Failed to read ModelServerConfig file: s3 (Minio) Tensorflow Serving (2.8->2.9)

[whoisltd]

I can't use s3 (minio) with TensorFlow serving version (2.8->2.9) on Docker. It gets an error:

Failed to start the server. Error: UNIMPLEMENTED: File system scheme 's3' not implemented (file: 's3://test/tf_model/models.config')

But when I downgrade image TensorFlow serving to the 2.6.0 version it will run.

docker run -p 8501:8501    --name tf_model\
     -e AWS_ACESS_KEY_ID=tQmkIzkPxWxiOMjb\
     -e AWS_SECRET_ACCESS_KEY=JvZ7IJm3QkXygirxz3ESXsf70FdcHY14\
     -e MODEL_BASE_PATH=s3://test/tf_model\
     -e S3_ENDPOINT=http://192.168.1.184/:9000\
     -e AWS_REGION=us-east-1\
     -t tensorflow/serving\
     --model_config_file=s3://test/tf_model/models.config\
     --allow_version_labels_for_unavailable_models=true\
     --model_config_file_poll_wait_seconds=10

How to use Tensorflow serving 2.8->2.9 version with minio ?

[singhniraj08]

@whoisltd,

Looks like this issue is similar to #1963.

As a workaround, as suggested here, please try installing and importing tensorflow-io. Kindly let us know if this resolves your issue. Thank you!

[whoisltd]

@singhniraj08 where i will install and import this?

[singhniraj08]

@whoisltd, tensorflow_io should be imported separately to your code to have s3 access.

[fsonntag]

@singhniraj08 As mentioned in the discussion you mentioned, this can only be done in a Python environment, while the Docker image is using the pre-built C++ TF Serving.
So you can't just "install and import stuff", you'd need to build it yourself with tensorflow_io being linked.

[singhniraj08]

@fsonntag,

Team is looking into this issue. In meanwhile, please use TF serving 2.6.2 docker image for accessing S3 file.

[1357810la]

这个是什么情况啊。大家有知道的吗

[whoisltd]

这个是什么情况啊。大家有知道的吗

@1357810la check this question.