Skip to content

Latest commit

 

History

History
239 lines (212 loc) · 9.09 KB

README.md

File metadata and controls

239 lines (212 loc) · 9.09 KB
Raw

pyboot

English | Chinese

The original purpose of this project was to call scientific computing models on the edge for edge computing scenarios.

Architecture

arch

htop

Features

1. starter

The system components can be expanded through the implementation of Starter. At present, the system provides several components for configuration, multi-process processing, and webserver. Finally, the system interrupts and stop signals are uniformly accepted to call the Stop function of HookStarter to recover all resources and stop the service.

  1. BaseConfStarter: Provide the relevant parameters of a unified custom configuration for the entire system. The relevant configuration information can be expanded by modifying the config.yaml and config.py files, and the system built-in parameters are provided by the settings.py file;
  2. ProcessorStarter: Provides multi-process and multi-threaded call functions for edge model calls. The default overall call logic of an edge model will be extended to an independent sub-process, and for a model call, its data transmission and reception and model call, then Use more in the multithreading of this child process;
  3. TornadoServer or FlaskStarter: Provides a unified web service for the main process, currently provides two optional frameworks based on Flask and Tornado;
  4. HookStarter: The Stop functions of all integrated components are registered and managed uniformly, and the resources can be recycled in order by priority, and finally the operation of the entire service is stopped;

2. boot

The boot program that the system starts, loads each Starter configured in brun/init.py, and executes the Init, Setup, and Start functions of all Starter components in turn when boot.Starter() is called. The penultimate Starter should be It is a component with blocking function (such as FlaskStarter, TornadoStarter), the last one should be HookStarter;

Configuration

v1

# mqtt broker resources configuration
mqtts:
  - name: b1
    broker: tcp://192.168.241.1:1883
    qos: 0
    retain: false
  - name: b2
    broker: tcp://192.168.241.1:1883
    qos: 0
    retain: false
# rules configuration
rules:
  # subprocess 1
    # input data from mqtt broker
  - sub:
      name: b1
      clientId: 5555
      topic: /gridsum/test/telm/in/thend_threshold_test
    # output data from mqtt broker
    pub:
      name: b1
      clientId: 6666
      timeout: 10s
      topic: /gridsum/test/telm/out/thend_threshold_test
  # subprocess 2
  - sub:
      name: b1
      clientId: 7777
      topic: /gridsum/test/telm/in/thend_multivar_test
    pub:
      name: b1
      clientId: 8888
      timeout: 10s
      topic: /gridsum/test/telm/out/thend_multivar_test
funcs:
  - model_address: http://10.200.60.18:22122/group1/M00/00/0F/Ch1hQF-NYuyAHmhzAAHaYnbCrrs964.zip
    model_md5: pZ4DJ+YhYM9ppHD+VhLi+A==
    # edge model config path: pyboot/modules/gridsum/science/{modelName}
    # edge model reference function entry name for importlib: pyboot.modules.gridsum.science.{modelName}.index.index(event, context)
    modelName: thend_threshold_zd
    deviceName: 设备3
    pointName: OPC温度

v3

expose:
- desc: monitor the container
  name: monitor
  outPort: 38080
  protocolPort: http://:8080
funcs:
- devices:
  - deviceName: "三厂-pyboot测试设备1"
    deviceAttr:
    - attrExpression: == data_to_dict['deviceInfo']['deviceName']
      attrName: deviceName
      attrValue: 三厂-pyboot测试设备1
    - attrExpression: in data_to_dict['telemetry']
      attrName: pointName
      attrValue: pyboot1
  modelAddress: http://10.200.60.18:9000/group1/M00/49/04/Csg8EmGdnIyAVJdVAAHaY0adxNc829.zip
  modelMd5: pZ4DJ+YhYM9ppHD+VhLi+A==
  modelName: industry
- devices:
  - deviceName: "三厂-pyboot测试设备2"
    deviceAttr:
    - attrExpression: == data_to_dict['deviceInfo']['deviceName']
      attrName: deviceName
      attrValue: 三厂-pyboot测试设备2
    - attrExpression: in data_to_dict['telemetry']
      attrName: pointName
      attrValue: pyboot2
  modelAddress: http://10.200.60.18:9000/group1/M00/49/04/Csg8EmGdnIyAVJdVAAHaY0adxNc829.zip
  modelMd5: pZ4DJ+YhYM9ppHD+VhLi+A==
  modelName: test
mqtts:
- broker: tcp://192.168.241.1:1883
  name: mosquitto-20336454-5413-4e2f-b52e-b945291359c7
  retain: false
rules:
- pub:
    clientId: ef2712a2-926d-4388-92d0-60470ab2613d
    name: mosquitto-20336454-5413-4e2f-b52e-b945291359c7
    timeout: 100
    topic: cloud/edge-pyboot-e060ff31-1644570929743-e54d4620-8b1a-11ec-9969-df2a36238047
  sub:
    clientId: ae03af75-13f5-4cf9-9145-0fea4e6ff1e4
    name: mosquitto-20336454-5413-4e2f-b52e-b945291359c7
    topic: cloud/edge-data-access-a60c2306-1644570929743-e54d4620-8b1a-11ec-9969-df2a36238047

