Np. Can you please try the patch and +1 it? I would like to get it into 1.8.
Thx Bolke. > On 14 Jan 2017, at 19:04, Sumit Maheshwari <[email protected]> wrote: > > Thanks a lot Bolke for looking into this issue and raising PR in such a short > time. > > > > On Sat, Jan 14, 2017 at 8:52 PM, Bolke de Bruin <[email protected] > <mailto:[email protected]>> wrote: > PR is out. Should fix your issue and fix a bug that was introduced with the > catchup feature. > > https://github.com/apache/incubator-airflow/pull/1994 > <https://github.com/apache/incubator-airflow/pull/1994> > <https://github.com/apache/incubator-airflow/pull/1994 > <https://github.com/apache/incubator-airflow/pull/1994>> > > Bolke > > > On 14 Jan 2017, at 14:22, Bolke de Bruin <[email protected] > > <mailto:[email protected]>> wrote: > > > > You will need to align your start_date with the cron interval. In this case > > your start date should include 30 minutes. This is expected behaviour in > > 1.7.0. and 1.7.1. In master we are auto aligning the start_date with the > > interval in the scheduler. However, the dependency checker doesn’t do this > > so there it is an issue. I have create AIRFLOW-759 to track this issue. > > > > Bolke. > > > > > >> On 14 Jan 2017, at 12:43, Sumit Maheshwari <[email protected] > >> <mailto:[email protected]>> wrote: > >> > >> Found that on our current prod version (1.7.0) and master as well, > >> scheduler doesn't fire TIs if *scheduler_interval* is a cron expression and > >> *depends_on_past* is True. > >> > >> The very simple DAG I used for testing can be found here: > >> http://pastebin.com/wPnFjRwD <http://pastebin.com/wPnFjRwD> > >> > >> However it creates new DAG runs as per the scheduler, see this: > >> https://drive.google.com/file/d/0B9EjxDCEDhERLWNRT196OWgtcnM/view?usp=sharing > >> > >> <https://drive.google.com/file/d/0B9EjxDCEDhERLWNRT196OWgtcnM/view?usp=sharing> > >> > >> Scheduler logs for the DAG can be seen here: http://pastebin.com/z5aRxh76 > >> <http://pastebin.com/z5aRxh76> > >> > >> Please let me know if this is a know issue, or there are any workaround. > >> What we do for workaround is to set *depends_on_past* to *False* at first > >> then set it to *True* later. > >> > >> Thanks, > >> Sumit > > > >
