Right, but I was asking about next_execution_date, not execution_date.
run time execution_date next_execution_date next_execution_date -
timedelta(days=1)
2017-01-02 02:00 2017-01-01 17:00 2017-01-02 02:00 2017-01-01 02:00
2017-01-02 05:00 2017-01-02 02:00 2017-01-02 05:00 2017-01-01 05:00
2017-01-02 08:00 2017-01-02 05:00 2017-01-02 08:00 2017-01-01 08:00
2017-01-02 11:00 2017-01-02 08:00 2017-01-02 11:00 2017-01-01 11:00
2017-01-02 14:00 2017-01-02 11:00 2017-01-02 14:00 2017-01-01 14:00
2017-01-02 17:00 2017-01-02 14:00 2017-01-02 17:00 2017-01-01 17:00
On Thu, Apr 27, 2017 at 4:04 PM, Alek Storm <alek.st...@gmail.com> wrote:
> It's `0 2-20/3 * * *`. The problem with using an offset from the execution
> date is that, for the very first run of the day, its execution date will
> *already* be the previous day (since it covers 8pm the previous day til 2am
> today) - subtracting from that would yield a date two days in the past,
> which is not what we want. Subsequent runs (at 5am, 8am, etc) have their
> execution date as the current day, so subtracting would work in those
> cases.
>
> Alek
>
> On Thu, Apr 27, 2017 at 2:55 PM, Paul Zaczkiewicz <paulz...@gmail.com>
> wrote:
>
> > First off, what schedule are you giving your DAG? I assume it's something
> > like "0 */4 * * *".
> >
> > Why don't you subtract 1 day from next_execution_date instead?
> >
> > On Thu, Apr 27, 2017 at 2:28 PM, Alek Storm <alek.st...@gmail.com>
> wrote:
> >
> > > Right, an easier way to do that would probably be to use the
> > > `prev_execution_date` macro, but either way, it would give the wrong
> > result
> > > on the first run of the day, since its start_date would already be the
> > > previous day.
> > >
> > > Alek
> > >
> > > On Thu, Apr 27, 2017 at 1:17 PM, Shah Altaf <mend...@gmail.com> wrote:
> > >
> > > > Hello. You can get access to the execution date of today and
> subtract a
> > > day
> > > > from it. You can use a PythonOperator with provide_context=True,
> that
> > > will
> > > > give your Python method a ds argument that has the execution date in
> > it.
> > > > Manipulate it and return an XCOM.
> > > >
> > > > For example, here's a filename being derived from execution date
> minus
> > > one
> > > > day:
> > > >
> > > > def get_filename_to_process(ds, **kwargs):
> > > > execdate=datetime.strptime(ds,"%Y-%m-%d")
> > > > my_file_path =
> > > > "dirname/{0}{1:02d}{2:02d}000000.csv".format(execdate.
> > > > year,execdate.month,execdate.day-1)
> > > > logging.info("my_file_path: ".format(my_file_path))
> > > > return my_file_path
> > > >
> > > > get_filename = PythonOperator(
> > > > task_id='the_filename',
> > > > python_callable=get_filename_to_process,
> > > > provide_context=True,
> > > > dag=dag)
> > > >
> > > >
> > > > Then, from other operators, you'd pull it in using some templating:
> > > >
> > > > {{ task_instance.xcom_pull(task_ids='the_filename') }}
> > > >
> > > > Hope that helps, sorry if I misunderstood your question.
> > > >
> > > >
> > > >
> > > > On Thu, Apr 27, 2017 at 6:28 PM Alek Storm <alek.st...@gmail.com>
> > wrote:
> > > >
> > > > > I'd like to create a job that runs six times a day, all with the
> same
> > > > > execution date (the previous day). Is there a way to do this
> without
> > > > > generating six separate DAGs, or running an `airflow clear` command
> > in
> > > > > between runs?
> > > > >
> > > > > Thanks,
> > > > > Alek
> > > > >
> > > >
> > >
> >
>
