sudo yum install epel-release
sudo yum -y install gcc gcc-c++ libffi-devel mariadb-devel cyrus-sasl-devel python-devel
easy_install -U setuptools
mysql一般默认安装了
配置：
创建数据库：airflow create database airflow default character set utf8mb4 collate utf8mb4_unicode_ci;
创建用户并授权：airflow
GRANT all privileges on airflow.* TO 'airflow'@'localhost' IDENTIFIED BY 'airflow';
yum install redis
开启redis服务
sudo systemctl start redis.server
测试redis安装成功
redis-cli ping
~/.bash_profile配置如下
export WORKON_HOME=$HOME/.virtualenvs
export PROJECT_HOME=$HOME/workspace
source /usr/bin/virtualenvwrapper.sh
workon py2
注意：用Python2，因为airflow[hdfs]依赖的snakebite不支持python3
安装airflow （kerberos模块，如果hadoop没使用kerberos认证服务，则不需安装）
pip install airflow[async,celery,crypto,druid,jdbc,hdfs,hive,kerberos,ldap,mysql,password,vertica]
使用CeleryExecutor模式（redis为broker）：
pip install futures
pip install redis
pip install -U "celery[redis]"
初始化db
airflow initdb
(py2) [airflow@AIRFLOW airflow]$ ipython
In [1]: import airflow
In [2]: from airflow import models, settings
In [3]: from airflow.contrib.auth.backends.password_auth import PasswordUser
In [4]: user = PasswordUser(models.User())
In [5]: user.username = 'airflow'
In [6]: user.email = 'airflow@airflow.com'
In [7]: user.password = '*******'
In [8]: session = settings.Session()
In [9]: session.add(user)
In [10]: session.commit()
In [11]: session.close()
In [12]: quit
配置方法参考：https://github.com/apache/incubator-airflow/tree/master/scripts/systemd
或https://github.com/JazzChen/InstallAirflow/tree/master/airflow/script/systemd
(py2) [airflow@AIRFLOW systemd]$ ls
airflow airflow.conf airflow-flower.service airflow-kerberos.service airflow-scheduler.service airflow-webserver.service airflow-worker.service
调整依赖服务，用户/用户组，启动命令（注意使用python虚拟环境，启动命名较复杂）