协程收包与tcp抓包的顺序偶尔乱序 about acl HOT 6 CLOSED

acl 协程底层也是通过 epoll 的事件触发的，上层虽是同步调用，在下层转成异步过程。你是否可以提供一个简单的示例演示一下？

from acl.

使用方式是线程+协程
//主线程accept连接
int main(int argc, char argv[])
{
acl::server_socket server;
if (server.open(sockAddr) == false) {
logger_error("open %s error\r\n", sockAddr);
return exit(1);
}
while (true) {
acl::socket_stream cstream = server.accept();
if (cstream == NULL) {
logger_error("accept failed: %s", acl::last_serror());
break;
}

    cstream->set_tcp_solinger(true, 10);
    cstream->set_tcp_nodelay(true);
    cstream->set_rw_timeout(0);
    cstream->set_tcp_non_blocking(false);
    unsigned int uiIdx = cstream->sock_handle() % g_CfgVar.uiMaxThreads;
    MPostMsg *ptrMsg = new MPostMsg(sockType, cstream);
    if (acl_mbox_send(g_mbox[uiIdx], ptrMsg) < 0) {
        delete cstream;
        delete ptrMsg;
    }   
}   

}

//启动两个MasterThread线程
void* MasterThread::run(void)
{
acl_fiber_create(fiber_callback, this, g_CfgVar.uiFiberStackSize);
acl_fiber_schedule();
return NULL;
}

//线程入口
void MasterThread::fiber_callback(ACL_FIBER * /* f */, void ctx)
{
MasterThread me = (MasterThread *)ctx;
me->mutex_ = acl_fiber_mutex_create();
for (;;) {
MPostMsg ptrMsg = (MPostMsg)acl_mbox_read(me->mbox_, 0, NULL);
if (ptrMsg == NULL)
continue;

    switch (ptrMsg->command)
    {
      case CMD_TCP_ID:
      {
        acl::socket_stream *client = (acl::socket_stream *)ptrMsg->ptrParam;
        if (g_CfgVar.uiMaxFibers > me->fibers_cnt_) {
            (new MasterTcpFiber(me, client))->start(g_CfgVar.uiFiberStackSize);
        }
        else {
            logger_error("fiber cnt=%d > max(%d), close client fd=%d", me->fibers_cnt_, g_CfgVar.uiMaxFibers, client->sock_handle());
            client->close();
            delete client;
        }
      }
  break;
  case ...:
  break;
}
delete ptrMsg;
ptrMsg = NULL;
 acl_fiber_mutex_free(me->mutex_);

}

// 协程入口，构造函数日志表示第一个socket在第二个socket之后才收到
MasterTcpFiber::MasterTcpFiber(MasterThread* t, acl::socket_stream* c)
:MFiberBase(t, c)
{
logger("accept fiber here delay...");
socket_stream_ = c;
buffer_ = acl_mymalloc(g_CfgVar.uiMaxReadBuffer);
acl_assert(buffer_ != NULL);
}

// 协程run
void MasterTcpFiber::run(void)
{
int ret = 0;
while(!socket_stream_->eof()){
ret = socket_stream_->read(buffer_, sizeof(NetPkgHeader));
...
}
}

from acl.

谢谢提供样例，我会研究测试一下。

from acl.

cstream->set_rw_timeout(0);
cstream->set_tcp_non_blocking(false);

阻塞模式设置rw timeout=0，源码中，会直接调用acl_socket_read()，就是 recv，会block住
int acl_socket_read(ACL_SOCKET fd, void *buf, size_t size,
int timeout, ACL_VSTREAM *fp, void *arg acl_unused)
{
#if 0
WSABUF wsaData;
DWORD dwBytes = 0;
DWORD flags = 0;
int ret;

    wsaData.len = (u_long) size;
    wsaData.buf = (char*) buf;
    ret = WSARecv(fd, &wsaData, 1, &dwBytes, &flags, NULL, NULL);
    if (ret == SOCKET_ERROR)
            return -1;
    if (dwBytes == 0)
            return -1;
    return dwBytes;

#else
if (fp != NULL && fp->read_ready)
fp->read_ready = 0;
else if (timeout > 0 && acl_read_wait(fd, timeout) < 0)
return -1;

    return recv(fd, buf, (int) size, 0);

#endif
}
应该是这里引起的，我修改下timeout再看看，谢谢

from acl.

我写了个类似的例子，没有发现你所说的问题，可参考：
https://github.com/acl-dev/demo/blob/master/thread_fiber_server.cpp
https://github.com/acl-dev/demo/blob/master/thread_fiber_client.cpp

from acl.

很详细的例子，谢谢，block阻塞模式下，设置timeout 2秒后，没有再出现了；后续修改为nonblock会不会致协程切换的频率相比block模式，过于频繁，增加context switch？

from acl.