diff --git a/doc/COMPILE.md b/doc/COMPILE.md
index aaf205eb3eeddc5d099384262fd26f31b291b370..51f2cc28ad959603b2f72d4bf32898023b940c08 100644
--- a/doc/COMPILE.md
+++ b/doc/COMPILE.md
@@ -12,7 +12,7 @@
 
 ``` python
 git clone https://github.com/PaddlePaddle/Serving
-git submodule update --init --recursive
+cd Serving && git submodule update --init --recursive
 ```
 
 ### 编译Server部分
@@ -44,3 +44,6 @@ make -j10
 
 ### 安装wheel包
 无论是client端还是server端，编译完成后，安装python/dist/下的whl包即可
+
+### 注意事项
+运行python端server时，会检查`SERVING_BIN`环境变量，如果想使用自己编译的二进制文件，请将设置该环境变量为对应二进制文件的路径，通常是`export SERVING_BIN=${BUILD_PATH}/core/general-server/serving`。
diff --git a/doc/DESIGN_DOC.md b/doc/DESIGN_DOC.md
index 7c460c129e74423eb94b4beb010a277778f62900..f6da3770acfce58cfa8a19fa683d246a4390350e 100644
--- a/doc/DESIGN_DOC.md
+++ b/doc/DESIGN_DOC.md
@@ -50,7 +50,6 @@ serving_io.save_model("serving_model", "client_conf",
 
 服务端的预测逻辑可以通过Paddle Serving Server端的API进行人工定义，一个例子：
 ``` python
-``` python
 import paddle_serving_server as serving
 op_maker = serving.OpMaker()
 read_op = op_maker.create('general_reader')
@@ -221,6 +220,3 @@ imdb_service.run_server()
 
 ### 5.3 向量检索、树结构检索
 在推荐与广告场景的召回系统中，通常需要采用基于向量的快速检索或者基于树结构的快速检索，Paddle Serving会对这方面的检索引擎进行集成或扩展。
-
-
-
diff --git a/python/paddle_serving_server_gpu/serve.py b/python/paddle_serving_server_gpu/serve.py
index 5d9d96d517d64b21313fda0b44a83b34142b014b..39d7a914b180d60442b0aa766a85115d7cc0c38c 100644
--- a/python/paddle_serving_server_gpu/serve.py
+++ b/python/paddle_serving_server_gpu/serve.py
@@ -91,15 +91,15 @@ if __name__ == "__main__":
     if args.name == "None":
         start_multi_card(args)
     else:
+        from .web_service import WebService
         web_service = WebService(name=args.name)
         web_service.load_model_config(args.model)
-        gpu_ids = []
-        if args.gpu_ids == "":
+        gpu_ids = args.gpu_ids
+        if gpu_ids == "":
             if "CUDA_VISIBLE_DEVICES" in os.environ:
                 gpu_ids = os.environ["CUDA_VISIBLE_DEVICES"]
         if len(gpu_ids) > 0:
-            gpus = [int(x) for x in gpu_ids.split(",")]
-            web_service.set_gpus(gpus)
+            web_service.set_gpus(gpu_ids)
         web_service.prepare_server(
             workdir=args.workdir, port=args.port, device=args.device)
         web_service.run_server()
diff --git a/tools/Dockerfile.gpu b/tools/Dockerfile.gpu
index 427ae83bcb805ec70c1e6d575e84234f17e9fb30..091f4a546b549a3dd53645e78ab49b1cd46bf5b3 100644
--- a/tools/Dockerfile.gpu
+++ b/tools/Dockerfile.gpu
@@ -10,6 +10,6 @@ RUN yum -y install wget && \
     curl https://bootstrap.pypa.io/get-pip.py -o get-pip.py && \
     python get-pip.py && rm get-pip.py && \
     ln -s /usr/local/cuda-9.0/lib64/libcublas.so.9.0 /usr/local/cuda-9.0/lib64/libcublas.so && \
-    echo 'export LD_LIBRARY_PATH=/usr/local/cuda/lib64':$LD_LIBRARY_PATH >> /root/.bashrc && \
+    echo 'export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH' >> /root/.bashrc && \
     ln -s /usr/local/cuda-9.0/targets/x86_64-linux/lib/libcudnn.so.7 /usr/local/cuda-9.0/targets/x86_64-linux/lib/libcudnn.so && \
     echo 'export LD_LIBRARY_PATH=/usr/local/cuda-9.0/targets/x86_64-linux/lib:$LD_LIBRARY_PATH' >> /root/.bashrc
