Angel Serving

Angel Serving is standalone industrial serving system for machine/deep learning models, it is designed to flexible and high-performance.

Architecture

---

Features

• One can access Angel Serving through gRPC and Restful API

• Angel Serving is a general machine learning serving framework which means models from other training platform can server on Angel Serving. There is a pluggable mechanism for the third party platform join in, now we support: Angel, PyTorch and PMML format. Through the PMML format, Angel can server Spark and XGBoost models.

• Similar to TensorFlow Serving, we provide fine grain version control: earliest, latest and specified versions.

• Apart from version control, angel serving also provide fine grain service monitoring:

  • QPS: Query per second
  • Success/Total requests
  • Response time distribution
  • Average response Time

Serve a model in 60 seconds

# Download the Angel Serving Docker image and repo
docker pull tencentangel/serving

git clone https://github.com/Angel-ML/serving.git
# Location of demo models
TESTDATA="$(pwd)/serving/models/angel/lr/lr-model"

# Start Angel Serving container and open the REST API port
docker run -t --rm -p 8501:8501 \
    -v "$TESTDATA:/models" \
    -e MODEL_NAME=lr \
    -e MODEL_PLATFORM=angel \
    tencentangel/serving &

# Query the model using the predict API
curl -H "Content-Type: application/json" -X POST -d '{"instances": [[0.51483303, 0.99900955, 0.9477888, 0.6912188, 0.41446745, 0.2525878, 0.6014038, 0.46847868, 0.12854028, 0.8306037, 0.3461753, 0.1129151, 0.6229094, 0.90299904, 0.50834644, 0.34843314, 0.95900637, 0.9437762, 0.31707388, 0.73501045, 0.05600065, 0.47225082, 0.28908283, 0.7371853, 0.55928135, 0.81367457, 0.91782594, 0.008230567, 0.9317811, 0.0061050057, 0.7060979, 0.51740277, 0.07297987, 0.34826308, 0.43395072, 0.5017575, 0.73248106, 0.7576818, 0.43087876, 0.9380423, 0.5226082, 0.9813176, 0.20717019, 0.42229313, 0.8274106, 0.6791944, 0.48174334, 0.77374876, 0.56179315, 0.6584269, 0.7635249, 0.9949779, 0.84034514, 0.7586089, 0.74443096, 0.21172583, 0.7850719, 0.5341459, 0.84134424, 0.06459451, 0.1270392, 0.41439575, 0.98234355, 0.5515572, 0.9594097, 0.18379861, 0.8221523, 0.23739898, 0.07713032, 0.66251403, 0.84977543, 0.905998, 0.21836805, 0.40002906, 0.6271626, 0.37708586, 0.20958215, 0.051997364, 0.6841619, 0.22454417, 0.34285623, 0.19205093, 0.35783356, 0.29280972, 0.19194472, 0.42898583, 0.27232456, 0.12662607, 0.74165606, 0.43464816, 0.8310301, 0.012846947, 0.9810947, 0.43377626, 0.3608846, 0.22756284, 0.6404164, 0.7243295, 0.68765146, 0.12439847, 0.25675082, 0.26143825, 0.41246158, 0.867953, 0.2895738, 0.3916427, 0.93816304, 0.27819514, 0.1989426, 0.62377095, 0.9969712, 0.4159639, 0.70966166, 0.29150474, 0.6492832, 0.10598481, 0.44674253, 0.03885162, 0.25127923, 0.60202503, 0.6067293, 0.94750637, 0.97315085]]}' \
  localhost:8501/v1/models/lr/versions/6:predict

# Returns => {"predictions": [{"trueLabel":"0.0","proba":"0.024818711775534966","pred":"-3.671025514602661","attached":"NaN","predLabel":"-1.0","sid":"0"}]}

Setup

1. Compile Environment Requirements

   • jdk >=1.8
   • maven >= 3.0.5
   • protobuf >= 3.5.1

2. Source Code Download

   git clone https://github.com/Angel-ML/serving.git
   

3. Compile

   Run the following command in the root directory of the source code

   mvn clean package -Dmaven.test.skip=true
   

   After compiling, a distribution package named serving-0.1.0-bin.zip will be generated under dist/target in the root directory.

4. Distribution Package Unpacking the distribution package, subdirectories will be generated under the root directory:

   • bin: contains Angel Serving start scripts.
   • conf: contains system config files.
   • lib: contains jars for Angel Serving and dependencies.
   • models: contains trained example models.
   • docs: contains user manual and restful api documentation.

Deployment Guide

1. Execution Environment Requirements

   • jdk >= 1.8
   • set JAVA_HOME

2. Start Server

   Run the serving-submit with args to start Angel Serving, example:

   $SERVING_HOME/bin/serving-submit \
      --port 8500 \
      --rest_api_port 8501 \
      --model_base_path $SERVING_HOME/models/angel/lr/lr-model/ \
      --model_name lr \ 
      --model_platform angel \
      --enable_metric_summary true
   

Documentation

• User Manual
• Restful API