Yeah, as a maintainer it is pain to have to maintain both sync and async
version of a hook for a Sensor in providers
On Wed, 13 Nov 2024 at 20:42, Daniel Standish
<daniel.stand...@astronomer.io.invalid> wrote:
> Yeah everything is related. I'm just saying for the purpose of picking a
> path forward i'm just trying to clarify the options.
>
> If we can rule out any of the options I presented we'll have made some
> progress.
>
> Do you have an opinion on any of those?
>
> I would probably lean towards keep reschedule interface, keep it on base
> sensor, remove it from providers.
>
>
>
> On Wed, Nov 13, 2024 at 12:37 PM Kaxil Naik <kaxiln...@gmail.com> wrote:
>
> > > I think we need to separate what to do about the sensors we have in
> > > providers from the core interface, essentially.
> >
> >
> > I think they are interconnected, especially if one of the goals is to
> have
> > the "most performant" option (deferral) as the default or the only option
> > for the users, regardless of the internal.
> >
> > From the user's POV, they can care about performance and visibility
> > ("rescheduled"/"deferral" as TI states).
> >
> > The secondary issue is what you are mentioning about the implementation
> of
> > "reschedule" method and what to do when reschedule happens, IIRC none of
> > the Operators in the "official" providers do anything special then
> > inheriting BaseSensorOperator.
> >
> > 1. remove the rescheduling interface in core. No more "up for
> reschedule"
> > > state etc. No more reschedule mode in base sensor. No more reschedule
> > mode
> > > in any derivative sensor.
> > > 2. keep rescheduling in core. remove from base sensor and derivatives.
> > > 3. keep reschedule in core. keep in base sensor. remove from all
> > > derivatives. ban the reintroduction of reschedule mode in derivative
> > > sensors.
> >
> >
> >
> >
> > On Wed, 13 Nov 2024 at 19:46, Ash Berlin-Taylor <a...@apache.org> wrote:
> >
> > > > Reschedulability in airflow (making up a word here) to my knowledge
> > only
> > > exists to support reschedule mode in sensors.
> > >
> > > Yes, this is true.
> > >
> > >
> > > On 13 November 2024 19:32:23 GMT, Daniel Standish
> > > <daniel.stand...@astronomer.io.INVALID> wrote:
> > > >The options are a bit muddled here.
> > > >
> > > >I think we need to separate what to do about the sensors we have in
> > > >providers from the core interface, essentially.
> > > >
> > > >Reschedulability in airflow (making up a word here) to my knowledge
> only
> > > >exists to support reschedule mode in sensors.
> > > >
> > > >Let's also separate poke from reschedule.
> > > >
> > > >For now, let's just look at reschedule.
> > > >
> > > >Let me enumerate some paths as i see them.
> > > >
> > > >1. remove the rescheduling interface in core. No more "up for
> > reschedule"
> > > >state etc. No more reschedule mode in base sensor. No more reschedule
> > > mode
> > > >in any derivative sensor.
> > > >2. keep rescheduling in core. remove from base sensor and derivatives.
> > > >3. keep reschedule in core. keep in base sensor. remove from all
> > > >derivatives. ban the reintroduction of reschedule mode in derivative
> > > >sensors.
> > >
> >
>