diff --git a/tools/Dockerfile.gpu.devel b/tools/Dockerfile.gpu.devel
new file mode 100644
index 0000000000000000000000000000000000000000..a2233908dbcff4f2f2bbd3edad24b83cb5252e16
--- /dev/null
+++ b/tools/Dockerfile.gpu.devel
@@ -0,0 +1,23 @@
+FROM nvidia/cuda:9.0-cudnn7-devel-centos7
+
+RUN yum -y install wget >/dev/null \
+    && yum -y install gcc gcc-c++ make glibc-static which >/dev/null \
+    && yum -y install git openssl-devel curl-devel bzip2-devel python-devel >/dev/null \
+    && wget https://cmake.org/files/v3.2/cmake-3.2.0-Linux-x86_64.tar.gz >/dev/null \
+    && tar xzf cmake-3.2.0-Linux-x86_64.tar.gz \
+    && mv cmake-3.2.0-Linux-x86_64 /usr/local/cmake3.2.0 \
+    && echo 'export PATH=/usr/local/cmake3.2.0/bin:$PATH' >> /root/.bashrc \
+    && rm cmake-3.2.0-Linux-x86_64.tar.gz \
+    && wget https://dl.google.com/go/go1.14.linux-amd64.tar.gz >/dev/null \
+    && tar xzf go1.14.linux-amd64.tar.gz \
+    && mv go /usr/local/go \
+    && echo 'export GOROOT=/usr/local/go' >> /root/.bashrc \
+    && echo 'export PATH=/usr/local/go/bin:$PATH' >> /root/.bashrc \
+    && rm go1.14.linux-amd64.tar.gz \
+    && yum -y install python-devel sqlite-devel >/dev/null \
+    && curl https://bootstrap.pypa.io/get-pip.py -o get-pip.py >/dev/null \
+    && python get-pip.py >/dev/null \
+    && pip install google protobuf setuptools wheel flask >/dev/null \
+    && rm get-pip.py \
+    && yum -y install epel-release && yum -y install patchelf \
+    && yum clean all
diff --git a/tools/serving_build.sh b/tools/serving_build.sh
index dd6a3f6da8b3e40f2e379cb9457c4e5f00bc900c..8bb305c750b7b6a60eaeb44bcbfa87746f7f25dc 100644
--- a/tools/serving_build.sh
+++ b/tools/serving_build.sh
@@ -1,12 +1,23 @@
 #!/usr/bin/env bash
 
+function unsetproxy() {
+    HTTP_PROXY_TEMP=$http_proxy
+    HTTPS_PROXY_TEMP=$https_proxy
+    unset http_proxy
+    unset https_proxy
+}
+
+function setproxy() {
+    export http_proxy=$HTTP_PROXY_TEMP
+    export https_proxy=$HTTPS_PROXY_TEMP
+}
+
 function init() {
     source /root/.bashrc
     set -v
-    #export http_proxy=http://172.19.56.199:3128
-    #export https_proxy=http://172.19.56.199:3128
     export PYTHONROOT=/usr
     cd Serving
+    export SERVING_WORKDIR=$PWD
 }
 
 function check_cmd() {
@@ -16,18 +27,40 @@ function check_cmd() {
     fi
 }
 
+function rerun() {
+    if [ $# -ne 2 ]; then
+        echo "usage: rerun command rerun-times"
+        exit 1
+    fi
+    local command=$1
+    local times=$2
+    for((i=1;i<=${times};i++))
+    do
+        if [ ${i} != 1 ]; then
+            echo "${i}-th run command: ${command}..."
+        fi
+        eval $command
+        if [ $? -eq 0 ]; then
+            return 0
+        fi
+        echo "${i}-th run(command: ${command}) failed."
+    done
+    exit 1
+}
+
 function build_client() {
     local TYPE=$1
     local DIRNAME=build-client-$TYPE
-    mkdir $DIRNAME && cd $DIRNAME
+    mkdir $DIRNAME # pwd: /Serving
+    cd $DIRNAME # pwd: /Serving/build-client-$TYPE
     case $TYPE in
         CPU|GPU)
             cmake -DPYTHON_INCLUDE_DIR=$PYTHONROOT/include/python2.7/ \
                   -DPYTHON_LIBRARIES=$PYTHONROOT/lib64/libpython2.7.so \
                   -DPYTHON_EXECUTABLE=$PYTHONROOT/bin/python \
                   -DCLIENT_ONLY=ON ..
