tensorflow - hdfsBuilderConnect error while using tfserving load model from hdfs

Question

there is my environment info:

• TensorFlow Serving version 1.14
• os mac10.15.7

i want to load modle from hdfs by using tfserving.

when i build a tensorflow-serving:hadoop docker image,like this:

FROM tensorflow/serving:2.2.0

RUN apt update && apt install -y openjdk-8-jre
RUN mkdir /opt/hadoop-2.8.2
COPY /hadoop-2.8.2 /opt/hadoop-2.8.2


ENV JAVA_HOME /usr/lib/jvm/java-8-openjdk-amd64
ENV HADOOP_HDFS_HOME /opt/hadoop-2.8.2
ENV HADOOP_HOME /opt/hadoop-2.8.2
ENV LD_LIBRARY_PATH 
${LD_LIBRARY_PATH}:${JAVA_HOME}/jre/lib/amd64/server
# ENV PATH $PATH:$HADOOP_HOME/sbin:$HADOOP_HOME/bin
RUN echo '#!/bin/bash 


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.sh

EXPOSE 8500
EXPOSE 9000
ENTRYPOINT ["/usr/bin/tf_serving_entrypoint.sh"]

and then run :

docker run -p 9001:9000 --name tensorflow-serving-11 -e MODEL_NAME=tfrest -e MODEL_BASE_PATH=hdfs://ip:port/user/cess2_test/workspace/cess/models -t tensorflow_serving:1.14-hadoop-2.8.2

i met this problem. ps:i have already modify hadoop config in hadoop-2.8.2

hdfsBuilderConnect(forceNewInstance=0, nn=ip:port, port=0, kerbTicketCachePath=(NULL), userName=(NULL)) 
error:(unable to get stack trace for java.lang.NoClassDefFoundError exception: ExceptionUtils::getStackTrace error.)

is there any suggestions how to solve this problem? thanks

Answer 1

i solved this problem by adding hadoop absolute path to classpath.