UnityToGymWrapper Completely Broken due to Online Registry Being Down

[Zach-Attach]

Describe the bug
Around 12pm EDT May 1, mlagents started breaking without any changes on my part. It appears that it is being caused by a request to "https://storage.googleapis.com/mlagents-test-environments/1.1.0/manifest.yaml" which is returning a 404.

To Reproduce
Steps to reproduce the behavior:

1. Run from mlagents_envs.envs.unity_gym_env import UnityToGymWrapper
2. Get back an Error

Console logs / stack traces

File "/home/zlaborde/run.py", line 15, in <module>
from mlagents_envs.envs.unity_gym_env import UnityToGymWrapper
File "/home/zlaborde/miniconda3/envs/test/lib/python3.10/site-packages/mlagents_envs/envs/__init__.py", line 5, in <module>
for key in default_registry:
File "/home/zlaborde/miniconda3/envs/test/lib/python3.10/site-packages/mlagents_envs/registry/unity_env_registry.py", line 118, in __iter__
self._load_all_manifests()
File "/home/zlaborde/miniconda3/envs/test/lib/python3.10/site-packages/mlagents_envs/registry/unity_env_registry.py", line 83, in _load_all_manifests
manifest = load_remote_manifest(path_to_yaml)
File "/home/zlaborde/miniconda3/envs/test/lib/python3.10/site-packages/mlagents_envs/registry/binary_utils.py", line 218, in load_remote_manifest
request = urllib.request.urlopen(url, timeout=30)
File "/home/zlaborde/miniconda3/envs/test/lib/python3.10/urllib/request.py", line 216, in urlopen
return opener.open(url, data, timeout)
File "/home/zlaborde/miniconda3/envs/test/lib/python3.10/urllib/request.py", line 525, in open
response = meth(req, response)
File "/home/zlaborde/miniconda3/envs/test/lib/python3.10/urllib/request.py", line 634, in http_response
response = self.parent.error(
File "/home/zlaborde/miniconda3/envs/test/lib/python3.10/urllib/request.py", line 563, in error
return self._call_chain(*args)
File "/home/zlaborde/miniconda3/envs/test/lib/python3.10/urllib/request.py", line 496, in _call_chain
result = func(*args)
File "/home/zlaborde/miniconda3/envs/test/lib/python3.10/urllib/request.py", line 643, in http_error_default
raise HTTPError(req.full_url, code, msg, hdrs, fp)
urllib.error.HTTPError: HTTP Error 404: Not Found

Environment (please complete the following information):

• Unity Version: Unity 2023.3.12
• OS + version: Ubuntu 22.04
• ML-Agents version: 1.0.0
• Torch version: 2.2.0

[miguelalonsojr]

On it.

[Zach-Attach]

Thank you. And I should clarify that I noticed it today at 12pm but it could have started as early as yesterday.

[Sungwoo-Lee]

I'm experiencing the same problem!

My Environment:
Unity Version: Unity 2022.3.16f1
OS + version: Ubuntu 22.04
ML-Agents version: 1.0.0
Torch version: 2.2.1

[Ffr0nt]

Exactly the same thing for me! However with

from mlagents_envs.envs.unity_parallel_env import UnityParallelEnv

mlagents 0.30.0
mlagents_envs 0.30.0
(!git clone --branch release_20 https://github.com/Unity-Technologies/ml-agents.git)
torch 2.3.0
Python 3.9.19
Ubuntu 22.04.3 LTS

more details here

[mchivuku]

Hi All, thank you so much prompt attention, we are really stuck on completing our work and any help to expedite this request is very much appreciated. Thank you.

[miguelalonsojr]

The the GCP project that hosts the registry is being decommissioned. This PR should fix the issue: #6107 I don't have an estimate as to when the registry will be back up, but this should unblock folks. You should install from develop and not any specific release.

[Zach-Attach]

@miguelalonsojr It appears that #6108 undid the change made in #6107, does this mean that the GCP URL directs to the new registry? If so, does that mean it is safe to use the latest version on PyPI (1.0.0) rather than using develop?

[miguelalonsojr]

Yes, the GCP registry should be back up. You can use either develop (I always recommend using develop since our release cadence is very variable since ML-Agents is in KTLO mode for the moment), or the latest version on pypi. There have been some updates since v1.0.0 so as I mentioned above, I recommend using develop.