Pthread线程使用模型之三 客户端/服务端模型(Client/Server)
场景
1.在客户端/服务端模型时,客户端向服务端请求一些数据集的操作. 服务端执行执行操作独立的(多进程或跨网络)– 客户端可以等待服务端响应再做其他任务或者做一些并行的操作,在一段时间后被通知时再去查询结果. 虽然客户端等待服务端是最简单的方式, 但是这种情况极少使用, 因为它不具备速度和性能优势, 同步的只适合获取一些普通的资源. 比如Socket i/O的非阻塞或异步模型.
2.在做一些复杂的命令行程序时(单元测试),也需要这种模型.比如实现并发的方法测试,或者需要异步网络通讯的测试,或者通按键的功能执行(1: 执行导入. 2: 执行输出 q:执行退出)
说明
- Server的作用就是并发的处理请求并分发任务.
- Client可以同步等待或者异步执行, 在实际的项目开发中, 一般是Socket i/o来实现 Client和Server.
例子
1.程序server.c, 每个线程都在反复的读,接着输出, 输入几行. 当程序在运行时, 你能看到线程在不同的顺序输出提醒, 也有其他线程在输出之前先提示. 但是你将不会发现一个提示或一个输出在prompt服务器里穿插在提示和读取(stdin)之间.
2.request_t结构体定义了对服务器的每个请求. 未处理的requests使用next来连接成链表.
3.tty_server_t结构体提供server线程的上下文.它有两个同步对象(mutex and request). 一个标记位表明了这个服务是否正在运行, 同时一个requests列表表明了有哪些未处理的请求.
4.主程序和客户端线程使用同步对象client_mutex和clients_done来通讯, 而不是使用pthread_join.
5.创建detach线程的两种方法:
// 方法1:
pthread_t thread;
pthread_attr_t detached_attr;
status = pthread_attr_init (&detached_attr);
if (status != 0)
err_abort (status, "Init attributes object");
status = pthread_attr_setdetachstate (
&detached_attr, PTHREAD_CREATE_DETACHED);
if (status != 0)
err_abort (status, "Set detach state");
tty_server.running = 1;
status = pthread_create (&thread, &detached_attr,
tty_server_routine, NULL);
// 方法2:
pthread_t thread;
status = pthread_create (&thread, NULL,
tty_server_routine, NULL);
pthread_detach(thread);
/*
* server.c
*
* Demonstrate a client/server threading model.
*
* Special notes: On a Solaris system, call thr_setconcurrency()
* to allow interleaved thread execution, since threads are not
* timesliced.
*/
#include <pthread.h>
#include <math.h>
#include "errors.h"
#define CLIENT_THREADS 4 /* Number of clients */
#define REQ_READ 1 /* Read with prompt */
#define REQ_WRITE 2 /* Write */
#define REQ_QUIT 3 /* Quit server */
/*
* Internal to server "package" -- one for each request.
*/
typedef struct request_tag {
struct request_tag *next; /* Link to next */
int operation; /* Function code */
int synchronous; /* Non-zero if synchronous */
int done_flag; /* Predicate for wait */
pthread_cond_t done; /* Wait for completion */
char prompt[32]; /* Prompt string for reads */
char text[128]; /* Read/write text */
} request_t;
/*
* Static context for the server
*/
typedef struct tty_server_tag {
request_t *first;
request_t *last;
int running;
pthread_mutex_t mutex;
pthread_cond_t request;
} tty_server_t;
tty_server_t tty_server = {
NULL, NULL, 0,
PTHREAD_MUTEX_INITIALIZER, PTHREAD_COND_INITIALIZER};
/*
* Main program data
*/
int client_threads;
pthread_mutex_t client_mutex = PTHREAD_MUTEX_INITIALIZER;
pthread_cond_t clients_done = PTHREAD_COND_INITIALIZER;
/*
* The server start routine. It waits for a request to appear
* in tty_server.requests using the request condition variable.
* It processes requests in FIFO order. If a request is marked
* "synchronous" (synchronous != 0), the server will set done_flag
* and signal the request's condition variable. The client is
* responsible for freeing the request. If the request was not
* synchronous, the server will free the request on completion.
