继续《Linux下使用ZMQ实践“生产者-消费者”模型》 文章之后进一步思考:
ZeroMQ通过隐藏了基础的socket操作,达到调用简明易懂的层次;
那么,如果某些场景下,又需要考虑到连接状态的维护,应该如何操作?
ZeroMQ给出的解决方案就是zmq_socket_monitor
支持监控的事件:
事件 | 描述 |
---|---|
ZMQ_EVENT_CONNECTED | socket已被成功连接 |
ZMQ_EVENT_CONNECT_DELAYED | 连接动作被挂起 |
ZMQ_EVENT_CONNECT_RETRIED | 连接失败,正在重试 |
ZMQ_EVENT_LISTENING | 监听成功 |
ZMQ_EVENT_BIND_FAILED | 绑定失败 |
ZMQ_EVENT_ACCEPTED | 接受新连接 |
ZMQ_EVENT_ACCEPT_FAILED | 接受新连接失败 |
ZMQ_EVENT_CLOSED | socket关闭(主动关闭) |
ZMQ_EVENT_CLOSE_FAILED | socket关闭失败 |
ZMQ_EVENT_DISCONNECTED | 连接意外关闭(被关闭) |
ZMQ_EVENT_MONITOR_STOPPED | 监控的socket消亡 |
使用思路:将要监听的sock跟monitor关联,然后创建一个额外的ZMQ_PAIR,通过pair来获取sock上的事件。
根据之前的“生产者-消费者”模型的一个改进:
在之前一对多的 Push-Pull 模式下,如果没有消费者连接,则生产者数据发送会一直阻塞,但如果有至少一个连接成功,则生产者进入发送数据阶段;在改进场景中,需求所有消费者就绪后,生产者才正式开始发送数据,达到一个理想的均衡状态。
这样,我们就依赖monitor机制的实现,监听消费者的Push套件,额外增加一个监听器monitor:
#define ADDR "tcp://127.0.0.1:555"
#define MONITOR "inproc://monitor-server"
...
void *sock = zmq_socket(ctx, ZMQ_PUSH);
void *mon = zmq_socket(ctx, ZMQ_PAIR);
...
zmq_bind(sock, ADDR);
zmq_socket_monitor(sock, MONITOR, ZMQ_EVENT_ALL);
zmq_connect(mon, MONITOR);
...
下来,我们通过monitor等待4个消费者的连接事件,成功后才发送数据;
发送数据完成后,我们通过发送“Quit”报文来通知消费者退出进程;
完整的生产者代码如下:
void test_producer(void *ctx, int times)
{
int ix = 0, cnt = 0, id = 0, event = 0;
char request[1024];
void *sock = zmq_socket(ctx, ZMQ_PUSH);
void *mon = zmq_socket(ctx, ZMQ_PAIR);
s_set_id_num(sock, id);
zmq_bind(sock, ADDR);
zmq_socket_monitor(sock, MONITOR, ZMQ_EVENT_ALL);
zmq_connect(mon, MONITOR);
LOGN("Producer %d setup\n", id);
for (cnt = 0; cnt < 4;) {
event = get_monitor_event(mon, NULL, NULL);
if (event == ZMQ_EVENT_ACCEPTED) {
LOGN("Producer accepted\n");
cnt++;
}
}
LOGN("Producer %d start\n", id);
for (ix = 0; ix < times; ix++) {
snprintf(request, sizeof(request), "Data-%03d-%03d", id, ix);
s_send(sock, request);
LOGN("Producer %d send: %s\n", id, request);
usleep(100 * 1000);
}
for (cnt = 0; cnt < 4;) {
s_send(sock, "Quit"); // 通知一个消费者,退出一个消费者
event = get_monitor_event(mon, NULL, NULL);
if (event == ZMQ_EVENT_DISCONNECTED) {
cnt++;
}
}
LOGN("Producer %d stop\n", id);
zmq_close(sock);
}
获取监听事件的接口为,get_monitor_event
,该函数从ZeroMQ帮助手册摘抄下来:
static int get_monitor_event (void *monitor, int *value, char **address)
{
// First frame in message contains event number and value
zmq_msg_t msg;
zmq_msg_init (&msg);
if (zmq_msg_recv (&msg, monitor, 0) == -1)
return -1; // Interrupted, presumably
assert (zmq_msg_more (&msg));
uint8_t *data = (uint8_t *) zmq_msg_data (&msg);
uint16_t event = *(uint16_t *) (data);
if (value)
*value = *(uint32_t *) (data + 2);
// Second frame in message contains event address
zmq_msg_init (&msg);
if (zmq_msg_recv (&msg, monitor, 0) == -1)
return -1; // Interrupted, presumably
assert (!zmq_msg_more (&msg));
if (address) {
