clearml-serving/clearml_serving/serving/entrypoint.sh

60 lines
2.4 KiB
Bash
Executable File

#!/bin/bash
# print configuration
echo CLEARML_SERVING_TASK_ID="$CLEARML_SERVING_TASK_ID"
echo CLEARML_SERVING_PORT="$CLEARML_SERVING_PORT"
echo CLEARML_USE_GUNICORN="$CLEARML_USE_GUNICORN"
echo CLEARML_EXTRA_PYTHON_PACKAGES="$CLEARML_EXTRA_PYTHON_PACKAGES"
echo CLEARML_SERVING_NUM_PROCESS="$CLEARML_SERVING_NUM_PROCESS"
echo CLEARML_SERVING_POLL_FREQ="$CLEARML_SERVING_POLL_FREQ"
echo CLEARML_DEFAULT_KAFKA_SERVE_URL="$CLEARML_DEFAULT_KAFKA_SERVE_URL"
SERVING_PORT="${CLEARML_SERVING_PORT:-8080}"
GUNICORN_NUM_PROCESS="${CLEARML_SERVING_NUM_PROCESS:-4}"
GUNICORN_SERVING_TIMEOUT="${GUNICORN_SERVING_TIMEOUT:-600}"
GUNICORN_MAX_REQUESTS="${GUNICORN_MAX_REQUESTS:-0}"
UVICORN_SERVE_LOOP="${UVICORN_SERVE_LOOP:-asyncio}"
UVICORN_LOG_LEVEL="${UVICORN_LOG_LEVEL:-warning}"
# set default internal serve endpoint (for request pipelining)
CLEARML_DEFAULT_BASE_SERVE_URL="${CLEARML_DEFAULT_BASE_SERVE_URL:-http://127.0.0.1:$SERVING_PORT/serve}"
CLEARML_DEFAULT_TRITON_GRPC_ADDR="${CLEARML_DEFAULT_TRITON_GRPC_ADDR:-127.0.0.1:8001}"
# print configuration
echo WEB_CONCURRENCY="$WEB_CONCURRENCY"
echo SERVING_PORT="$SERVING_PORT"
echo GUNICORN_NUM_PROCESS="$GUNICORN_NUM_PROCESS"
echo GUNICORN_SERVING_TIMEOUT="$GUNICORN_SERVING_PORT"
echo GUNICORN_MAX_REQUESTS="$GUNICORN_MAX_REQUESTS"
echo GUNICORN_EXTRA_ARGS="$GUNICORN_EXTRA_ARGS"
echo UVICORN_SERVE_LOOP="$UVICORN_SERVE_LOOP"
echo UVICORN_EXTRA_ARGS="$UVICORN_EXTRA_ARGS"
echo UVICORN_LOG_LEVEL="$UVICORN_LOG_LEVEL"
echo CLEARML_DEFAULT_BASE_SERVE_URL="$CLEARML_DEFAULT_BASE_SERVE_URL"
echo CLEARML_DEFAULT_TRITON_GRPC_ADDR="$CLEARML_DEFAULT_TRITON_GRPC_ADDR"
# runtime add extra python packages
if [ ! -z "$CLEARML_EXTRA_PYTHON_PACKAGES" ]
then
python3 -m pip install $CLEARML_EXTRA_PYTHON_PACKAGES
fi
if [ -z "$CLEARML_USE_GUNICORN" ]
then
echo "Starting Uvicorn server"
PYTHONPATH=$(pwd) python3 -m uvicorn \
clearml_serving.serving.main:app --log-level $UVICORN_LOG_LEVEL --host 0.0.0.0 --port $SERVING_PORT --loop $UVICORN_SERVE_LOOP \
$UVICORN_EXTRA_ARGS
else
echo "Starting Gunicorn server"
# start service
PYTHONPATH=$(pwd) python3 -m gunicorn \
--preload clearml_serving.serving.main:app \
--workers $GUNICORN_NUM_PROCESS \
--worker-class uvicorn.workers.UvicornWorker \
--max-requests $GUNICORN_MAX_REQUESTS \
--timeout $GUNICORN_SERVING_TIMEOUT \
--bind 0.0.0.0:$SERVING_PORT \
$GUNICORN_EXTRA_ARGS
fi