Airflow 从入门到精通-03-完整 ETL 实例

本节将讲述使用 Connection、MyqLOperator、XComs 来实现一个完整的airflow ETL。

一、将数据存入数据库的原始方法

1、创建表

CREATE database demodb;

use demodb;

create table stock_prices_stage(
ticker varchar(30),
as_of_date date,
open_price double,
high_price double,
low_price double,
close_price double

)  COMMENT = '股票价格缓冲区表';

create table stock_prices(
id int not null AUTO_INCREMENT,
ticker varchar(30),
as_of_date date  COMMENT '当前日期',
open_price double,
high_price double,
low_price double,
close_price double,
created_at timestamp default now(),
updated_at timestamp default now(),
primary key (id)
)COMMENT = '股票价格表';

create index ids_stockprices on stock_prices(ticker, as_of_date);

create index ids_stockpricestage on stock_prices_stage(ticker, as_of_date);

二、使用 airflow Connection 管理数据库连接信息

在上一节代码的基础上,将保存到文件的数据转存到数据库中,V2版本的代码如下:

download_stock_price_v2.py

2.1 传统连接方法

"""Example DAG demonstrating the usage of the BashOperator."""

from datetime import timedelta
from textwrap import dedent
import yfinance as yf

from airflow import DAG
from airflow.operators.python import PythonOperator
from airflow.utils.dates import days_ago
from airflow.models import Variable
import mysql.connector


def download_price(*args, **context):
    stock_list = get_tickers(context)
    for ticker in stock_list:
        dat = yf.Ticker(ticker)
        hist = dat.history(period="1mo")
        # print(type(hist))
        # print(hist.shape)
        # print(os.getcwd())

        with open(get_file_path(ticker), 'w') as writer:
            hist.to_csv(writer, index=True)
        print("Finished downloading price data for " + ticker)


def get_file_path(ticker):
    # NOT SAVE in distributed system
    return f'./{ticker}.csv'


def load_price_data(ticker):
    with open(get_file_path(ticker), 'r') as reader:
        lines = reader.readlines()
        return [[ticker] + line.split(',')[:5] for line in lines if line[:4] != 'Date']


def get_tickers(context):
    # 获取配置的变量Variables
    stock_list = Variable.get("stock_list_json", deserialize_json=True)

    # 如果有配置参数,则使用配置参数的数据(Trigger DAG with parameters)
    stocks = context["dag_run"].conf.get("stocks")
    if stocks:
        stock_list = stocks
    return stock_list


def save_to_mysql_stage(*args, **context):
    tickers = get_tickers(context)

    # 连接数据库
    mydb = mysql.connector.connect(
        host="98.14.13.15",
        user="root",
        password="Quant888",
        database="demodb",
        port=3307
    )

    mycursor = mydb.cursor()
    for ticker in tickers:
        val = load_price_data(ticker)
        print(f"{ticker} length={len(val)} {val[1]}")

        sql = """INSERT INTO stock_prices_stage
        (ticker, as_of_date, open_price, high_price, low_price, close_price)
        VALUES (%s,%s,%s,%s,%s,%s)"""
        mycursor.executemany(sql, val)

        mydb.commit()
        print(mycursor.rowcount, "record inserted.")


default_args = {
    'owner': 'airflow'
}

# [START instantiate_dag]
with DAG(
        dag_id='download_stock_price_v2',
        default_args=default_args,
        description='download stock price and save to local csv files and save to database',
        schedule_interval=None,
        start_date=days_ago(2),
        tags=['quantdata'],
) as dag:
    # [END instantiate_dag]

    dag.doc_md = """
    This DAG download stock price
    """

    download_task = PythonOperator(
        task_id="download_prices",
        python_callable=download_price,
        provide_context=True
    )

    save_to_mysql_task = PythonOperator(
        task_id="save_to_database",
        python_callable=save_to_mysql_stage,
        provide_context=True
    )

    download_task >> save_to_mysql_task

然后在 airflow 后台手动触发执行,前两次执行失败,后边调试后,执行成功了
file

可以看到数据已经入库了:
file

2.2 airflow Connection管理连接信息

上边的demo有些问题,将数据库的连接直接硬编码到代码中了,这样后期维护不是很好,airflow给我们提供了 Connections 连接方法,可以使用该方法将连接信息直接写入到这里即可。

file

选择连接类型,缺少了MySQL连接类型:
file

Conn Type missing? Make sure you've installed the corresponding Airflow Provider Package.