uint8_t *data = (uint8_t *) zmq_msg_data (&msg);
size_t size = zmq_msg_size (&msg);
*address = (char *) malloc (size + 1);
memcpy (*address, data, size);
(*address)[size] = 0;
}
return event;
}
然后消费者的实现,跟先前的例子差不多,就多了一个退出的判断:
int test_consumer(void *ctx, int id)
{
int cnt = 0;
char request[1024];
void *sock = zmq_socket(ctx, ZMQ_PULL);
s_set_id_num(sock, id);
zmq_connect(sock, ADDR);
LOGN("Consumer %d start\n", id);
while (++cnt) {
s_recv(sock, request);
LOGN("Consumer %d recv: %s\n", id, request);
usleep(300 * 1000);
if (strcmp(request, "Quit") == 0) {
break;
}
}
LOGN("Consumer %d stop\n", id);
zmq_close(sock);
}
最后,main函数功能,主要为fork,主进程做生产者,子进程做消费者;
同时,为了方便起见,省略了waitpid回收子进程的动作;
int main(int argc, char *argv[])
{
int ix = 0;
void *ctx = zmq_ctx_new();
srandom(time(NULL));
/* 1x producter vs 4x consumer */
for (ix= 1; ix <= 4; ix++) {
pid_t pid = fork();
if (pid == 0) {
test_consumer(ctx, ix);
goto out;
}
}
test_producer(ctx, atoi(argv[1]));
// TODO waitpid
out:
zmq_ctx_destroy(ctx);
exit(EXIT_SUCCESS);
}
实际运行情况如下:
[ 1561228921.433 ]: Consumer 1 start
[ 1561228921.433 ]: Consumer 2 start
[ 1561228921.434 ]: Consumer 4 start
[ 1561228921.434 ]: Consumer 3 start
[ 1561228921.434 ]: Producer 0 setup
[ 1561228921.435 ]: Producer accepted
[ 1561228921.496 ]: Producer accepted
[ 1561228921.572 ]: Producer accepted
[ 1561228921.572 ]: Producer accepted
[ 1561228921.572 ]: Producer 0 start
[ 1561228921.572 ]: Producer 0 send: Data-000-000
[ 1561228921.574 ]: Consumer 3 recv: Data-000-000
[ 1561228921.673 ]: Producer 0 send: Data-000-001
[ 1561228921.774 ]: Producer 0 send: Data-000-002
[ 1561228921.775 ]: Consumer 2 recv: Data-000-002
[ 1561228921.876 ]: Producer 0 send: Data-000-003
[ 1561228921.877 ]: Consumer 1 recv: Data-000-003
[ 1561228921.978 ]: Producer 0 send: Data-000-004
[ 1561228921.978 ]: Consumer 4 recv: Data-000-004
[ 1561228922.079 ]: Producer 0 send: Data-000-005
[ 1561228922.081 ]: Consumer 3 recv: Data-000-005
[ 1561228922.183 ]: Producer 0 send: Data-000-006
[ 1561228922.284 ]: Producer 0 send: Data-000-007
[ 1561228922.285 ]: Consumer 2 recv: Data-000-007
[ 1561228922.386 ]: Producer 0 send: Data-000-008
[ 1561228922.387 ]: Consumer 1 recv: Data-000-008
[ 1561228922.488 ]: Producer 0 send: Data-000-009
[ 1561228922.488 ]: Consumer 4 recv: Data-000-009
[ 1561228922.590 ]: Consumer 3 recv: Quit
[ 1561228922.892 ]: Consumer 3 stop
[ 1561228922.894 ]: Consumer 2 recv: Quit
[ 1561228923.195 ]: Consumer 2 stop
[ 1561228923.196 ]: Consumer 1 recv: Quit
[ 1561228923.497 ]: Consumer 1 stop
[ 1561228923.499 ]: Consumer 4 recv: Quit
[ 1561228923.800 ]: Consumer 4 stop
[ 1561228923.802 ]: Producer 0 stop
可以看出,程序第一阶段,启动进程;第二阶段,发送数据,负载均衡;第三阶段,回收资源。
ZMQ监控事件的方法,提供了一种可选的扩展场景支持,实际使用可以放主线程处理,也可以放独立的子线程处理。