How to prevent GCP compute instances from resetting by itself?

Question

I\'m training a RL agent (more precisely, a stable baselines agent written using Tensorflow) on a GCP VM instance. The agent runs fine if I train for a small number of steps (e.