Airflow 2.x distributed deployment combat (qbit)

Preface

• Start of the test used MySQL 8 as a database, as of 2021.5.13, airflow 2.0.2 of this problem is not resolved, it is converted to use PostgreSQL 12
• airflow is a task management system of DAG (directed acyclic graph), simple understanding is an advanced version of crontab.
• Airflow solves the task dependency problem that crontab cannot solve.
• airflow basic architecture
  
• airflow + celery architecture
  

Environment and components

• Ubuntu 20.04
• Python-3.8（Anaconda3-2020.11-Linux-x86_64）
• PostgreSQL 12.6
• apache-airflow 2.0.2
• celery 4.4.7

Cluster planning

installation steps

• Create an account (node0/node1/node2)

  sudo useradd airflow -m -s /bin/bash
  sudo passwd airflow

• Switch account (node0/node1/node2)

  su airflow

• Configure Anaconda environment variables (node0/node1/node2)

  # /home/airflow/.bashrc
  export PATH=/home/airflow/anaconda3/bin:$PATH

• Upgrade pip (node0/node1/node2)

  pip install pip --upgrade  -i https://mirrors.aliyun.com/pypi/simple/

• Configure pip domestic mirror (node0/node1/node2)

  pip3 config set global.index-url https://mirrors.aliyun.com/pypi/simple/

• Install airflow (node0/node1/node2) dependencies: https://airflow.apache.org/docs/apache-airflow/2.0.2/extra-packages-ref.html

  # 全家桶（master）
  pip3 install "apache-airflow[all]~=2.0.2"
  # OR 选择性安装
  pip3 install "apache-airflow[async,postgres,mongo,redis,rabbitmq,celery,dask]~=2.0.2"

• Add PATH environment variable for airflow (node0/node1/node2)

  # 在 /home/airflow/.bashrc 文件尾追加以下内容：
  export PATH=/home/airflow/.local/bin:$PATH

• Check the airflow version and create the home directory of airflow (node0/node1/node2)

  # 默认 ~/airflow 目录
  airflow version

• Set Ubuntu system time zone (node0/node1/node2)

  timedatectl set-timezone Asia/Shanghai

• Modify the time zone in airflow (/home/airflow/airflow/airflow.cfg) (node0/node1/node2)

  [core]
  # 改为 system 或 Asia/Shanghai
  default_timezone = system

• At this point, the installation is complete

PostgreSQL configuration

• Create database

  CREATE DATABASE airflow_db;

• Create user

  CREATE USER airflow_user WITH PASSWORD 'airflow_pass';
  GRANT ALL PRIVILEGES ON DATABASE airflow_db TO airflow_user;

• Modify PostgreSQL connection (/home/airflow/airflow/airflow.cfg) (node0/node1/node2)

  [core]
  sql_alchemy_conn = postgresql+psycopg2://airflow:airflow@192.168.x.y/airflow

• Initialize the database table (node0)

  airflow db init

• Check whether the database is initialized successfully
  

WEB UI login

• creates an administrator user (node0)

  # 角色表： ab_role
  # 用户表： ab_user
  # 创建 Admin 角色用户 
  airflow users create \
   --lastname user \
   --firstname admin \
   --username admin \
   --email [e-mail] \
   --role Admin \
   --password admin123
  # 创建 Viewer 角色用户 
  airflow users create \
   --lastname user \
   --firstname view \
   --username view \
   --email [e-mail] \
   --role Viewer \
   --password view123

• Start webserver (node0)

  airflow webserver -p 8080

• Log in with the created account in the browser
  

Configuration CeleryExecutor

• official document: 161adab97db308 https://docs.celeryproject.org/en/v4.4.7/userguide/configuration.html
• Create an airflow account on RabbitMQ and assign a virtual host
  

• Modify the configuration file (/home/airflow/airflow/airflow.cfg) (node0/node1/node2)

  [core]
  executor = CeleryExecutor
  [celery]
  broker_url = amqp://airflow:mq_pwd@192.168.y.z:5672/vhost_airflow
  result_backend = db+postgresql://airflow:airflow@192.168.x.y/airflow

Test Case

• Create a test script (/home/airflow/airflow/dags/send_msg.py) (node0/node1/node2), and send the local IP to the enterprise WeChat.

