Merge branch 'develop' into image-reader

python/examples/bert/benchmark.sh

-rm profile_log
+rm profile_log*
 export CUDA_VISIBLE_DEVICES=0,1,2,3
 export FLAGS_profile_server=1
 export FLAGS_profile_client=1
 export FLAGS_serving_latency=1
-python3 -m paddle_serving_server_gpu.serve --model $1 --port 9292 --thread 4 --gpu_ids 0,1,2,3 --mem_optim --ir_optim 2> elog > stdlog &
-hostname=`echo $(hostname)|awk -F '.baidu.com' '{print $1}'`
-sleep 5
+
 gpu_id=0
+#save cpu and gpu utilization log
+if [ -d utilization ];then
+    rm -rf utilization
+else
+    mkdir utilization
+fi
+#start server
+$PYTHONROOT/bin/python3 -m paddle_serving_server_gpu.serve --model $1 --port 9292 --thread 4 --gpu_ids 0,1,2,3 --mem_optim  --ir_optim >  elog  2>&1 &
+sleep 5
 
 #warm up
-python3 benchmark.py --thread 8 --batch_size 1 --model $2/serving_client_conf.prototxt --request rpc > profile 2>&1
-
-for thread_num in 4 8 16
+$PYTHONROOT/bin/python3 benchmark.py --thread 4 --batch_size 1 --model $2/serving_client_conf.prototxt --request rpc > profile 2>&1
+echo -e "import psutil\ncpu_utilization=psutil.cpu_percent(1,False)\nprint('CPU_UTILIZATION:', cpu_utilization)\n" > cpu_utilization.py
+for thread_num in 1 4 8 16
 do
-for batch_size in 1 4 16 64 256
+for batch_size in 1 4 16 64
 do
     job_bt=`date '+%Y%m%d%H%M%S'`
-    nvidia-smi --id=$gpu_id --query-compute-apps=used_memory --format=csv -lms 100 > gpu_use.log 2>&1 &
+    nvidia-smi --id=0 --query-compute-apps=used_memory --format=csv -lms 100 > gpu_use.log 2>&1 &
+    nvidia-smi --id=0 --query-gpu=utilization.gpu --format=csv -lms 100 > gpu_utilization.log 2>&1 &
     gpu_memory_pid=$!
-    python3 benchmark.py --thread $thread_num --batch_size $batch_size --model $2/serving_client_conf.prototxt --request rpc > profile 2>&1
+    $PYTHONROOT/bin/python3 benchmark.py --thread $thread_num --batch_size $batch_size --model $2/serving_client_conf.prototxt --request rpc > profile 2>&1
     kill ${gpu_memory_pid}
+    kill `ps -ef|grep used_memory|awk '{print $2}'`
     echo "model_name:" $1
     echo "thread_num:" $thread_num
     echo "batch_size:" $batch_size
     echo "=================Done===================="
     echo "model_name:$1" >> profile_log_$1
     echo "batch_size:$batch_size" >> profile_log_$1
+    $PYTHONROOT/bin/python3 cpu_utilization.py >> profile_log_$1
     job_et=`date '+%Y%m%d%H%M%S'`
-    awk 'BEGIN {max = 0} {if(NR>1){if ($1 > max) max=$1}} END {print "MAX_GPU_MEMORY_USE:", max}' gpu_use.log >> profile_log_$1
-    monquery -n ${hostname} -i GPU_AVERAGE_UTILIZATION -s $job_bt -e $job_et -d 10 > gpu_log_file_${job_bt}
-    monquery -n ${hostname} -i CPU_USER -s $job_bt -e $job_et  -d 10 > cpu_log_file_${job_bt}
-    cpu_num=$(cat /proc/cpuinfo | grep processor | wc -l)
-    gpu_num=$(nvidia-smi -L|wc -l)
-    python ../util/show_profile.py profile $thread_num >> profile_log_$1
+    awk 'BEGIN {max = 0} {if(NR>1){if ($1 > max) max=$1}} END {print "MAX_GPU_MEMORY:", max}' gpu_use.log >> profile_log_$1
+    awk 'BEGIN {max = 0} {if(NR>1){if ($1 > max) max=$1}} END {print "GPU_UTILIZATION:", max}' gpu_utilization.log >> profile_log_$1
+    rm -rf gpu_use.log gpu_utilization.log
+    $PYTHONROOT/bin/python3 ../util/show_profile.py profile $thread_num >> profile_log_$1
     tail -n 8 profile >> profile_log_$1
     echo "" >> profile_log_$1
 done
 done
 