Start run

pip install -r /home/requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simple

export PYTHONPATH=$PYTHONPATH:`pwd`:'pwd'/pyboot

python pyboot/brun/main.py

# If you want to customize the incoming configuration file, you can pass in parameters. If you don’t pass it, 
# the configuration file of pyboot/conf/config.yaml will be used by default.
python pyboot/brun/main.py --config pyboot/conf/config.yaml

# If you want the program to automatically download and decompress the model file (currently supports .zip), 
# you need to declare environment variables
export DOWNLOAD_MODEL=True

check the performance

You can send get request to the http://localhost:5888/queue_size_metrics. To obtain whether the performance of service message processing is blocked, if the value of pre_queue and post_queue is greater than 0, it means that the current service has a performance backlog. You can adjust the number of service instances or edge model threads to improve efficiency. However, post_queue generally writes messages to mqtt , When the pressure is high, there may still be some instantaneous backlogs;

performance

test

  1. Install the docker environment locally and download the data generator image:
export GOFAKE_IMAGE_TAG=v6.5.0-7-g4224d58
# Pull image
docker pull docker.gridsumdissector.com/kubeedge/gofakeit-server@${GOFAKE_IMAGE_TAG}
# run
docker run -itd -p 18080:8080 --restart always --name gofaker docker.gridsumdissector.com/kubeedge/gofakeit-server:${GOFAKE_IMAGE_TAG}
  1. Install the mqtt middleware mosquitto locally
yum install mosquitto -y
  1. Download the message forwarding program
# Download the message forwarding program
mkdir $HOME/benthos/bin -p
cd $HOME/benthos/bin
curl --noproxy "*" -X GET -u domainAccount:domainAccountPassword -O http://repository.gridsum.com/repository/cps/pkg/tools/benthos/x86_64/v3.49.0-3-g84709014/linux/amd64/bin/benthos
chmod +x benthos

# Download the transfer program configuration file
mkdir $HOME/benthos/conf -p
cd $HOME/benthos/conf
## Test model 1-pure python function
curl --noproxy "*" -X GET -u 域账号:域账号密码 -O http://repository.gridsum.com/repository/cps/pkg/tools/benthos/x86_64/v3.49.0-3-g84709014/linux/amd64/conf/http_mqtt_for_pyboot.yaml
## Test model 2-load the pickle model file
curl --noproxy "*" -X GET -u 域账号:域账号密码 -O http://repository.gridsum.com/repository/cps/pkg/tools/benthos/x86_64/v3.49.0-3-g84709014/linux/amd64/conf/http_json_mqtt_for_pyboot.yaml
  1. Start the message transfer program
cd $HOME/benthos/bin
## Run the data required by the transfer model 1 to the local mosquitto: topic=/gridsum/test/telm/in/m1
./benthos -c ../conf/http_mqtt_for_pyboot.yaml
## Run the data required by the transfer model 2 to the local mosquitto: topic=/gridsum/test/telm/in/m_test
./benthos -c ../conf/http_json_mqtt_for_pyboot.yaml
  1. Data format can be observed through mqttbox after startup
  • Model 1 transfer data Model 1 transfer data
  • Model 2 transfer data Model 2 transfer data
  1. At this moment, you can go to the pyboot project and start brun/main.py to test the model
  • Model 1 running results Model 1 running results
  • Model 2 running results Model 2 running results
  1. The qos parameters per unit time in the configuration file can be adjusted for performance testing
rate_limit_resources:
  - label: foobar
    local:
      count: 1
      interval: 6s
  • View the backlog of data processing through the interface to help understand model performance: Model running performance

docker

Provides a Makefile file, which can be used to easily generate docker images through the make images command. Currently, python:3.6-slim is used as the basic image;

docker run --name pyboot-local-test -itd -v /lyndon/iProject/pypath/config/goboot.yml:/etc/edge/goboot.yml -e DOWNLOAD_MODEL="True" docker.gridsumdissector.com/kubeedge/pyboot:v1.0.2 --config /etc/edge/goboot.yml

TODO

  • Persistence framework integration
  • Integration testing for some edge models
  • Integrate with kubernetes and kubeedge, and write related yaml resource files;
  • ...