The code was pretended to have been executed.

We set up 20~60 sandboxes and then use shell api to execute code to all sandbox simultaneously.

We found that for the same code, there is approximately a 1% chance that the code will not be executed properly: Looking at the logs, it is indeed that the python code.py was called, but it immediately exited with exit_code = 0. However, the code was not actually executed (the first line of the code was not executed at all).

The resources of the server where the sandbox is located are sufficient.