+#Divided log
+awk 'BEGIN{RS="\n\n"}{i++}{print > "bert_log_"i}' profile_log_$1
+mkdir bert_log && mv bert_log_* bert_log
 ps -ef|grep 'serving'|grep -v grep|cut -c 9-15 | xargs kill -9

python/examples/imagenet/benchmark.sh

-rm profile_log
+rm profile_log*
 export CUDA_VISIBLE_DEVICES=0,1,2,3
 export FLAGS_profile_server=1
 export FLAGS_profile_client=1
 python -m paddle_serving_server_gpu.serve --model $1 --port 9292 --thread 4 --gpu_ids 0,1,2,3 --mem_optim --ir_optim  2> elog > stdlog &
 
 sleep 5
+gpu_id=0
+#save cpu and gpu utilization log
+if [ -d utilization ];then
+    rm -rf utilization
+else
+    mkdir utilization
+fi
 
 #warm up
-$PYTHONROOT/bin/python benchmark.py --thread 8 --batch_size 1 --model $2/serving_client_conf.prototxt --request rpc > profile 2>&1
+$PYTHONROOT/bin/python3 benchmark.py --thread 4 --batch_size 1 --model $2/serving_client_conf.prototxt --request rpc > profile 2>&1
+echo -e "import psutil\ncpu_utilization=psutil.cpu_percent(1,False)\nprint('CPU_UTILIZATION:', cpu_utilization)\n" > cpu_utilization.py
 
 for thread_num in 1 4 8 16
 do
 for batch_size in 1 4 16 64
 do
+    job_bt=`date '+%Y%m%d%H%M%S'`
+    nvidia-smi --id=0 --query-compute-apps=used_memory --format=csv -lms 100 > gpu_use.log 2>&1 &
+    nvidia-smi --id=0 --query-gpu=utilization.gpu --format=csv -lms 100 > gpu_utilization.log 2>&1 &
+    gpu_memory_pid=$!
     $PYTHONROOT/bin/python benchmark.py --thread $thread_num --batch_size $batch_size --model $2/serving_client_conf.prototxt --request rpc > profile 2>&1
+    kill ${gpu_memory_pid}
+    kill `ps -ef|grep used_memory|awk '{print $2}'`
     echo "model name :" $1
     echo "thread num :" $thread_num
     echo "batch size :" $batch_size
     echo "=================Done===================="
     echo "model name :$1" >> profile_log
     echo "batch size :$batch_size" >> profile_log
+    job_et=`date '+%Y%m%d%H%M%S'`
+    awk 'BEGIN {max = 0} {if(NR>1){if ($1 > max) max=$1}} END {print "MAX_GPU_MEMORY:", max}' gpu_use.log >> profile_log_$1
+    awk 'BEGIN {max = 0} {if(NR>1){if ($1 > max) max=$1}} END {print "GPU_UTILIZATION:", max}' gpu_utilization.log >> profile_log_$1
+    rm -rf gpu_use.log gpu_utilization.log
     $PYTHONROOT/bin/python ../util/show_profile.py profile $thread_num >> profile_log
     tail -n 8 profile >> profile_log
+    echo "" >> profile_log_$1
 done
 done
 
