How to switch between Model Storage engine HDFS and S3 by TensorFlow-Serving

Editor to share with you how TensorFlow-Serving switches model storage engines HDFS and S3. I hope you will get something after reading this article. Let's discuss it together.

1. TensorFlow-Serving uses S3 storage model

Note: S3 here not only refers to Amazon S3, but also supports minio and other object storage that implements S3 protocol. It is worth noting that Tencent Cloud cos supports 2.2.0 and previous versions during testing. There is a compatibility problem with 2.3.0 and there is no problem with minio.

# Please set the following environment variable AWS_ACCESS_KEY_ID=XXXXX # to access the ID credential AWS_SECRET_ACCESS_KEY=XXXXX # required by S3 to access the KEY credential AWS_REGION=us-east-1 # S3 bucket region information required by S3 (optional) whether the S3_ENDPOINT=s3.us-east-1.amazonaws.com # S 3 access address S3_USE_HTTPS=1 # uses the HTTPS protocol. If the shutdown setting is 0S3_VERIFY_SSL=1 # if the HTTPS protocol is used, control whether SSL should be enabled. If the shutdown setting is set to 02, TensorFlow-Serving uses the HDFS storage model.

Currently, if you need to use HDFS as the storage engine, you need to repackage the existing docker images.

1. Download the corresponding Hadoop package and decompress it

Https://archive.apache.org/dist/hadoop/core/hadoop-2.6.0/hadoop-2.6.0.tar.gz

2. Dockerfile is as follows:

FROM tensorflow/serving:2.3.0RUN apt update & & apt install-y openjdk-8-jreCOPY hadoop-2.6.0 / root/hadoopENV JAVA_HOME / usr/lib/jvm/java-8-openjdk-amd64/ENV HADOOP_HDFS_HOME / root/hadoopENV LD_LIBRARY_PATH ${LD_LIBRARY_PATH}: ${JAVA_HOME} / jre/lib/amd64/serverRUN echo'#! / bin/bash\ n\ n CLASSPATH=$ (${HADOOP_HDFS_HOME} / bin/hadoop classpath-- Glob) tensorflow_model_server-- port=8500-- rest_api_port=9000\-- model_name=$ {MODEL_NAME}-- model_base_path=$ {MODEL_BASE_PATH} / ${MODEL_NAME}\ "$@" > / usr/bin/tf_serving_entrypoint.sh\ & & chmod + x / usr/bin/tf_serving_entrypoint.shEXPOSE 8500EXPOSE 9000ENTRYPOINT ["/ usr/bin/tf_serving_entrypoint.sh"]

3 、 Docker Build

Docker build-t tensorflow_serving:2.3.0-hadoop-2.6.0.

4. Run TensorFlow-Serving

Docker run-p 9001MODEL_BASE_PATH=hdfs://172.16.36.234:8020/user/model/tensorflow/ 9000-- name tensorflow-serving-e MODEL_NAME=resnet-e MODEL_BASE_PATH=hdfs://172.16.36.234:8020/user/model/tensorflow/-t tensorflow_serving:2.3.0-hadoop-2.6.0 after reading this article, I believe you have some understanding of "how TensorFlow-Serving switches model storage engines HDFS and S3". If you want to know more about it, welcome to follow the industry information channel. Thank you for reading!