  # encoding: utf-8
  # author: qbit
  # date: 2021-05-13
  # summary: 发送/分配任务到任务结点
  
  import os
  import time
  import json
  import psutil
  import requests
  from datetime import timedelta
  from airflow.utils.dates import days_ago
  from airflow.models import DAG
  from airflow.operators.python_operator import PythonOperator
  
  
  def GetLocalIPByPrefix(prefix):
    r"""
  多网卡情况下，根据前缀获取IP
  测试可用：Windows、Linux，Python 3.6.x，psutil 5.4.x
  ipv4/ipv6 地址均适用
  注意如果有多个相同前缀的 ip，只随机返回一个
  """
    localIP = ''
    dic = psutil.net_if_addrs()
    for adapter in dic:
        snicList = dic[adapter]
        for snic in snicList:
            if not snic.family.name.startswith('AF_INET'):
                continue
            ip = snic.address
            if ip.startswith(prefix):
                localIP = ip
  
    return localIP
  
  
  def send_msg(msg='default msg', **context):
    r""" 发送 message 到企业微信 """
    print(context)
    run_id = context['run_id']
    nowTime = time.strftime('%Y-%m-%d %H:%M:%S', time.localtime())
    message = '%s\n%s\n%s_%d\n%s' % (
        run_id, nowTime, GetLocalIPByPrefix('192.168.'), os.getpid(), msg)
    print(message)
  
    '''
    发送代码（涉及账号，本段代码隐藏）
    '''
  
  default_args = {
    'owner': 'qbit',
    # depends_on_past 是否依赖于过去。
    # 如果为True，那么必须要上次的 DAG 执行成功了，这次的 DAG 才能执行。
    'depends_on_past': False
  }
  
  with DAG(dag_id='send_msg',
         default_args=default_args,
         start_date=days_ago(1),
         schedule_interval=timedelta(seconds=60),
         # catchup 是否回补（backfill）开始时间到现在的任务
         catchup=False,
         tags=['qbit']
  ) as dag:
    first = PythonOperator(
        task_id='send_msg_1',
        python_callable=send_msg,
        op_kwargs={'msg': '111'},
        provide_context=True,
        dag=dag,
    )
  
    second = PythonOperator(
        task_id='send_msg_2',
        python_callable=send_msg,
        op_kwargs={'msg': '222'},
        provide_context=True,
        dag=dag,
    )
  
    third = PythonOperator(
        task_id='send_msg_3',
        python_callable=send_msg,
        op_kwargs={'msg': '333'},
        provide_context=True,
        dag=dag,
    )
  
    [third, first] >> second

• View dag information (node0)

  # 打印出所有正在活跃状态的 DAGs
  $ airflow dags list
  
  # 打印出 'send_msg' DAG 中所有的任务
  $ airflow tasks list send_msg
  [2021-05-13 16:00:47,123] {dagbag.py:451} INFO - Filling up the DagBag from /home/airflow/airflow/dags
  send_msg_1
  send_msg_2
  send_msg_3
  
  # 打印出 'send_msg' DAG 的任务层次结构
  $ airflow tasks list send_msg --tree

• Test a single task (node0)

  airflow tasks test send_msg send_msg_1 20210513

• Test a single dag (node0)

  airflow dags test send_msg 20210513

• Cluster test

  # node0
  airflow webserver -p 8080
  airflow scheduler
  airflow celery flower  # 默认端口 5555
  # node1/node2
  airflow celery worker
  # 指定 hostname 启动
  airflow celery worker --celery-hostname node1

  

references

• Related reading: Airflow2.0.0 + Celery cluster building
• Compare CeleryExecutor and DaskExecutor in airflow: A Gentle Introduction To Understand Airflow Executor

This article is from qbit snap