diff --git a/doc/PIPELINE_SERVING_CN.md b/doc/PIPELINE_SERVING_CN.md
index 22e4b33050ac64ae7c53454ab5110ee623b93ce7..46a21a103777bafd4ed200e75861e6c09251d037 100644
--- a/doc/PIPELINE_SERVING_CN.md
+++ b/doc/PIPELINE_SERVING_CN.md
@@ -130,7 +130,7 @@ def postprocess(self, input_dicts, fetch_dict):
 
 **preprocess** 的参数是前继 Channel 中的数据 `input_dicts`，该变量是一个以前继 OP 的 name 为 Key，对应 OP 的输出为 Value 的字典。
 
-**process ** 的参数是 Paddle Serving Client 预测接口的输入变量 `fetch_dict`（preprocess 函数的返回值），该变量是一个以 feed_name 为 Key，对应 ndarray 格式的数据为 Value 的字典。
+**process** 的参数是 Paddle Serving Client 预测接口的输入变量 `fetch_dict`（preprocess 函数的返回值），该变量是一个以 feed_name 为 Key，对应 ndarray 格式的数据为 Value 的字典。
 
 **postprocess** 的参数是 `input_dicts` 和 `fetch_dict`，`input_dicts` 与 preprocess 的参数一致，`fetch_dict` 是 process 函数的返回值（如果没有执行 process ，则该值为 preprocess 的返回值）。
 
diff --git a/tools/serving_build.sh b/tools/serving_build.sh
index ee70cfc0b9e61fda0436eb84ed13200287175680..c2eee61959856c885103ed5bdd031603178b02e5 100644
--- a/tools/serving_build.sh
+++ b/tools/serving_build.sh
@@ -137,6 +137,15 @@ function kill_server_process() {
     sleep 1
 }
 
+function kill_process_by_pid() {
+    if [ $# != 1 ]; then
+        echo "usage: kill_process_by_pid <PID>"
+        exit 1
+    fi
+    local PID=$1
+    lsof -i:$PID | awk 'NR == 1 {next} {print $2}' | xargs kill
+}
+
 function python_test_fit_a_line() {
     # pwd: /Serving/python/examples
     cd fit_a_line # pwd: /Serving/python/examples/fit_a_line
@@ -521,6 +530,7 @@ function python_test_grpc_impl() {
             check_cmd "python test_batch_client.py > /dev/null"
             check_cmd "python test_timeout_client.py > /dev/null"
             kill_server_process
+            kill_process_by_pid 9393
 
             check_cmd "python test_server.py uci_housing_model > /dev/null &"
             sleep 5 # wait for the server to start
@@ -531,6 +541,7 @@ function python_test_grpc_impl() {
             check_cmd "python test_batch_client.py > /dev/null"
             check_cmd "python test_timeout_client.py > /dev/null"
             kill_server_process
+            kill_process_by_pid 9393
 
             cd .. # pwd: /Serving/python/examples/grpc_impl_example
 
@@ -579,6 +590,7 @@ function python_test_grpc_impl() {
             check_cmd "python test_batch_client.py > /dev/null"
             check_cmd "python test_timeout_client.py > /dev/null"
             kill_server_process
+            kill_process_by_pid 9393
 
             check_cmd "python test_server_gpu.py uci_housing_model > /dev/null &"
             sleep 5 # wait for the server to start
@@ -589,7 +601,8 @@ function python_test_grpc_impl() {
             check_cmd "python test_batch_client.py > /dev/null"
             check_cmd "python test_timeout_client.py > /dev/null"
             kill_server_process
-            ps -ef | grep "test_server_gpu" | grep -v serving_build | grep -v grep | awk '{print $2}' | xargs kill
+            kill_process_by_pid 9393
+            #ps -ef | grep "test_server_gpu" | grep -v serving_build | grep -v grep | awk '{print $2}' | xargs kill
 
             cd .. # pwd: /Serving/python/examples/grpc_impl_example
 
@@ -702,7 +715,6 @@ function python_test_pipeline(){
             # start paddle serving service (brpc)
             sh get_data.sh
             python -m paddle_serving_server.serve --model imdb_cnn_model --port 9292 --workdir test9292 &> cnn.log &
-            sleep 5
             python -m paddle_serving_server.serve --model imdb_bow_model --port 9393 --workdir test9393 &> bow.log &
             sleep 5
             
@@ -771,10 +783,11 @@ EOF
             ps -ef | grep "pipeline_server" | grep -v grep | awk '{print $2}' | xargs kill
             
             kill_server_process
+            kill_process_by_pid 9292
+            kill_process_by_pid 9393
 
             # start paddle serving service (grpc)
             python -m paddle_serving_server.serve --model imdb_cnn_model --port 9292 --use_multilang --workdir test9292 &> cnn.log &
-            sleep 5
             python -m paddle_serving_server.serve --model imdb_bow_model --port 9393 --use_multilang --workdir test9393 &> bow.log &
             sleep 5
             python test_pipeline_server.py > /dev/null &
@@ -782,6 +795,8 @@ EOF
             check_cmd "python test_pipeline_client.py"
             ps -ef | grep "pipeline_server" | grep -v grep | awk '{print $2}' | xargs kill
             kill_server_process
+            kill_process_by_pid 9292
+            kill_process_by_pid 9393
             ;;
         GPU)
             echo "pipeline ignore GPU test"