请看官方文档:
https://airflow.apache.org/do...
https://airflow.apache.org/do...
https://airflow.apache.org/do...

file

file

$ pip install apache-airflow-providers-mysql

然后重新刷新连接页面,可以看到连接类型 MySQL 已经出现了:

file

然后填入相关的数据库连接信息:
file

然后对代码进行修改:

def save_to_mysql_stage(*args, **context):
    tickers = get_tickers(context)

    """
    # 连接数据库(硬编码方式连接)
    mydb = mysql.connector.connect(
        host="98.14.14.145",
        user="root",
        password="Quant888",
        database="demodb",
        port=3307
    )
    """

    # 使用airflow 的 Connections 动态获取配置信息
    from airflow.hooks.base_hook import BaseHook
    conn = BaseHook.get_connection('demodb')

    mydb = mysql.connector.connect(
        host=conn.host,
        user=conn.login,
        password=conn.password,
        database=conn.schema,
        port=conn.port
    )

    mycursor = mydb.cursor()
    for ticker in tickers:
        val = load_price_data(ticker)
        print(f"{ticker} length={len(val)} {val[1]}")

        sql = """INSERT INTO stock_prices_stage
        (ticker, as_of_date, open_price, high_price, low_price, close_price)
        VALUES (%s,%s,%s,%s,%s,%s)"""
        mycursor.executemany(sql, val)

        mydb.commit()
        print(mycursor.rowcount, "record inserted.")

完整代码:

"""Example DAG demonstrating the usage of the BashOperator."""

from datetime import timedelta
from textwrap import dedent
import yfinance as yf

from airflow import DAG
from airflow.operators.python import PythonOperator
from airflow.utils.dates import days_ago
from airflow.models import Variable
import mysql.connector


def download_price(*args, **context):
    stock_list = get_tickers(context)
    for ticker in stock_list:
        dat = yf.Ticker(ticker)
        hist = dat.history(period="1mo")
        # print(type(hist))
        # print(hist.shape)
        # print(os.getcwd())

        with open(get_file_path(ticker), 'w') as writer:
            hist.to_csv(writer, index=True)
        print("Finished downloading price data for " + ticker)


def get_file_path(ticker):
    # NOT SAVE in distributed system
    return f'./{ticker}.csv'


def load_price_data(ticker):
    with open(get_file_path(ticker), 'r') as reader:
        lines = reader.readlines()
        return [[ticker] + line.split(',')[:5] for line in lines if line[:4] != 'Date']


def get_tickers(context):
    # 获取配置的变量Variables
    stock_list = Variable.get("stock_list_json", deserialize_json=True)

    # 如果有配置参数,则使用配置参数的数据(Trigger DAG with parameters)
    stocks = context["dag_run"].conf.get("stocks")
    if stocks:
        stock_list = stocks
    return stock_list


def save_to_mysql_stage(*args, **context):
    tickers = get_tickers(context)

    """
    # 连接数据库(硬编码方式连接)
    mydb = mysql.connector.connect(
        host="98.14.13.14",
        user="root",
        password="Quan888",
        database="demodb",
        port=3307
    )
    """

    # 使用airflow 的 Connections 动态获取配置信息
    from airflow.hooks.base_hook import BaseHook
    conn = BaseHook.get_connection('demodb')

    mydb = mysql.connector.connect(
        host=conn.host,
        user=conn.login,
        password=conn.password,
        database=conn.schema,
        port=conn.port
    )

    mycursor = mydb.cursor()
    for ticker in tickers:
        val = load_price_data(ticker)
        print(f"{ticker} length={len(val)} {val[1]}")

        sql = """INSERT INTO stock_prices_stage
        (ticker, as_of_date, open_price, high_price, low_price, close_price)
        VALUES (%s,%s,%s,%s,%s,%s)"""
        mycursor.executemany(sql, val)

        mydb.commit()
        print(mycursor.rowcount, "record inserted.")