-            check_cmd "make -j2 >/dev/null"
-            pip install python/dist/paddle_serving_client* >/dev/null
+            rerun "make -j2 >/dev/null" 3 # due to some network reasons, compilation may fail
+            pip install -U python/dist/paddle_serving_client* >/dev/null
             ;;
         *)
             echo "error type"
@@ -35,22 +68,24 @@ function build_client() {
             ;;
     esac
     echo "build client $TYPE part finished as expected."
-    cd ..
-    rm -rf $DIRNAME
+    cd .. # pwd: /Serving
+    # rm -rf $DIRNAME
 }
 
 function build_server() {
     local TYPE=$1
     local DIRNAME=build-server-$TYPE
-    mkdir $DIRNAME && cd $DIRNAME
+    mkdir $DIRNAME # pwd: /Serving
+    cd $DIRNAME # pwd: /Serving/build-server-$TYPE
     case $TYPE in
         CPU)
             cmake -DPYTHON_INCLUDE_DIR=$PYTHONROOT/include/python2.7/ \
                   -DPYTHON_LIBRARIES=$PYTHONROOT/lib64/libpython2.7.so \
                   -DPYTHON_EXECUTABLE=$PYTHONROOT/bin/python \
                   -DCLIENT_ONLY=OFF ..
-            check_cmd "make -j2 >/dev/null && make install -j2 >/dev/null"
-            pip install python/dist/paddle_serving_server* >/dev/null
+            rerun "make -j2 >/dev/null" 3 # due to some network reasons, compilation may fail
+            check_cmd "make install -j2 >/dev/null"
+            pip install -U python/dist/paddle_serving_server* >/dev/null
             ;;
         GPU)
             cmake -DPYTHON_INCLUDE_DIR=$PYTHONROOT/include/python2.7/ \
@@ -58,8 +93,9 @@ function build_server() {
                   -DPYTHON_EXECUTABLE=$PYTHONROOT/bin/python \
                   -DCLIENT_ONLY=OFF \
                   -DWITH_GPU=ON ..
-            check_cmd "make -j2 >/dev/null && make install -j2 >/dev/null"
-            pip install python/dist/paddle_serving_server* >/dev/null
+            rerun "make -j2 >/dev/null" 3 # due to some network reasons, compilation may fail
+            check_cmd "make install -j2 >/dev/null"
+            pip install -U python/dist/paddle_serving_server* >/dev/null
             ;;
         *)
             echo "error type"
@@ -67,30 +103,62 @@ function build_server() {
             ;;
     esac
     echo "build server $TYPE part finished as expected."
-    cd ..
+    cd .. # pwd: /Serving
+    # rm -rf $DIRNAME    for export SERVING_BIN
+}
+
+function kill_server_process() {
+    ps -ef | grep "serving" | grep -v serving_build | grep -v grep | awk '{print $2}' | xargs kill
 }
 
 function python_test_fit_a_line() {
-    cd fit_a_line
+    # pwd: /Serving/python/examples
+    cd fit_a_line # pwd: /Serving/python/examples/fit_a_line
     sh get_data.sh
     local TYPE=$1
-    echo $TYPE
+    export SERVING_BIN=${SERVING_WORKDIR}/build-server-${TYPE}/core/general-server/serving
     case $TYPE in
         CPU)
             # test rpc
-            check_cmd "python test_server.py uci_housing_model/ > /dev/null &"
-            sleep 5
+            check_cmd "python -m paddle_serving_server.serve --model uci_housing_model --port 9393 --thread 4 > /dev/null &"
+            sleep 5 # wait for the server to start
             check_cmd "python test_client.py uci_housing_client/serving_client_conf.prototxt > /dev/null"
-            ps -ef | grep "paddle_serving_server" | grep -v grep | awk '{print $2}' | xargs kill
+            kill_server_process
+            
             # test web
-            check_cmd "python -m paddle_serving_server.serve --model uci_housing_model/ --name uci --port 9399 --name uci > /dev/null &"
-            sleep 5
-            check_cmd "curl -H \"Content-Type:application/json\" -X POST -d '{\"x\": [0.0137, -0.1136, 0.2553, -0.0692, 0.0582, -0.0727, -0.1583, -0.0584, 0.6283, 0.4919, 0.1856, 0.0795, -0.0332], \"fetch\":[\"price\"]}' http://127.0.0.1:9399/uci/prediction"
-            ps -ef | grep "paddle_serving_server" | grep -v grep | awk '{print $2}' | xargs kill
+            unsetproxy # maybe the proxy is used on iPipe, which makes web-test failed.