*/
void *tty_server_routine (void *arg)
{
static pthread_mutex_t prompt_mutex = PTHREAD_MUTEX_INITIALIZER;
request_t *request;
int operation, len;
int status;
while (1) {
status = pthread_mutex_lock (&tty_server.mutex);
if (status != 0)
err_abort (status, "Lock server mutex");
/*
* Wait for data
*/
while (tty_server.first == NULL) {
status = pthread_cond_wait (
&tty_server.request, &tty_server.mutex);
if (status != 0)
err_abort (status, "Wait for request");
}
request = tty_server.first;
tty_server.first = request->next;
if (tty_server.first == NULL)
tty_server.last = NULL;
status = pthread_mutex_unlock (&tty_server.mutex);
if (status != 0)
err_abort (status, "Unlock server mutex");
/*
* Process the data
*/
operation = request->operation;
switch (operation) {
case REQ_QUIT:
break;
case REQ_READ:
if (strlen (request->prompt) > 0)
printf (request->prompt);
if (fgets (request->text, 128, stdin) == NULL)
request->text[0] = '\0';
/*
* Because fgets returns the newline, and we don't want it,
* we look for it, and turn it into a null (truncating the
* input) if found. It should be the last character, if it is
* there.
*/
len = strlen (request->text);
if (len > 0 && request->text[len-1] == '\n')
request->text[len-1] = '\0';
break;
case REQ_WRITE:
puts (request->text);
break;
default:
break;
}
if (request->synchronous) {
status = pthread_mutex_lock (&tty_server.mutex);
if (status != 0)
err_abort (status, "Lock server mutex");
request->done_flag = 1;
status = pthread_cond_signal (&request->done);
if (status != 0)
err_abort (status, "Signal server condition");
status = pthread_mutex_unlock (&tty_server.mutex);
if (status != 0)
err_abort (status, "Unlock server mutex");
} else
free (request);
if (operation == REQ_QUIT)
break;
}
return NULL;
}
5.tty_server_routine用来持续处理客户端请求直到退出请求发出. 如果一个请求标记为同步(synchronous不为0),那么server设置done_flag的值并且通知done条件变量. 当请求是同步的, 那么客户端有责任去释放request包. 如果请求是异步的, 那么server在完成request时会释放.
void tty_server_request (
int operation,
int sync,
const char *prompt,
char *string)
{
request_t *request;
int status;
status = pthread_mutex_lock (&tty_server.mutex);
if (status != 0)
err_abort (status, "Lock server mutex");
if (!tty_server.running) {
pthread_t thread;
pthread_attr_t detached_attr;
status = pthread_attr_init (&detached_attr);
if (status != 0)
err_abort (status, "Init attributes object");
status = pthread_attr_setdetachstate (
&detached_attr, PTHREAD_CREATE_DETACHED);
if (status != 0)
err_abort (status, "Set detach state");
tty_server.running = 1;
status = pthread_create (&thread, &detached_attr,
tty_server_routine, NULL);
if (status != 0)
err_abort (status, "Create server");
/*
* Ignore an error in destroying the attributes object.
* It's unlikely to fail, there's nothing useful we can
* do about it, and it's not worth aborting the program
* over it.
*/
pthread_attr_destroy (&detached_attr);
}
/*
* Create and initialize a request structure.
*/
request = (request_t*)malloc (sizeof (request_t));
if (request == NULL)
errno_abort ("Allocate request");
request->next = NULL;
request->operation = operation;
request->synchronous = sync;
if (sync) {
request->done_flag = 0;
status = pthread_cond_init (&request->done, NULL);
if (status != 0)
err_abort (status, "Init request condition");
}
if (prompt != NULL)
strncpy (request->prompt, prompt, 32);
else
request->prompt[0] = '\0';
if (operation == REQ_WRITE && string != NULL)
strncpy (request->text, string, 128);
else
request->text[0] = '\0';
/*
* Add the request to the queue, maintaining the first and
* last pointers.
*/
if (tty_server.first == NULL) {
tty_server.first = request;
tty_server.last = request;
} else {
(tty_server.last)->next = request;
tty_server.last = request;
}
/*
* Tell the server that a request is available.
