AutoML - Forecasting -- Some child jobs are taking much longer. Why?

Question

I am setting up a AutoML Forecasting job, and I really like the automatic approach.

I have set a 3-fold cross validation, only using the RandomForest and some featurization. I understand that setting up the environment takes time. After an initial delay, it trains a set of models really quick (<1 minute per training), but then it gets stuck.

I am opening the child jobs, but have randomly assigned names and I find no information on why that particular run was so slow. I am looking through logs, trying to find the bottleneck to adjust my experimental setup. For instance, in the failed job below, the "code + logs" tap just include on file, automl_driver.py, which doesn't provide me with alot of information.

I have followed the FAQ about AutoML Forecasting, but it's not helping when I can't debug my AutoML experiment.

QUESTION: How do I know what each steps does in order to mitigate bottlenecks?