+            check_cmd "python -m paddle_serving_server.serve --model uci_housing_model --name uci --port 9393 --thread 4 --name uci > /dev/null &"
+            sleep 5 # wait for the server to start
+            check_cmd "curl -H \"Content-Type:application/json\" -X POST -d '{\"x\": [0.0137, -0.1136, 0.2553, -0.0692, 0.0582, -0.0727, -0.1583, -0.0584, 0.6283, 0.4919, 0.1856, 0.0795, -0.0332], \"fetch\":[\"price\"]}' http://127.0.0.1:9393/uci/prediction"
+            # check http code
+            http_code=`curl -H "Content-Type:application/json" -X POST -d '{"x": [0.0137, -0.1136, 0.2553, -0.0692, 0.0582, -0.0727, -0.1583, -0.0584, 0.6283, 0.4919, 0.1856, 0.0795, -0.0332], "fetch":["price"]}' -s -w "%{http_code}" -o /dev/null http://127.0.0.1:9393/uci/prediction`
+            setproxy # recover proxy state
+            kill_server_process
+            if [ ${http_code} -ne 200 ]; then
+                echo "HTTP status code -ne 200"
+                exit 1
+            fi
             ;;
         GPU)
-            echo "not support yet"
-            exit 1
+            # test rpc
+            check_cmd "python -m paddle_serving_server_gpu.serve --model uci_housing_model --port 9393 --thread 4 --gpu_ids 0 > /dev/null &"
+            sleep 5 # wait for the server to start
+            check_cmd "python test_client.py uci_housing_client/serving_client_conf.prototxt > /dev/null"
+            kill_server_process
+            
+            # test web
+            unsetproxy # maybe the proxy is used on iPipe, which makes web-test failed.
+            check_cmd "python -m paddle_serving_server_gpu.serve --model uci_housing_model --port 9393 --thread 2 --gpu_ids 0 --name uci > /dev/null &"
+            sleep 5 # wait for the server to start
+            check_cmd "curl -H \"Content-Type:application/json\" -X POST -d '{\"x\": [0.0137, -0.1136, 0.2553, -0.0692, 0.0582, -0.0727, -0.1583, -0.0584, 0.6283, 0.4919, 0.1856, 0.0795, -0.0332], \"fetch\":[\"price\"]}' http://127.0.0.1:9393/uci/prediction"
+            # check http code
+            http_code=`curl -H "Content-Type:application/json" -X POST -d '{"x": [0.0137, -0.1136, 0.2553, -0.0692, 0.0582, -0.0727, -0.1583, -0.0584, 0.6283, 0.4919, 0.1856, 0.0795, -0.0332], "fetch":["price"]}' -s -w "%{http_code}" -o /dev/null http://127.0.0.1:9393/uci/prediction`
+            setproxy # recover proxy state
+            kill_server_process
+            if [ ${http_code} -ne 200 ]; then
+                echo "HTTP status code -ne 200"
+                exit 1
+            fi
             ;;
         *)
             echo "error type"
@@ -99,57 +167,69 @@ function python_test_fit_a_line() {
     esac
     echo "test fit_a_line $TYPE part finished as expected."
     rm -rf image kvdb log uci_housing* work*
-    cd ..
