about curl_multi_poll

[xuquan316]

code:

for(;;){
      ...
      curl_multi_perform(...);
      ...
      mc = curl_multi_poll(m_multi_handle, NULL, 0, 1000, &nums_fds);
      if(mc == CURLM_OK && nums_fds >0){
            while((handle =curl_multi_info_read(....) )!= null){
                int64_t current_time_recv = now;
                if (resp->msg == CURLMSG_DONE){
                         task_queue.push(resp->easy_handle);  //  find map <CURL*,void* data>,data used callback
                }
            }
      }
      .....
}

size_t rest_read_callback(....){
       ....
       int64_t current_time_callback = now;    //The callback is called once and the message is received.
       ....
}

Message request interval is 100ms
test: current_time_recv - current_time_callback （> 1ms and <200ms，40ms delay for most) ， Is there something I'm not using correctly?

init func:
curl_multi_setopt(m_multi_handle, CURLMOPT_PIPELINING, CURLPIPE_MULTIPLEX);
curl_multi_setopt(m_multi_handle, CURLMOPT_MAX_HOST_CONNECTIONS, m_connect_num); //1 or 10，http1.1 dont use one
curl_multi_setopt(m_multi_handle, CURLMOPT_MAXCONNECTS, m_connect_num);

[xuquan316]

code:

for(;;){
        request_queue.pop(data);
        curl_easy_deal...
        curl_multi_add_handle(.....);
        ...
        curl_multi_perform(...);
        ...
        mc = curl_multi_poll(m_multi_handle, NULL, 0, 1000, &nums_fds);
        if(mc == CURLM_OK && nums_fds >0){
                while((handle =curl_multi_info_read(....) )!= null){
                        int64_t current_time_recv = now;
                        if (resp->msg == CURLMSG_DONE){
                        task_queue.push(resp->easy_handle); // find map <CURL*,void* data>,data used callback
                        curl_multi_remove_handle(...);
                }
        }
    }
    .....
}

size_t rest_read_callback(....){
        ....
        int64_t current_time_callback = now; //The callback is called once and the message is received.
        ....
}

Message request interval is 100ms
test: current_time_recv - current_time_callback （> 1ms and <200ms，40ms delay for most) ， Is there something I'm not using correctly?

init func:
curl_multi_setopt(m_multi_handle, CURLMOPT_PIPELINING, CURLPIPE_MULTIPLEX);
curl_multi_setopt(m_multi_handle, CURLMOPT_MAX_HOST_CONNECTIONS, m_connect_num); //1 or 10，http1.1 dont use one
curl_multi_setopt(m_multi_handle, CURLMOPT_MAXCONNECTS, m_connect_num);
another thread:
request_queue.push(data);
curl_multi_wakeup(m_multi_handle);

[jay]

Message request interval is 100ms
test: current_time_recv - current_time_callback （> 1ms and <200ms，40ms delay for most) ， Is there something I'm not using correctly?

I don't understand the question

[xuquan316]

Message request interval is 100ms
test: current_time_recv - current_time_callback （> 1ms and <200ms，40ms delay for most) ， Is there something I'm not using correctly?

I don't understand the question

Callback timing current_time_callback, and curl_multi_poll interface call timing current_time_recv, the time difference between the two 40ms

[bagder]

I would probably recommend you don't use such architecture/design: let the multi interface run in a single thread.

[jay]

Maybe curl_multi_wakeup would help?

[xuquan316]

Every time "curl" is reinitialized, the "wakeup api" is called. But the effect is the same.

[xuquan316]

Some of the data is provided by another thread, using a lockless queue, which uses "add perform poll remove" to the same effect.

[jay]

No information in this discussion indicates a bug in curl. Also, a lock free queue may have worse performance than locking.

[xuquan316]

Maybe curl_multi_wakeup would help?

Some of the data is provided by another thread, using a lockless queue, which uses "add perform poll remove" to the same effect.

I would probably recommend you don't use such architecture/design: let the multi interface run in a single thread.