+#Divided log
+awk 'BEGIN{RS="\n\n"}{i++}{print > "ResNet_log_"i}' profile_log_$1
+mkdir $1_log && mv ResNet_log_* $1_log
 ps -ef|grep 'serving'|grep -v grep|cut -c 9-15 | xargs kill -9

python/examples/imdb/benchmark.sh

-rm profile_log
-export CUDA_VISIBLE_DEVICES=0,1,2,3
+rm profile_log*
 export FLAGS_profile_server=1
 export FLAGS_profile_client=1
 export FLAGS_serving_latency=1
-python -m paddle_serving_server_gpu.serve --model $1 --port 9292 --thread 4 --gpu_ids 0,1,2,3 --mem_optim --ir_optim  2> elog > stdlog &
+$PYTHONROOT/bin/python3 -m paddle_serving_server.serve --model $1 --port 9292 --thread 4 --mem_optim --ir_optim  2> elog > stdlog &
 hostname=`echo $(hostname)|awk -F '.baidu.com' '{print $1}'`
-
+#save cpu and gpu utilization log
+if [ -d utilization ];then
+    rm -rf utilization
+else
+    mkdir utilization
+fi
 sleep 5
 
+
+#warm up
+$PYTHONROOT/bin/python3 benchmark.py --thread 4 --batch_size 1 --model $2/serving_client_conf.prototxt --request rpc > profile 2>&1
+echo -e "import psutil\ncpu_utilization=psutil.cpu_percent(1,False)\nprint('CPU_UTILIZATION:', cpu_utilization)\n" > cpu_utilization.py
+
 for thread_num in 1 4 8 16
 do
 for batch_size in 1 4 16 64
 do
     job_bt=`date '+%Y%m%d%H%M%S'`
-    python benchmark.py --thread $thread_num --batch_size $batch_size --model $2/serving_client_conf.prototxt --request rpc > profile 2>&1
+    $PYTHONROOT/bin/python3 benchmark.py --thread $thread_num --batch_size $batch_size --model $2/serving_client_conf.prototxt --request rpc > profile 2>&1
     echo "model_name:" $1
     echo "thread_num:" $thread_num
     echo "batch_size:" $batch_size
@@ -21,15 +30,14 @@ do
     echo "model_name:$1" >> profile_log_$1
     echo "batch_size:$batch_size" >> profile_log_$1
     job_et=`date '+%Y%m%d%H%M%S'`
-    awk 'BEGIN {max = 0} {if(NR>1){if ($1 > max) max=$1}} END {print "MAX_GPU_MEMORY_USE:", max}' gpu_use.log >> profile_log_$1
-    monquery -n ${hostname} -i GPU_AVERAGE_UTILIZATION -s $job_bt -e $job_et -d 10 > gpu_log_file_${job_bt}
-    monquery -n ${hostname} -i CPU_USER -s $job_bt -e $job_et  -d 10 > cpu_log_file_${job_bt}
-    cpu_num=$(cat /proc/cpuinfo | grep processor | wc -l)
-    gpu_num=$(nvidia-smi -L|wc -l)
-    python ../util/show_profile.py profile $thread_num >> profile_log_$1
+    $PYTHONROOT/bin/python3 ../util/show_profile.py profile $thread_num >> profile_log_$1
+    $PYTHONROOT/bin/python3 cpu_utilization.py >> profile_log_$1
     tail -n 8 profile >> profile_log_$1
     echo "" >> profile_log_$1
 done
 done
 
+#Divided log
+awk 'BEGIN{RS="\n\n"}{i++}{print > "imdb_log_"i}' profile_log_$1
+mkdir $1_log && mv imdb_log_* $1_log
 ps -ef|grep 'serving'|grep -v grep|cut -c 9-15 | xargs kill -9

python/paddle_serving_app/reader/image_reader.py

@@ -692,7 +692,7 @@ class Resize(object):
 
     Args:
         size (sequence or int): Desired output size. If size is a sequence like
-            (h, w), output size will be matched to this. If size is an int,
+            (w, h), output size will be matched to this. If size is an int,
             smaller edge of the image will be matched to this number.
             i.e, if height > width, then image will be rescaled to
             (size * height / width, size)