[ 
https://issues.apache.org/jira/browse/AIRFLOW-2279?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16748143#comment-16748143
 ] 

Gabriel Silk commented on AIRFLOW-2279:
---------------------------------------

At Dropbox, we have the exact same need. We were going to embark on building 
this, but perhaps it would make sense to use your patch as a starting point?

One hard requirement we have is the ability to limit the number of tasks 
cleared when doing a cross-DAG clear. At Dropbox, a lot of the tasks we run are 
*very* data- and compute- intensive, so if we (for example) cleared 10,000 
tasks accidentally, it would be difficult to recover from.

Another aspect of this is the time window of the ExternalTaskSensor – for 
example if I have a task that runs at time _t_ in dag A and depends on the set 
of tasks in another dag B between [_t - 7, t_), then clearing the task at time 
_t - 4_ in dag B should also clear the task at time _t_ in dag A. Does your 
patch do this already?

> Clearing Tasks Across DAGs
> --------------------------
>
>                 Key: AIRFLOW-2279
>                 URL: https://issues.apache.org/jira/browse/AIRFLOW-2279
>             Project: Apache Airflow
>          Issue Type: Improvement
>            Reporter: Achal Soni
>            Priority: Major
>         Attachments: cross_dag_ui_screenshot.png
>
>
> At Stripe, we commonly have discrete dags that depend on each other by 
> leveraging ExternalTaskSensors. We also find ourselves routinely wanting to 
> not only clear tasks and their downstream tasks in a particular dag, but also 
> their downstream tasks in their dependent dags (linked by 
> ExternalTaskSensors). 
> We currently have extended Airflow to handle this by modifying the webapp and 
> cli tool to optionally clear dependent tasks across multiple dags (see 
> attached screenshot). 
> We want to open the floor for discussion with the larger Airflow community 
> about the usage of ExternalTaskSensors and specifically how to handle 
> clearing across dags. We are interested in learning more about the accepted 
> practices in this regard, and are very open/willing to contribute in this 
> area if there is interest!



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Reply via email to