩個不相關的網頁),而服務單元有很大可能出錯,但願出錯時不影響其餘單元。 linux
2. 高性能:服務器若支持多CPU或超線程,多線程沒法徹底利用機器性能,多進程則可讓服務器滿載。 ios
最近使用多進程開發流媒體服務器,性能很不錯,預計能支持1000個客戶端併發播放1000kbps,能跑到千兆網的極限。參考這篇文章:http://blog.csdn.net/winlinvip/article/details/7840285 。 nginx
支持高併發的多進程服務器,應遵循如下設計要點: chrome
- 並不是每一個connection新建一個進程,這樣支持的併發並不高。
- 每一個進程處理多個connection,使用event-base來管理這些鏈接。
- master由信號驅動,worker由epoll驅動(固然信號會讓epoll_wait返回)。
- 多進程用來利用多CPU硬件,因此按照業務邊界來劃分進程,或者就按CPU個數配置。
- 如有特殊業務邏輯,譬如chrome的站點抽象,多進程有提升系統容錯性的優勢。
- 每一個connection的fd(socket)使用非阻塞方式(即異步方式,參考 http://blog.csdn.net/winlinvip/article/details/7843000 )。
- 每一個進程是單線程的:event-based, non-blocking, async-io 這些足以免多線程。
- 關鍵點就是儘可能使用異步。
關於chrome的多進程結構的背景: shell
http://blog.chromium.org/2008/09/multi-process-architecture.html 編程
http://www.charlesreis.com/research/publications/eurosys-2009-talk.pdf?attredirects=0 瀏覽器
總結以下: bash
1. 應該對系統分析,先搞清楚系統的目標,根據目標劃分邊界,再設計系統結構。
例如瀏覽器模型中,多進程模型分爲:核心進程(存儲,網絡,UI)、插件進程、展示進程(DOM,JS引擎,HTML展現)。
這樣設計的緣由在於對於瀏覽器系統的分析————系統的邊界能夠定義在相關的頁面,即將可相互操做的頁面分爲Site組。
區分系統邊界的緣由在於瀏覽器的實際問題「網頁和腳本極易出錯致使崩潰「、」加載速度隨網頁增多變慢「。
因此,先應分析系統的主要設計目標(要解決的問題),而後分析邊界(可分解的子系統),最後才能設計系統的結構(進程的模型之類的)。
插件進程做爲單獨進程的緣由,也是由系統的目標決定的:插件在邊界上很明顯,獨立於HTML展示,並且也有崩潰的可能性。
核心進程一旦崩潰,整個系統也就崩潰了,因此核心進程的邊界是也避開那些容易崩潰的系統(譬如把插件獨立出去)。
2. 系統的目標和內涵:穩定性,性能。
穩定性:包括容錯性,內存管理,可問責性。
容錯性:即某個子系統崩潰時是否能不影響其餘系統。譬如某個頁面崩潰是否其餘頁面還能不受影響。
可問責性:即當系統出問題時,是否能快速定位是某個子系統,並採起措施解決問題。譬如,當瀏覽器佔用過多資源時(內存泄漏啦之類),是否能快速找到出問題的頁面,關閉它解決問題。
內存管理:是否能彙報各個子系統的內存使用狀況,內存泄漏時是否能重啓或關閉子系統解決問題。
性能:包括實時性,吞吐速度,多CPU加速。
實時性:即系統對於用戶的響應是否實時。譬如,A頁面如今很慢,但用戶在B頁面中操做是否能實時響應。
吞吐速度:當任務增多時系統的總體吞吐能力。譬如,開不少個頁面,系統是否還能性能良好。
多CPU加速:系統是否能利用多CPU的服務器加速。譬如,是否能利用24CPU的機器能力,仍是說24CPU能力和4CPU能力同樣?