+    unset SERVING_BIN
+    cd .. # pwd: /Serving/python/examples
 }
 
 function python_run_criteo_ctr_with_cube() {
+    # pwd: /Serving/python/examples
     local TYPE=$1
     yum install -y bc >/dev/null
-    cd criteo_ctr_with_cube
-    check_cmd "wget https://paddle-serving.bj.bcebos.com/unittest/ctr_cube_unittest.tar.gz"
-    check_cmd "tar xf ctr_cube_unittest.tar.gz"
-    check_cmd "mv models/ctr_client_conf ./"
-    check_cmd "mv models/ctr_serving_model_kv ./"
-    check_cmd "mv models/data ./cube/"
-    check_cmd "mv models/ut_data ./"
-    cp ../../../build-server-$TYPE/output/bin/cube* ./cube/ 
-    mkdir -p $PYTHONROOT/lib/python2.7/site-packages/paddle_serving_server/serving-cpu-avx-openblas-0.1.3/
-    yes | cp ../../../build-server-$TYPE/output/demo/serving/bin/serving $PYTHONROOT/lib/python2.7/site-packages/paddle_serving_server/serving-cpu-avx-openblas-0.1.3/
+    cd criteo_ctr_with_cube # pwd: /Serving/python/examples/criteo_ctr_with_cube
+    case $TYPE in
+        CPU)
+            check_cmd "wget https://paddle-serving.bj.bcebos.com/unittest/ctr_cube_unittest.tar.gz"
+            check_cmd "tar xf ctr_cube_unittest.tar.gz"
+            check_cmd "mv models/ctr_client_conf ./"
+            check_cmd "mv models/ctr_serving_model_kv ./"
+            check_cmd "mv models/data ./cube/"
+            check_cmd "mv models/ut_data ./"
+            cp ../../../build-server-$TYPE/output/bin/cube* ./cube/ 
+            mkdir -p $PYTHONROOT/lib/python2.7/site-packages/paddle_serving_server/serving-cpu-avx-openblas-0.1.3/
+            yes | cp ../../../build-server-$TYPE/output/demo/serving/bin/serving $PYTHONROOT/lib/python2.7/site-packages/paddle_serving_server/serving-cpu-avx-openblas-0.1.3/
 
-    sh cube_prepare.sh &
-    check_cmd "mkdir work_dir1 && cp cube/conf/cube.conf ./work_dir1/"    
-    python test_server.py ctr_serving_model_kv &
-    check_cmd "python test_client.py ctr_client_conf/serving_client_conf.prototxt ./ut_data >score"
-    AUC=$(tail -n 2  score | awk 'NR==1')
-    VAR2="0.70"
-    RES=$( echo "$AUC>$VAR2" | bc )
-    if [[ $RES -eq 0 ]]; then
-        echo "error with criteo_ctr_with_cube inference auc test, auc should > 0.70"
-        exit 1
-    fi
-    echo "criteo_ctr_with_cube inference auc test success"
-    ps -ef | grep "paddle_serving_server" | grep -v grep | awk '{print $2}' | xargs kill
-    ps -ef | grep "cube" | grep -v grep | awk '{print $2}' | xargs kill
+            sh cube_prepare.sh &
+            check_cmd "mkdir work_dir1 && cp cube/conf/cube.conf ./work_dir1/"    
+            python test_server.py ctr_serving_model_kv &
+            check_cmd "python test_client.py ctr_client_conf/serving_client_conf.prototxt ./ut_data >score"
+            AUC=$(tail -n 2  score | awk 'NR==1')
+            VAR2="0.70"
+            RES=$( echo "$AUC>$VAR2" | bc )
+            if [[ $RES -eq 0 ]]; then
+                echo "error with criteo_ctr_with_cube inference auc test, auc should > 0.70"
+                exit 1
+            fi
+            echo "criteo_ctr_with_cube inference auc test success"
+            ps -ef | grep "paddle_serving_server" | grep -v grep | awk '{print $2}' | xargs kill
+            ps -ef | grep "cube" | grep -v grep | awk '{print $2}' | xargs kill
+            ;;
+        GPU)
+            ;;
+        *)
+            echo "error type"
+            exit 1
+            ;;
+    esac
+    echo "test criteo_ctr_with_cube $TYPE part finished as expected."
+    cd .. # pwd: /Serving/python/examples
 }
 
 function python_run_test() {
-    cd python/examples
-    local TYPE=$1
-    # Frist time run, downloading PaddleServing components ...
-    python -c "from paddle_serving_server import Server; server = Server(); server.download_bin()"
-    python_test_fit_a_line $TYPE
-    python_run_criteo_ctr_with_cube $TYPE
+    # Using the compiled binary
+    local TYPE=$1 # pwd: /Serving
+    cd python/examples # pwd: /Serving/python/examples
+    python_test_fit_a_line $TYPE # pwd: /Serving/python/examples
+    python_run_criteo_ctr_with_cube $TYPE # pwd: /Serving/python/examples
     echo "test python $TYPE part finished as expected."
-    cd ../..
+    cd ../.. # pwd: /Serving
 }
 
 function main() {
-    local TYPE=$1
-    init
-    build_client $TYPE
-    build_server $TYPE
-    cd Serving/
-    python_run_test $TYPE
+    local TYPE=$1 # pwd: /
+    init # pwd: /Serving
+    build_client $TYPE # pwd: /Serving
+    build_server $TYPE # pwd: /Serving
+    python_run_test $TYPE # pwd: /Serving
     echo "serving $TYPE part finished as expected."
 }