shahar1 commented on PR #68749: URL: https://github.com/apache/airflow/pull/68749#issuecomment-4782880256
> > re-running for a 3-month-old event would just read today's data anyway > > This entirely depends on how the task actually reads data, doesn’t it? If it just reads the current data, the same argument can be made against scheduled catchups too. > > For catchup to work, you need to implement the task in a way that makes catchup possible in the first place. With scheduled dags, this means it should use the data interval or logical date. With assets, it should use the asset event date, which would be in the past. I don’t think this argument holds. Gotcha, it makes sense - I've probably overstated. Assuming that it's a valid case and we gate this behavior with a flag: 1. Would we rather use the existing `catchup` flag or create a new one? I tend towards a new flag (with a different name, like `consume_asset_history`), because of the different semantics and retain compatibility with existing behavior (see next). 2. What should be the default behavior, now and in the future? Considering that it's a technically valid use case, I tend towards defaulting it to `True` to avoid breaking changes (with a future warning when unset), and flip it in Airflow 3.4+ / 4.0+ for parity with the `catchup`'s default. WDYT? I'm fine with putting the above for a discussion and lazy consensus in the dev list. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
