Modify in process_queue(void* shard_num)

[ancientcc]

Code: Select all

if(min_threads && waiting_threads[shard] >= min_threads) {
        DBG_NW << "worker thread exiting... not enough jobs\n";
        to_clear[shard].push_back(threading::get_current_thread_id());
        return 0;
}

modify to

Code: Select all

if(outgoing_bufs[shard].empty() && min_threads && waiting_threads[shard] >= min_threads) {
        DBG_NW << "worker thread exiting... not enough jobs\n";
        to_clear[shard].push_back(threading::get_current_thread_id());
        return 0;
}

Add "outgoing_bufs[shard].empty()" to exit condition.

Why do it? when work in server mode, consider below command sequence. Assume min_threads = 2.

• Thread#1 detect outgoing_buf isn't empty, so set socket to SOCKET_LOCKED, and send data.
  More data push into outgoing_buf, and Thread#2 is waken up, but socket is in SOCKET_LOCKED, go to bed continue.
  More data push into outgoing_buf, and Thread#3 is waken up, but socket is in SOCKET_LOCKED, go to bed continue.
  Thread#1 finish send, goto while entry and decide whether exit self. Satisfy "min_threads && waiting_threads[shard] >= min_threads", Thread#1 exit!
  Thread#2 and Thred#3 are in bed, until other receive/send wake it up. But duration it, data in outgoing_buf is delay.

[Iris]

You might want to provide more context for your post and/or send us a patch or pull request. If there is a bug involved, either include a link to the bug report or file a new one. Also note that many of the C++ coders in the development team only occasionally use the forums and prefer IRC or the mailing list instead.