Docker image error: " File system scheme 's3' not implemented. "

[ekaynar]

I built my own docker image from the source code, and it works fine with local filesystem. However, when I try use the s3 interface I am getting the following error:

Steps to reproduce the problem:
docker build -t dlio .

docker run -e S3_ENDPOINT=http://xxxx:9000 -e AWS_ACCESS_KEY_ID=xxxx-e AWS_SECRET_ACCESS_KEY=xxx -e S3_VERIFY_SSL=0 -e S3_USE_HTTPS=0 -e AWS_REGION=us-east-1 -t dlio dlio_benchmark workload=resnet50 ++workload.workflow.generate_data=True ++workload.workflow.train=False ++workload.storage.storage_type=s3 ++workload.storage.storage_root=mybucket

Error:
File "/usr/local/lib/python3.10/site-packages/dlio_benchmark/framework/tf_framework.py", line 130, in create_node tf.io.gfile.mkdir(id) File "/usr/local/lib/python3.10/site-packages/tensorflow/python/lib/io/file_io.py", line 483, in create_dir_v2 _pywrap_file_io.CreateDir(compat.path_to_bytes(path)) tensorflow.python.framework.errors_impl.UnimplementedError: File system scheme 's3' not implemented (file: 's3://mybucket/resnet50')

When I build and install the source code on bare-metal node, S3 interface works fine. I can read/write data to S3 without any issue. The requirement.txt includes tensorflow-io and I could not figure it out why the docker image does not work.

[hariharan-devarajan]

Can u check pip freeze on your source install and from docker. I think there is a version mismatch of tensorflow packages.