default_args = {
    'owner': 'airflow'
}

# [START instantiate_dag]
with DAG(
        dag_id='download_stock_price_v2',
        default_args=default_args,
        description='download stock price and save to local csv files and save to database',
        schedule_interval=None,
        start_date=days_ago(2),
        tags=['quantdata'],
) as dag:
    # [END instantiate_dag]

    dag.doc_md = """
    This DAG download stock price
    """

    download_task = PythonOperator(
        task_id="download_prices",
        python_callable=download_price,
        provide_context=True
    )

    save_to_mysql_task = PythonOperator(
        task_id="save_to_database",
        python_callable=save_to_mysql_stage,
        provide_context=True
    )

    download_task >> save_to_mysql_task

三、使用 MyqLOperator 执行数据库操作

dags/ 目录下新建sql文件,用来合并缓冲表(stage)的数据到正式表。

file

merge_stock_price.sql

-- update the existing rows
UPDATE stock_prices p, stock_prices_stage s
SET p.open_price = s.open_price,
    p.high_price = s.high_price,
        p.low_price = s.low_price,
        p.close_price = s.close_price,
        updated_at = now()
WHERE p.ticker = s.ticker
AND p.as_of_date = s.as_of_date;

-- inserting new rows
INSERT INTO stock_prices
(ticker,as_of_date,open_price,high_price,low_price,close_price)
SELECT ticker,as_of_date,open_price,high_price,low_price,close_price
FROM stock_prices_stage s
WHERE NOT EXISTS
(SELECT 1 FROM stock_prices p
  WHERE p.ticker = s.ticker
    AND p.as_of_date = s.as_of_date);

-- truncate the stage table;
TRUNCATE TABLE stock_prices_stage;

download_stock_price_v2.py 文件新建 MySQL task 任务:
需要先引入:

from airflow.providers.mysql.operators.mysql import MySqlOperator
 mysql_task = MySqlOperator(
        task_id="merge_stock_price",
        mysql_conn_id='demodb',
        sql="merge_stock_price.sql",
        dag=dag,
    )

    download_task >> save_to_mysql_task >> mysql_task

完整代码:

"""Example DAG demonstrating the usage of the BashOperator."""

from datetime import timedelta
from textwrap import dedent
import yfinance as yf
import mysql.connector

from airflow import DAG
from airflow.operators.python import PythonOperator
# from airflow.operators.mysql_operator import MySqlOperator
from airflow.providers.mysql.operators.mysql import MySqlOperator

from airflow.utils.dates import days_ago
from airflow.models import Variable

def download_price(*args, **context):
    stock_list = get_tickers(context)
    for ticker in stock_list:
        dat = yf.Ticker(ticker)
        hist = dat.history(period="1mo")
        # print(type(hist))
        # print(hist.shape)
        # print(os.getcwd())

        with open(get_file_path(ticker), 'w') as writer:
            hist.to_csv(writer, index=True)
        print("Finished downloading price data for " + ticker)


def get_file_path(ticker):
    # NOT SAVE in distributed system
    return f'./{ticker}.csv'


def load_price_data(ticker):
    with open(get_file_path(ticker), 'r') as reader:
        lines = reader.readlines()
        return [[ticker] + line.split(',')[:5] for line in lines if line[:4] != 'Date']


def get_tickers(context):
    # 获取配置的变量Variables
    stock_list = Variable.get("stock_list_json", deserialize_json=True)

    # 如果有配置参数,则使用配置参数的数据(Trigger DAG with parameters)
    stocks = context["dag_run"].conf.get("stocks")
    if stocks:
        stock_list = stocks
    return stock_list


def save_to_mysql_stage(*args, **context):
    tickers = get_tickers(context)

    """
    # 连接数据库(硬编码方式连接)
    mydb = mysql.connector.connect(
        host="98.14.14.15",
        user="root",
        password="Quan888",
        database="demodb",
        port=3307
    )
    """

    # 使用airflow 的 Connections 动态获取配置信息
    from airflow.hooks.base_hook import BaseHook
    conn = BaseHook.get_connection('demodb')

    mydb = mysql.connector.connect(
        host=conn.host,
        user=conn.login,
        password=conn.password,
        database=conn.schema,
        port=conn.port
    )

    mycursor = mydb.cursor()
    for ticker in tickers:
        val = load_price_data(ticker)
        print(f"{ticker} length={len(val)} {val[1]}")

        sql = """INSERT INTO stock_prices_stage
        (ticker, as_of_date, open_price, high_price, low_price, close_price)
        VALUES (%s,%s,%s,%s,%s,%s)"""
        mycursor.executemany(sql, val)

        mydb.commit()
        print(mycursor.rowcount, "record inserted.")


