Can RLLib writing Environment Data to gcs?

How severe does this issue affect your experience of using Ray?

  • Medium: It contributes to significant difficulty to complete my task, but I can work around it.

Hi All,

I am trying to write environment data to gcs, but not sure RLLIb currently supports this. I want to make sure I am not missing something.

I set-up a minimum example using ray[tune,rllib]>=1.13.0,<1.14.0

from ray.rllib.agents.bandit import BanditLinTSTrainer
path = "gs://ml-adhoc/test_path"
bandit_config = {
    "env": SimpleEnvironment, # A custom environment
    "output": "dataset",
    "output_config": {
        "format":  "parquet",
        "path": path,
         "store_info" : True
     "horizon": 1,
    "env_config": {
        "seed": 0,       
    "seed": 0,
    'timesteps_per_iteration': 10,
    'num_workers': 2,
bandit_trainer = BanditLinTSTrainer(config=bandit_config)

I noticed that the data is actually written to the directory /app/project/gs://ml-adhoc/test_path, meaning that it is written locally. Is there a way to write the data to a gcs bucket?

If not, is there a recommended workaround?


Tagging @kourosh because he’s worked on datasets.

Hey @Diego_Ardila This is not supported right now. but I created a PR which should address your concern. [RLlib] Enable remote ds output writer by kouroshHakha · Pull Request #34793 · ray-project/ray · GitHub In the meantime is it possible to write into local and then upload the local files to the remote directory with python code in the end of your script?