3. 多進程架構會佔用更多內存。
4. Chrome其實支持單進程和多進程結構。 服務器
nginx也是多進程,雖然單個進程也能夠支持高併發,但在實際服務器上,通常仍是會使用多進程——這也是爲何用多進程的緣由之一。參考nginx的系統結構一文:http://www.aosabook.org/en/nginx.html (個人朋友雷總分享給個人好東西,感謝他)。
參考《Unix環境高級編程》中對fork的總結,即多進程的使用場景:
fork有兩種用法:
(1) 一個父進程但願複製本身,使父、子進程同時執行不一樣的代碼段。父進程偵聽請求,當請求到達時,父進程fork子進程處理此請求,父進程繼續偵聽下一個請求。
(2) 一個進程要執行一個不一樣的程序。這對shell是常見的狀況。在這種狀況下,子進程在從fork返回後當即調用exec。
其實也提到了服務器的應用。可見網絡服務器使用多進程是有很長曆史了。
兩個多進程服務器原型:http://blog.csdn.net/winlinvip/article/details/7764526
linux多進程服務器真的很給力,贊一個!
寫了一個原型,讓服務器保持10200個穩定鏈接的原型。
開啓了15個worker進程(伺服進程):
ps -axf
1540 ? Ss 0:00 /usr/sbin/sshd
1959 ? Ss 0:00 \_ sshd: winlin [priv]
1961 ? S 0:00 \_ sshd: winlin@pts/0
1962 pts/0 Ss 0:00 \_ -bash
30127 pts/0 S+ 0:00 \_ ./multiple_process multi
30128 pts/0 S+ 0:00 \_ ./multiple_process multi
30129 pts/0 S+ 0:00 \_ ./multiple_process multi
30130 pts/0 S+ 0:00 \_ ./multiple_process multi
30131 pts/0 S+ 0:00 \_ ./multiple_process multi
30132 pts/0 S+ 0:00 \_ ./multiple_process multi
30133 pts/0 S+ 0:00 \_ ./multiple_process multi
30134 pts/0 S+ 0:00 \_ ./multiple_process multi
30135 pts/0 S+ 0:00 \_ ./multiple_process multi
30136 pts/0 S+ 0:00 \_ ./multiple_process multi
30137 pts/0 S+ 0:00 \_ ./multiple_process multi
30138 pts/0 S+ 0:00 \_ ./multiple_process multi
30139 pts/0 S+ 0:00 \_ ./multiple_process multi
30140 pts/0 S+ 0:00 \_ ./multiple_process multi
30141 pts/0 S+ 0:00 \_ ./multiple_process multi
30142 pts/0 S+ 0:00 \_ ./multiple_process multi
process #30128 serving 387 clients
process #30129 serving 732 clients
process #30130 serving 1020 clients
process #30131 serving 443 clients
process #30132 serving 437 clients
process #30133 serving 434 clients
process #30134 serving 883 clients
process #30135 serving 678 clients
process #30136 serving 958 clients
process #30137 serving 329 clients
process #30138 serving 758 clients
process #30139 serving 1020 clients
process #30140 serving 1020 clients
process #30141 serving 411 clients
process #30142 serving 690 clients
共有10200個client。
開啓了10個程序,每一個創建1020個鏈接:
[1] Running ./test_client 127.0.0.1 1990 1020 > /dev/null 2>&1 &
[2] Running ./test_client 127.0.0.1 1990 1020 > /dev/null 2>&1 &
[3] Running ./test_client 127.0.0.1 1990 1020 > /dev/null 2>&1 &
[4] Running ./test_client 127.0.0.1 1990 1020 > /dev/null 2>&1 &
[5] Running ./test_client 127.0.0.1 1990 1020 > /dev/null 2>&1 &
[6] Running ./test_client 127.0.0.1 1990 1020 > /dev/null 2>&1 &
[7] Running ./test_client 127.0.0.1 1990 1020 > /dev/null 2>&1 &
