What is your question?

Ray version and other system information (Python version, TensorFlow version, OS):

Python version: Python 3.6.10 :: Anaconda, Inc.
Anaconda version: 4.7.12
TensorFlow version: 1.15.0
OS: Ubuntu 16.04
ray version: 0.8.1

Hello
I've trained an RL model using rllib
I have tested the breakout environment and the agent is running successfully.

Here is the python code I used. I used Rllib's python API
If you test at openai gym after learning, it works successfully.

I want to record the results of the agent.
Agent test was executed using the rollout command of rllib.
My result photo is below.
This is the command I executed

rllib rollout checkpoint_4301/checkpoint-4301 \
    --run PPO --env BreakoutNoFrameskip-v4 --monitor --config '{"monitor": true}'

When testing with rollout, on average 8 games run.
Many episodes ran, but only the results of one game (four lives in the case of breakout) were saved as videos.
And I think it's a mix of different game parts.

I additionally tested ms pacman.
Pacman's recording showed the same problem.

I have attached my rllib model and the recorded results file.
I would be grateful if you could tell me a document or a way to help me record the video.

PPO_BreakoutNoFrameskip-v4_2020-02-11_21-32-52m48o1b48.zip
BreakoutModel.zip