Airflow使用指南

1.只执行单个任务

将downstream和recursive按钮的点击状态取消，然后点击clear，最后选择Ignore All Deps，然后点击run

2.从一个任务开始，执行它以及它的下游任务

将downstream和recursive按钮的点击状态取消，然后点击clear，最后选择Ignore Task Deps，然后点击run

3.airflow命令行

https://airflow.apache.org/docs/apache-airflow/stable/cli-and-env-variables-ref.html#dags

1.第一次登录创建airflow用户

airflow users create --username airflow --role Admin --password airflow --email airflow@xxx.com --lastname airflow --firstname airflow

2.根据dag id删除一个dag

airflow dags delete {dag_id}

3.触发一个airflow dag

airflow dags trigger --help
usage: airflow dags trigger [-h] [-c CONF] [-e EXEC_DATE] [-r RUN_ID]
                            [-S SUBDIR]
                            dag_id

Trigger a DAG run

positional arguments:
  dag_id                The id of the dag

optional arguments:
  -h, --help            show this help message and exit
  -c CONF, --conf CONF  JSON string that gets pickled into the DagRun's conf attribute
  -e EXEC_DATE, --exec-date EXEC_DATE
                        The execution date of the DAG
  -r RUN_ID, --run-id RUN_ID
                        Helps to identify this run
  -S SUBDIR, --subdir SUBDIR
                        File location or directory from which to look for the dag. Defaults to '[AIRFLOW_HOME]/dags' where [AIRFLOW_HOME] is the value you set for 'AIRFLOW_HOME' config you set in 'airflow.cfg'

airflow dags trigger -e '2022-07-19T08:00:00' your_dag_id

注意execution_time要在start_date和end_date之间，否则会报

ValueError: The execution_date [2022-07-19T08:00:00+00:00] should be >= start_date [2022-07-20T00:00:00+00:00] from DAG's default_args

4.airflow按start_date和end_date触发backfill任务

airflow dags backfill -s 2022-01-01 -e 2022-01-10 DAG_ID

执行的任务是从2022-01-01到2022-01-09，包含startr_date，不包含end_date　　

5.测试airflow task

4.airflow会的connection配置参数

from airflow.hooks.base import BaseHook

connection = BaseHook.get_connection("username_connection")
password = connection.password

5.在airflow界面上触发特定execution date的任务

点击DAG Runs

add a new record

add dag run

state注意填写running

run id格式参考其他的dag：scheduled__2022-09-18T23:00:00+00:00，最后点击save

成功触发一个特定execution date的dag

如果需要回溯的dag特别多，也可以将dag的start_date设置好，并将catchup=True，这样如果是新建的dag的会，就会从start_date开始运行

6.sensor的reschedule mode

对于长时间检测的sensor，如果sensor的条件不满足，reschedule mode的sensor会释放worker给其他的task

In sensor mode='reschedule' means that if the criteria of the sensor isn't True then the sensor will release the worker to other tasks.

This is very useful for cases when sensor may wait for a long time. 参考：

https://stackoverflow.com/questions/66715894/how-does-the-mode-reschedule-in-airflow-sensors-work

7.sensor的retry

sensor中可以指定重试的次数，比如

timeout=60 * 60,
retries=2,
retry_delay=timedelta(minutes=5),

retries=2表示总共会尝试运行3轮，超时之后运行下一轮，所以将会在airflow日志中看到

[2022-09-06 00:02:19,350] {taskinstance.py:1068} INFO - Starting attempt 1 of 3

[2022-09-06 01:09:58,153] {taskinstance.py:1068} INFO - Starting attempt 2 of 3

[2022-09-06 02:16:38,812] {taskinstance.py:1068} INFO - Starting attempt 3 of 3

timeout=60 * 60 表示每一轮尝试的超时时间会1个小时，但是在同一轮中会尝试多次，timeout默认值是60*60*24*7（1周）

retry_delay=timedelta(minutes=5) 表示超时后retry下一轮的间隔是5分钟，所以可以看到上面3轮重试的间隔是1小时5分钟

poke_interval 表示sensor check的间隔，默认值是60（1分钟）

参考：https://help.aliyun.com/document_detail/336856.html

8.出现报错：The scheduler does not appear to be running. Last heartbeat was received X minutes ago

查看/health接口之后scheduler的状态是unhealthy

文章介绍主要原因是配置中parsing_processes的数量过大，大于cpu的核数，将scheduler上报心跳的线程的cpu占用了导致

参考：记一次 Airflow 性能调优

9.airflow运行hive operator任务出现报错：airflow.exceptions.AirflowException: /usr/bin/env: ‘bash’: No such file or directory

这可能是由于使用的是supervisor来启动的airflow，所以需要额外的airflow中配置environment，如下

environment=PATH="/home/xxx/envs/your_env/bin:/usr/local/bin:/bin:/usr/bin:/usr/local/sbin:/usr/sbin"

参考：supervisor环境变量

10.airflow配置parsing_processes，parallelism，dag_concurrencyd参数的调整