[8] Running ./test_client 127.0.0.1 1990 1020 > /dev/null 2>&1 &
[9]- Running ./test_client 127.0.0.1 1990 1020 > /dev/null 2>&1 &
[10]+ Running ./test_client 127.0.0.1 1990 1020 > /dev/null 2>&1 &
該模型能保持穩定的10200個鏈接,因爲每一個進程只支持1024個fd(可設置),因此每一個進程限定最多鏈接爲1020個fd。
- //multiple_process.cpp
-
- #include <stdio.h>
- #include <stdlib.h>
- #include <string.h>
- #include <iostream>
- #include <vector>
- using namespace std;
-
- // for socket
- #include <sys/types.h>
- #include <sys/socket.h>
- #include <netinet/in.h>
- // for select
- #include <sys/time.h>
- // for process control
- #include <unistd.h>
- #include <sys/wait.h>
-
- int prepare_server_socket(int server_port){
- int server_fd = socket(AF_INET, SOCK_STREAM, 0);
- if(server_fd == -1){
- cout << "init socket error" << endl;
- exit(1);
- }
- cout << "socket init success" << endl;
-
- int reuse_socket = 1;
- if(setsockopt(server_fd, SOL_SOCKET, SO_REUSEADDR, &reuse_socket, sizeof(int)) == -1){
- cout << "reuse socket error" << endl;
- exit(1);
- }
- cout << "socket set to reuse success" << endl;
-
- struct sockaddr_in addr;
- memset(&addr, 0, sizeof(sockaddr_in));
- addr.sin_family = AF_INET;
- addr.sin_port = htons(server_port);
- addr.sin_addr.s_addr = INADDR_ANY;
- if(bind(server_fd, (const struct sockaddr*)&addr, sizeof(addr)) == -1){
- cout << "bind socket error" << endl;
- exit(1);
- }
- cout << "socket bind success" << endl;
-
- if(listen(server_fd, 10) == -1){
- cout << "listen socket error" << endl;
- exit(1);
- }
- cout << "socket listen success" << endl;
-
- return server_fd;
- }
-
- bool can_socket_read(int socket_fd, int timeout_ms){
- fd_set set;
- FD_ZERO(&set);
- FD_SET(socket_fd, &set);
-
- timeval timeout;
- timeout.tv_sec = 0;
- timeout.tv_usec = timeout_ms * 1000;
-
- if(select(socket_fd + 1, &set, NULL, NULL, &timeout) == -1){
- cout << "select socket error " << endl;
- exit(1);
- }
-
- if(!FD_ISSET(socket_fd, &set)){
- return false;
- }
-
- return true;
- }
-
- vector<int> clients;
- void process_cycle(int server_fd){
- cout << "process #" << getpid() << " accept and process client from socket " << server_fd << endl;
-
- while(true){
- // accept
- while(clients.size() < 1020){
- if(!can_socket_read(server_fd, 500)){
- break;
- }
-
- //cout << "get a client, accept it" << endl;
- int client_fd = accept(server_fd, NULL, 0);
- if(client_fd == -1){
- cout << "failed to accept client" << endl;
- exit(1);
- }
- //cout << "accept client success: #" << client_fd << endl;
- clients.push_back(client_fd);
- }
- // process: PING.