*/
status = pthread_cond_signal (&tty_server.request);
if (status != 0)
err_abort (status, "Wake server");
/*
* If the request was "synchronous", then wait for a reply.
*/
if (sync) {
while (!request->done_flag) {
status = pthread_cond_wait (
&request->done, &tty_server.mutex);
if (status != 0)
err_abort (status, "Wait for sync request");
}
if (operation == REQ_READ) {
if (strlen (request->text) > 0)
strcpy (string, request->text);
else
string[0] = '\0';
}
status = pthread_cond_destroy (&request->done);
if (status != 0)
err_abort (status, "Destroy request condition");
free (request);
}
status = pthread_mutex_unlock (&tty_server.mutex);
if (status != 0)
err_abort (status, "Unlock mutex");
}6.tty_server_request 用来初始化一个发送到tty server的请求. 如果tty server线程没有在运行, 启动一个, 并创建一个属性对象(detached_attr)， 并设置detachstate属性值为 PTHREAD_CREATE_DETACHED.
PTHREAD_CREATE_JOINABLE – pthread_create创建时默认就是joinable, 意味着通过pthread_create创建的线程id能被用来join with这个线程,查询线程的返回值或者取消它.
PTHREAD_CREATE_DETACHED – 通过pthread_create创建的id是不能再用饿了,当线程终止时, 任何被线程使用的资源都会被系统迅速回收. 也可以在pthread_create之后通过pthread_detach转换为detach线程. 建议如果不需要控制线程, 就把它设置为detach.
7.分配和初始化一个server的请求(request_t)包. 如果请求是同步的, 初始化request里的条件变量done(注意配合使用的还是server的mutex,这里和server共用mutex,如果是同步请求的话,server只能等待client的请求完成并释放mutex才能继续执行下一个request,效率并不高)
/*
* Client routine -- multiple copies will request server.
*/
void *client_routine (void *arg)
{
int my_number = (int)arg, loops;
char prompt[32];
char string[128], formatted[128];
int status;
sprintf (prompt, "Client %d> ", my_number);
while (1) {
tty_server_request (REQ_READ, 1, prompt, string);
if (strlen (string) == 0)
break;
for (loops = 0; loops < 4; loops++) {
sprintf (
formatted, "(%d#%d) %s", my_number, loops, string);
tty_server_request (REQ_WRITE, 0, NULL, formatted);
sleep (1);
}
}
status = pthread_mutex_lock (&client_mutex);
if (status != 0)
err_abort (status, "Lock client mutex");
client_threads--;
if (client_threads <= 0) {
status = pthread_cond_signal (&clients_done);
if (status != 0)
err_abort (status, "Signal clients done");
}
status = pthread_mutex_unlock (&client_mutex);
if (status != 0)
err_abort (status, "Unlock client mutex");
return NULL;
}
int main (int argc, char *argv[])
{
pthread_t thread;
int count;
int status;
#ifdef sun
/*
* On Solaris 2.5, threads are not timesliced. To ensure
* that our threads can run concurrently, we need to
* increase the concurrency level to CLIENT_THREADS.
*/
DPRINTF (("Setting concurrency level to %d\n", CLIENT_THREADS));
thr_setconcurrency (CLIENT_THREADS);
#endif
/*
* Create CLIENT_THREADS clients.
*/
client_threads = CLIENT_THREADS;
for (count = 0; count < client_threads; count++) {
status = pthread_create (&thread, NULL,
client_routine, (void*)count);
if (status != 0)
err_abort (status, "Create client thread");
}
status = pthread_mutex_lock (&client_mutex);
if (status != 0)
err_abort (status, "Lock client mutex");
while (client_threads > 0) {
status = pthread_cond_wait (&clients_done, &client_mutex);
if (status != 0)
err_abort (status, "Wait for clients to finish");
}
status = pthread_mutex_unlock (&client_mutex);
if (status != 0)
err_abort (status, "Unlock client mutex");
printf ("All clients done\n");
tty_server_request (REQ_QUIT, 1, NULL, NULL);
return 0;
}