default_args = {
    'owner': 'airflow'
}

# [START instantiate_dag]
with DAG(
        dag_id='download_stock_price_v2',
        default_args=default_args,
        description='download stock price and save to local csv files and save to database',
        schedule_interval=None,
        start_date=days_ago(2),
        tags=['quantdata'],
) as dag:
    # [END instantiate_dag]

    dag.doc_md = """
    This DAG download stock price
    """

    download_task = PythonOperator(
        task_id="download_prices",
        python_callable=download_price,
        provide_context=True
    )

    save_to_mysql_task = PythonOperator(
        task_id="save_to_database",
        python_callable=save_to_mysql_stage,
        provide_context=True
    )

    mysql_task = MySqlOperator(
        task_id="merge_stock_price",
        mysql_conn_id='demodb',
        sql="merge_stock_price.sql",
        dag=dag,
    )

    download_task >> save_to_mysql_task >> mysql_task

然后手动执行airflow,可以看到已经执行成功了:
file

然后看相关表数据,也已经更新成功了
file

四、使用 XComs 在任务之间传递数据

XComs 概念

XComs(“交叉通信”的缩写)是一种让任务相互通信的机制,因为默认情况下任务是完全隔离的,并且可能运行在完全不同的机器上。

XCom 由一个键(本质上是它的名称)以及它来自的 task_id 和 dag_id 标识。它们可以具有任何(可序列化的)值,但它们仅适用于少量数据;不要使用它们来传递大值,例如数据帧。

简单一句话,XComs可以在多个task之间进行通信(数据的传递)

XComs are explicitly "pushed" and "pulled" to/from their storage using the xcom_push and xcom_pull methods on Task Instances. Many operators will auto-push their results into an XCom key called return_value if the do_xcom_push argument is set to True (as it is by default), and @task functions do this as well.

# Pulls the return_value XCOM from "pushing_task"
value = task_instance.xcom_pull(task_ids='pushing_task')

实战应用

使用场景:增加一支不存在股票,然后对这只股票进行验证,存在的股票才可以传入到后边。

修改 download_stock_price_v2.py 文件下载代码:
file

然后将股票保存到MySQL stage 时,通过上一步返回的股票来获取已经过滤的ticker。

file

download_stock_price_v2.py 完整代码

"""Example DAG demonstrating the usage of the BashOperator."""

from datetime import timedelta
from textwrap import dedent
import yfinance as yf
import mysql.connector

from airflow import DAG
from airflow.operators.python import PythonOperator
# from airflow.operators.mysql_operator import MySqlOperator
from airflow.providers.mysql.operators.mysql import MySqlOperator

from airflow.utils.dates import days_ago
from airflow.models import Variable



def download_price(*args, **context):
    stock_list = get_tickers(context)

    # 新增正常的股票(没有退市的或不存在的)
    valid_tickers = []
    for ticker in stock_list:
        dat = yf.Ticker(ticker)
        hist = dat.history(period="1mo")
        # print(type(hist))
        # print(hist.shape)
        # print(os.getcwd())

        if hist.shape[0] > 0:
            valid_tickers.append(ticker)
        else:
            continue

        with open(get_file_path(ticker), 'w') as writer:
            hist.to_csv(writer, index=True)
        print("Finished downloading price data for " + ticker)
    # 增加返回值(用于任务之间数据的传递)
    return valid_tickers


def get_file_path(ticker):
    # NOT SAVE in distributed system
    return f'./{ticker}.csv'


def load_price_data(ticker):
    with open(get_file_path(ticker), 'r') as reader:
        lines = reader.readlines()
        return [[ticker] + line.split(',')[:5] for line in lines if line[:4] != 'Date']


def get_tickers(context):
    # 获取配置的变量Variables
    stock_list = Variable.get("stock_list_json", deserialize_json=True)

    # 如果有配置参数,则使用配置参数的数据(Trigger DAG with parameters)
    stocks = context["dag_run"].conf.get("stocks")
    if stocks:
        stock_list = stocks
    return stock_list


def save_to_mysql_stage(*args, **context):
    # tickers = get_tickers(context)
    # Pull the return_value XCOM from "pulling_task"
    tickers = context['ti'].xcom_pull(task_ids='download_prices')
    print(f"received tickers:{tickers}")

    """
    # 连接数据库(硬编码方式连接)
    mydb = mysql.connector.connect(
        host="98.14.14.15",
        user="root",
        password="Quant888",
        database="demodb",
        port=3307
    )
    """