- if(true){
- vector<int>::iterator ite;
- for(ite = clients.begin(); ite != clients.end(); ite++){
- int client_fd = *ite;
-
- if(!can_socket_read(client_fd, 1)){
- continue;
- }
-
- char buf[1024];
- if(recv(client_fd, buf, sizeof(buf), 0) <= 0){
- cout << "read client error, exit" << endl;
- exit(1);
- }
- //cout << "process #" << getpid() << " (" << clients.size() << " clients) recv from client: " << buf << endl;
- if(send(client_fd, buf, strlen(buf), 0) <= 0){
- cout << "send data error, exit" << endl;
- exit(1);
- }
- }
- }
- cout << "process #" << getpid() << " serving " << clients.size() << " clients" << endl;
- }
- }
-
- void start_worker_process(int server_fd, int process_num){
- for(int i = 0; i < process_num; i++){
- pid_t pid = fork();
- // child process
- if(pid == 0){
- process_cycle(server_fd);
- exit(0);
- }
- // parent process
- else if(pid == -1){
- cout << "fork child process error" << endl;
- exit(1);
- }
- else{
- cout << "fork child process success: #" << pid << endl;
- }
- }
- }
-
- int main(int argc, char** argv){
- if(argc <= 3){
- cout << "usage: " << argv[0] << " <port> <single_process_mode> <num_of_processes> " << endl
- << "port: the port to listen at" << endl
- << "single_process_mode: whether use the single process mode" << endl
- << "num_of_processes: if multiple processes mode, how many process we will create" << endl
- << "for example: " << argv[0] << " 1990 false 15" << endl;
- exit(1);
- }
- int server_port = atoi(argv[1]);
- char* single_process_mode = argv[2];
- int num_of_processes = atoi(argv[3]);
- bool is_single_process = !strcmp(single_process_mode, "true");
- cout << "[config] " << (is_single_process? "single process" : "multiple processes") << " mode, "
- << "listening at " << server_port
- << ", create " << num_of_processes << " worker processes" << endl;
-
- char* server_ip = argv[1];
- cout << "[remark] multiple processes prototype. " << endl
- << "[remark] master process listen to get a socket fd(file descriptor), worker process accept and serve client." << endl
- << "[remark] this prototype also show the architecture swithing between multiple and single process." << endl;
-
- // master process: bind and listen.
- int server_fd = prepare_server_socket(server_port);
-
- // worker process: accept and process.
- if(is_single_process){
- process_cycle(server_fd);
- }
- else{
- start_worker_process(server_fd, num_of_processes);
- }
-
- for(int i = 0; i < num_of_processes; i++){
- int status = 0;
- pid_t pid = waitpid(-1, &status, 0);
- cout << "child process #" << pid << " terminated" << endl;
- }
-
- cout << "server terminated" << endl;
- close(server_fd);
-
- return 0;
- }
- // test_client.cpp
-
- #include <stdio.h>
- #include <stdlib.h>
- #include <string.h>
- #include <iostream>
- #include <vector>
- using namespace std;
-
- #include <sys/types.h>
- #include <sys/socket.h>
- #include <netinet/ip.h>
- #include <arpa/inet.h>
-
- int main(int argc, char** argv)
- {
- if(argc <= 3){
- cout << "usage: " << argv[0] << " <server_ip> <server_port> <num_of_clients> " << endl
- << "server_ip: the ip of server to connect" << endl
- << "server_port: the port of server to connect" << endl
- << "num_of_clients: how many client we will start" << endl
- << "for example: " << argv[0] << " 127.0.0.1 1990 1020" << endl;
- exit(1);
- }
- char* server_ip = argv[1];
- int server_port = atoi(argv[2]);
- int num_of_clients = atoi(argv[3]);
-
- vector<int> clients;
- for(int i = 0; i < num_of_clients; i++){
- int client = socket(AF_INET, SOCK_STREAM, 0);
- if(client == -1){
- cout << "create socket error" << endl;
- exit(1);
- }
- cout << "create socket success" << endl;
-
- sockaddr_in addr;
- memset(&addr, 0, sizeof(addr));
- addr.sin_family = AF_INET;
- addr.sin_port = htons(server_port);
- addr.sin_addr.s_addr = inet_addr(server_ip);
-
- if(connect(client, (const struct sockaddr*)&addr, sizeof(addr)) == -1){
- cout << "connect to server error" << endl;
- exit(1);
- }
- cout << "connect server success" << endl;
-
- clients.push_back(client);
- }
-
- const char msg[] = "hello, this is client";
- char buf[1024];
- memset(buf, 0, sizeof(buf));
- memcpy(buf, msg, sizeof(msg));
-
- while(true){
- for(vector<int>::iterator ite = clients.begin(); ite != clients.end(); ite++){
- int client = *ite;
-
- if(send(client, buf, sizeof(msg), 0) < 0){
- cout << "send error" << endl;
- exit(1);
- }
- if(recv(client, buf, sizeof(buf), 0) < 0){
- cout << "recv error" << endl;
- exit(1);
- }
- cout << "recv from server: " << buf << endl;
-
- sleep(3);
- }
-
- sleep(120);
- }
-
- return 0;
- }