Worker Error and Training Failure on Running Script

[zxkyjimmy]

Hi, I attempted to run your script but encountered an unexpected error. The attached terminal output screenshot indicated that the training process failed to complete.

Below is the command line input that led to the error:

scripts/start-container.sh
tools/quick-run.sh train go az 300 -n go_9x9_az_n200 -conf_str env_board_size=9:actor_num_simulation=200

Please help me understand what is going wrong or provide any guidance on resolving this issue.

Thank you so much for your help.

[moporgic]

Hi, thanks for your report.

This is a known issue caused by the zero server when it receives an unexpected incoming message.
Since the zero server opens a public port for worker connections, any client (including unwanted clients) may connect to the server and send a message to it.

In your case, an unexpected message "_\sFjiS$NqFY:A'*+/." was sent to the server, which triggered a false alarm.
It is likely that this message was sent by a security software, which was checking network vulnerabilities just during the training.

This issue will be fixed in a future release. At current, a workaround is to remove these lines:

minizero/minizero/zero/zero_server.cpp

Lines 129 to 133 in 3c0a9d6

	std::string error_message = message;
	std::replace(error_message.begin(), error_message.end(), '\r', ' ');
	std::replace(error_message.begin(), error_message.end(), '\n', ' ');
	shared_data_.logger_.addWorkerLog("[Worker Error] \"" + error_message + "\"");
	close();

PS, you can run the same command and follow the displayed instructions to continue the training at the point of interruption.

Hope you enjoy using minizero.
Have a nice day :)