    # 使用airflow 的 Connections 动态获取配置信息
    from airflow.hooks.base_hook import BaseHook
    conn = BaseHook.get_connection('demodb')

    mydb = mysql.connector.connect(
        host=conn.host,
        user=conn.login,
        password=conn.password,
        database=conn.schema,
        port=conn.port
    )

    mycursor = mydb.cursor()
    for ticker in tickers:
        val = load_price_data(ticker)
        print(f"{ticker} length={len(val)} {val[1]}")

        sql = """INSERT INTO stock_prices_stage
        (ticker, as_of_date, open_price, high_price, low_price, close_price)
        VALUES (%s,%s,%s,%s,%s,%s)"""
        mycursor.executemany(sql, val)

        mydb.commit()
        print(mycursor.rowcount, "record inserted.")


default_args = {
    'owner': 'airflow'
}

# [START instantiate_dag]
with DAG(
        dag_id='download_stock_price_v2',
        default_args=default_args,
        description='download stock price and save to local csv files and save to database',
        schedule_interval=None,
        start_date=days_ago(2),
        tags=['quantdata'],
) as dag:
    # [END instantiate_dag]

    dag.doc_md = """
    This DAG download stock price
    """

    download_task = PythonOperator(
        task_id="download_prices",
        python_callable=download_price,
        provide_context=True
    )

    save_to_mysql_task = PythonOperator(
        task_id="save_to_database",
        python_callable=save_to_mysql_stage,
        provide_context=True
    )

    mysql_task = MySqlOperator(
        task_id="merge_stock_price",
        mysql_conn_id='demodb',
        sql="merge_stock_price.sql",
        dag=dag,
    )

    download_task >> save_to_mysql_task >> mysql_task

然后在 Variables 增加一个不存在的 ticker(FBXXOO),以此来验证Xcom数据传递进行验证:
file

手动执行DAG,可以通过日志打印看到已经获取到了 Xcom tickers = context['ti'].xcom_pull(task_ids='download_prices')
上一个任务传递过来的数据了。

file

file


相关文章:
Airflow 相关概念文档
Airflow XComs官方文档


Corwien
为者常成,行者常至!

为者常成,行者常至。

6.3k 声望
1.6k 粉丝
0 条评论
推荐阅读
CDH6 离线安装
Cloudera Manager是一个拥有集群自动化安装、中心化管理、集群监控、报警功能的一个工具,使得安装集群从几天的时间缩短在几个小时内,运维人员从数十人降低到几人以内,极大的提高集群管理的效率。

Corwien2阅读 1.9k

Ubuntu20.04 从源代码编译安装 python3.10
Ubuntu 22.04 Release DateUbuntu 22.04 Jammy Jellyfish is scheduled for release on April 21, 2022If you’re ready to use Ubuntu 22.04 Jammy Jellyfish, you can either upgrade your current Ubuntu syste...

ponponon1阅读 4k评论 1

日常Python 代码片段整理
1、简单的 HTTP Web 服务器 {代码...} 2、单行循环List {代码...} 3、更新字典 {代码...} 4、拆分多行字符串 {代码...} 5、跟踪列表中元素的频率 {代码...} 6、不使用 Pandas 读取 CSV 文件 {代码...} 7、将列表...

墨城2阅读 340

Unicode 正则表达式(qbit)
前言本文根据《精通正则表达式》和 Unicode Regular Expressions 整理。本文的示例默认以 Python3 为实现语言,用到 Python3 的 re 模块或 regex 库。基本的 Unicode 属性分类 {代码...} 基本的 Unicode 子属性Le...

qbit阅读 4.4k

Python + Sqlalchemy 对数据库的批量插入或更新(Upsert)
由于不同数据库对这种 upsert 的实现机制不同,Sqlalchemy 也就不再试图做一致性的封装了,而是提供了各自的方言 API,具体到 Mysql,就是给 insert statement ,增加了 on_duplicate_key_update 方法。

songofhawk1阅读 2.1k评论 4

封面图
打脸了兄弟们,Go1.20 arena 来了!
大家好,我是煎鱼。大概半年前,我写过一篇文章《Go 要违背初心吗?新提案:手动管理内存》。有兴趣了深入解的同学,可以再回顾一下。当时我们还想着 Go 团队应该不会接纳,至少不会那么快:懒得翻也可以看我再次...

煎鱼1阅读 3.3k

uwsgi 注意事项
http 和 http-socket 选项是完全不同的。第一个生成一个额外的进程,转发请求到一系列的worker (将它想象为一种形式的盾牌,与apache或者nginx同级),而第二个设置worker为原生使用http协议。

zed2015阅读 2.2k

为者常成,行者常至。

6.3k 声望
1.6k 粉丝
宣传栏