How to read multiple s3 bucket?

What I need help with / What I was wondering
I want to read multiple s3 bucket within a single process.
Here is psudo code:

  ```
  import tensorflow_datasets as tfds
  import tensorflow_io as tfio
  os.environ["AWS_ACCESS_KEY_ID"] = "xxxxxxxxx"
  os.environ["AWS_SECRET_ACCESS_KEY"] = "xxxxxxxxxx"
  # os.environ["AWS_REGION"] = "us-east-1"
  os.environ["S3_ENDPOINT"] = "http://ip1:80"
  os.environ["S3_USE_HTTPS"] = "0"
  os.environ["S3_VERIFY_SSL"] = "0"
  
  builder = tfds.builder_from_directory(builder_dir='s3://open_x_embodiment_origin/cmu_stretch/0.1.0/')
  
  os.environ["AWS_ACCESS_KEY_ID"] = "yyyyy"
  os.environ["AWS_SECRET_ACCESS_KEY"] = "yyyyyyyyyy"
  # os.environ["AWS_REGION"] = "us-east-1"
  os.environ["S3_ENDPOINT"] = "http://ip2:80"
  
  builder = tfds.builder_from_directory(builder_dir='s3://droid/1.0.0/')
  ```

I can only get the first bucket. When I builder from the second bucket, I fails with the message: FileNotFoundError: Could not load dataset info from s3://droid/1.0.0/dataset_info.json.

It seems like tfds can not read two buckets with different access keys. Is it right? or I should do something to connect two different buckets.

Best,
Looking forward to your reply.

The report uses Python with tensorflow_datasets.builder_from_directory and tensorflow_io, targeting two s3 paths with different AWS credentials and endpoints. Start by reproducing the second call after the first and inspect the reported dataset_info.json FileNotFoundError; done means establishing whether two bucket configurations are supported in one process and documenting or testing the supported behavior.