You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@airflow.apache.org by "kasim (Jira)" <ji...@apache.org> on 2019/09/23 03:26:00 UTC
[jira] [Updated] (AIRFLOW-5538) Add a flag to make scheduling
trigger on start_date instead of execution_date (make execution_date equal
to start_date)
[ https://issues.apache.org/jira/browse/AIRFLOW-5538?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
kasim updated AIRFLOW-5538:
---------------------------
Description:
From [https://airflow.apache.org/scheduler.html] :
> Note that if you run a DAG on a schedule_interval of one day, the run
> stamped 2016-01-01 will be trigger soon after 2016-01-01T23:59. In
> other words, the job instance is started once the period it covers has
> ended.
This feature is very hurt .
For example I have etl job which run every day, schedule_interval is `0 1 * * *`, so it will trigger 2019-09-22 01:00:00 job on 2019-09-23 01:00:00 . But my etl is processing all data before start_date , means data range is between (history, 2019-09-23 00:00:00) , and we can't use `datetime.now()` because this is unable to reproduce. This force me add 1 day to execution_date:
```python
etl_end_time = "\{{ (execution_date + macros.timedelta(days=1)).strftime('%Y-%m-%d 00:00:00') }}"
```
However, when I need run a job with schedule_interval `45 2,3,4,5,6 * * *` , the `2019-09-22 06:45:00` job would run on `2019-09-23 02:45:00`, which is the day after execution_date . Instead of adding a day, I had to changed schedule_interval `45 2,3,4,5,6,7 * * *` and put a dummy operator on last run.
And in this situation , you don't need add one day to execution_date , this means you have to define two `etl_end_time` to represent a same date in jobs with different schedule_interval .
All these are very uncomfortable for me , adding a config or built-in method to make execution_date equal to start_date. would be very nice .
was:
From https://airflow.apache.org/scheduler.html :
> Note that if you run a DAG on a schedule_interval of one day, the run
> stamped 2016-01-01 will be trigger soon after 2016-01-01T23:59. In
> other words, the job instance is started once the period it covers has
> ended.
This feature is very hurt .
For example I have etl job which run every day, schedule_interval is `0 1 * * *`, so it will trigger 2019-09-22 01:00:00 job on 2019-09-23 01:00:00 . But my etl is processing all data before start_date , means data range is between (history, 2019-09-23 00:00:00) , and we can't use `datetime.now()` because this is unable to reproduce. This force me add 1 day to execution_date:
```python
etl_end_time = "\{{ (execution_date + macros.timedelta(days=1)).strftime('%Y-%m-%d 00:00:00') }}"
```
However, when I need run a job with schedule_interval `45 2,3,4,5,6 * * *` , the `2019-09-22 06:45:00` job would run on `2019-09-23 02:45:00`, which is the day after execution_date . Instead of adding a day, I had to changed schedule_interval `45 2,3,4,5,6,7 * * *` and put a dummy operator on last run.
And in this situation , you don't need add one day to execution_date , this means you have to define two `etl_end_time` to represent a same date in jobs with different schedule_interval .
All these are very uncomfortable for me , is there any config or built-in method to make execution_date equal to start_date ? Or I have to modify airflow source code ...
> Add a flag to make scheduling trigger on start_date instead of execution_date (make execution_date equal to start_date)
> -----------------------------------------------------------------------------------------------------------------------
>
> Key: AIRFLOW-5538
> URL: https://issues.apache.org/jira/browse/AIRFLOW-5538
> Project: Apache Airflow
> Issue Type: Improvement
> Components: DagRun
> Affects Versions: 1.10.5
> Reporter: kasim
> Priority: Major
>
> From [https://airflow.apache.org/scheduler.html] :
> > Note that if you run a DAG on a schedule_interval of one day, the run
> > stamped 2016-01-01 will be trigger soon after 2016-01-01T23:59. In
> > other words, the job instance is started once the period it covers has
> > ended.
> This feature is very hurt .
> For example I have etl job which run every day, schedule_interval is `0 1 * * *`, so it will trigger 2019-09-22 01:00:00 job on 2019-09-23 01:00:00 . But my etl is processing all data before start_date , means data range is between (history, 2019-09-23 00:00:00) , and we can't use `datetime.now()` because this is unable to reproduce. This force me add 1 day to execution_date:
> ```python
> etl_end_time = "\{{ (execution_date + macros.timedelta(days=1)).strftime('%Y-%m-%d 00:00:00') }}"
> ```
> However, when I need run a job with schedule_interval `45 2,3,4,5,6 * * *` , the `2019-09-22 06:45:00` job would run on `2019-09-23 02:45:00`, which is the day after execution_date . Instead of adding a day, I had to changed schedule_interval `45 2,3,4,5,6,7 * * *` and put a dummy operator on last run.
> And in this situation , you don't need add one day to execution_date , this means you have to define two `etl_end_time` to represent a same date in jobs with different schedule_interval .
> All these are very uncomfortable for me , adding a config or built-in method to make execution_date equal to start_date. would be very nice .
--
This message was sent by Atlassian Jira
(v8.3.4#803005)