"ProtocolError" and "InvalidMapData"

[gigg1]

Hello, I encountered an issue at runtime when I ran the MARL algorithm.
When I run the algorithm for a while, it sometimes shows that the temp map file is not valid, and the program ends.

Because I can successfully start and restart the SC2 game while the program is running, I think the maps I installed are correct. For example, when it is successfully restarted:

----------- ID: 113, Training for 284642/2050000
[INFO 16:50:39] absl Shutdown gracefully.
[INFO 16:50:39] absl Shutdown with return code: 1
[INFO 16:50:39] absl Launching SC2: C:/Program Files (x86)/StarCraft II\Versions/Base94137\SC2_x64.exe -listen 127.0.0.1 -port 58017 -dataDir C:/Program Files (x86)/StarCraft II\ -tempDir C:\Users\XXX\AppData\Local\Temp\sc-zeu63d73\ -displayMode 0 -windowwidth 1920 -windowheight 1200 -windowx 50 -windowy 50
[INFO 16:50:39] absl Connecting to: ws://127.0.0.1:58017/sc2api, attempt: 0, running: True
[INFO 16:50:42] absl Connecting to: ws://127.0.0.1:58017/sc2api, attempt: 1, running: True
----------- ID: 113, Training for 284652/2050000

This error is usually random in the middle of the algorithm running. The following is the error information:
Error:

----------- ID: 113, Training for 339975/2050000
[INFO 17:28:26] absl Shutdown gracefully.
[INFO 17:28:26] absl Shutdown with return code: 1
[INFO 17:28:26] absl Launching SC2: C:/Program Files (x86)/StarCraft II\Versions/Base94137\SC2_x64.exe -listen 127.0.0.1 -port 61797 -dataDir C:/Program Files (x86)/StarCraft II\ -tempDir C:\Users\XXX~1\AppData\Local\Temp\sc-4_gk43t4\ -displayMode 0 -windowwidth 1920 -windowheight 1200 -windowx 50 -windowy 50
[INFO 17:28:26] absl Connecting to: ws://127.0.0.1:61797/sc2api, attempt: 0, running: True
[INFO 17:28:29] absl Connecting to: ws://127.0.0.1:61797/sc2api, attempt: 1, running: True
[DEBUG 17:28:30] pymarl Stopping Heartbeat
[ERROR 17:28:30] pymarl Failed after 4:03:09!
Traceback (most recent calls WITHOUT Sacred internals):
  File "C:\Users\XXX\anaconda3\envs\gacg_gpu_smac\lib\site-packages\smac\env\starcraft2\starcraft2.py", line 477, in step
    self._controller.actions(req_actions)
  File "C:\Users\XXX\anaconda3\envs\gacg_gpu_smac\lib\site-packages\pysc2\lib\remote_controller.py", line 81, in _skip_status
    return func(self, *args, **kwargs)
  File "C:\Users\XXX\anaconda3\envs\gacg_gpu_smac\lib\site-packages\pysc2\lib\remote_controller.py", line 94, in _valid_status
    func.__name__, self.status, ",".join(map(str, valid))))
**pysc2.lib.protocol.ProtocolError: `actions` called while in state: Status.ended, valid: (Status.in_game)**

During handling of the above exception, another exception occurred:

Traceback (most recent calls WITHOUT Sacred internals):
  File "src/main.py", line 41, in my_main
    run(_run, config, _log)
  File "D:\X\GACG-main\src\run.py", line 164, in run
    run_sequential(args=args, logger=logger)
  File "D:\X\GACG-main\src\run.py", line 404, in run_sequential
    episode_batch, return_list, _ = runner.run(test_mode=False)
  File "D:\X\GACG-main\src\runners\episode_runner.py", line 80, in run
    reward, terminated, env_info = self.env.step(actions[0])
  File "C:\Users\XXX\anaconda3\envs\gacg_gpu_smac\lib\site-packages\smac\env\starcraft2\starcraft2.py", line 483, in step
    self.full_restart()
  File "C:\Users\XXX\anaconda3\envs\gacg_gpu_smac\lib\site-packages\smac\env\starcraft2\starcraft2.py", line 449, in full_restart
    self._launch()
  File "C:\Users\XXX\anaconda3\envs\gacg_gpu_smac\lib\site-packages\smac\env\starcraft2\starcraft2.py", line 343, in _launch
    self._controller.create_game(create)
  File "C:\Users\XXX\anaconda3\envs\gacg_gpu_smac\lib\site-packages\pysc2\lib\remote_controller.py", line 95, in _valid_status
    return func(self, *args, **kwargs)
  File "C:\Users\XXX\anaconda3\envs\gacg_gpu_smac\lib\site-packages\pysc2\lib\remote_controller.py", line 70, in _check_error
    return check_error(func(*args, **kwargs), error_enum)
  File "C:\Users\XXX\anaconda3\envs\gacg_gpu_smac\lib\site-packages\pysc2\lib\remote_controller.py", line 61, in check_error
    raise RequestError("%s.%s: '%s'" % (enum_name, error_name, details), res)
**pysc2.lib.remote_controller.RequestError: SC2APIProtocol.ResponseCreateGame.Error.InvalidMapData: 'temporary map 'C:\Users\XXX\AppData\Local\Temp\StarCraft II\TempLaunchMap.SC2Map' has invalid data.'**

[INFO 17:28:30] absl Shutdown gracefully.
[INFO 17:28:30] absl Shutdown with return code: 1

[gigg1]

I would like to know how to address this problem. Many thanks for any help!

[Balxkodehodet]

the Map error happens to me too frequently and it suddenly stops training... The reason is because the training ended abruptly or something like that, which corrupts the TEMP SC2Maps
The fix is to delete all TEMP maps in the TEMP folder (which the path of should be specified in the error) (So that it can create new temp maps which is fresh and working)
I got chatgpt to write me code which deleted the folder of corrupt maps automatically but usually doing it manually is the most secure way of knowing you are ok with training again.

[Balxkodehodet]

NOTE: Just delete the SC2 TEMP files, not the folders

[gigg1]

Thanks a lot for your help! It is really helpful. It has confused me for a long time, specifically when the training of algorithms was almost done, but this error happened suddenly.
So, do you mean that, once the error happens, delete the file "TempLaunchMap.SC2Map" in the folder 'C:\Users\XXX\AppData\Local\Temp\StarCraft II' ?

[Balxkodehodet]

Hey, you are welcome..
Yes, as corrupted SC2 temp map files can cause other errors also, its better to just delete them all before training again.

Its ok because you only delete a temporary map file that gets created again from the original map when it starts training again.

So yes delete the file, its ok if you delete all of them.. I usually have to delete alot of them because I run multiple envs.

Everytime I do this the training resumes successfully and no errors.

[Balxkodehodet]

can I ask what map you are training on?

this worked for me at simple64 map and with stablebaselines3 PPO agent learning with pysc2

So I am just assuming it will work with you aswell, in most cases deleting temporary files wont do damage:)

[gigg1]

I trained the MAPPO-based algorithms on almost all the maps of SMAC. Such as 8m, corridor, MMM2, and etc (as shown in the following picture). This error usually occurs in several maps, such as 3s5z, MMM2. Some of the experiments were trained on the computing servers without output, and I could not track their error information when they stopped unexpectedly.