Hi Kenn, Thanks for the explanation. Yes, I will test the runner directly.
Shen On Wed, Jun 7, 2017 at 12:35 PM, Kenneth Knowles <k...@google.com.invalid> wrote: > Hi Shen, > > There are four points (at least) to why we don't currently have custom > triggers: > > 1. The existing trigger state machine implementations use low-level and > difficult APIs that are not available to users. > 2. If we make triggers a UDF then we need to cross the Fn API. It makes a > lot more portability work in terms of engineering and also is really the > center of a tight loop. Instead, runners can very efficiently execute them > directly. > 3. There are many state machines that are not valid triggers. > 4. Some triggers can be more efficiently executed directly from their > syntax than from a state machine. > > Eventually, we may add some kind of triggering UDF. But we may not, because > if you want something like a custom trigger for some use case, then you can > do it with state & timers in ParDo. > > But it sounds like you need it for some quite specific testing. Since > "watermark advance event" is a runner-specific concept not part of the > model (a runner doesn't have to actually reify such events) maybe you just > want to test your runner details directly. > > Kenn > > On Wed, Jun 7, 2017 at 10:22 AM, Shen Li <cs.she...@gmail.com> wrote: > > > Hi Lukasz, > > > > Thanks again for the suggestion. Is there any reason for not allowing > users > > create custom triggers? > > > > Shen > > > > On Wed, Jun 7, 2017 at 12:13 PM, Lukasz Cwik <lc...@google.com.invalid> > > wrote: > > > > > You should really take a look at TestStream and have runners integrate > > with > > > it instead. > > > > > > There are already several tests which validate TestStream compatible > > > runners to make sure their trigger evaluations are correct. > > > > > > On Wed, Jun 7, 2017 at 10:10 AM, Shen Li <cs.she...@gmail.com> wrote: > > > > > > > Hi Lukasz, > > > > > > > > Thanks for the suggestion. I am trying to test how the runner > generates > > > > watermarks. So I would like to have the trigger to fire on every > > > watermark > > > > advancing event. > > > > > > > > Shen > > > > > > > > On Wed, Jun 7, 2017 at 10:49 AM, Lukasz Cwik > <lc...@google.com.invalid > > > > > > > wrote: > > > > > > > > > Look into the AfterPane#elementCountAtLeast trigger as it seems to > be > > > the > > > > > closest to your description. It fires as soon as any data is > > available. > > > > > > > > > > Are you sure you don't want some kind of watermark based trigger > with > > > > just > > > > > with a small interval size? > > > > > > > > > > > > > > > > > > > > > > > > > On Wed, Jun 7, 2017 at 8:22 AM, Shen Li <cs.she...@gmail.com> > wrote: > > > > > > > > > > > Hi Lukasz, > > > > > > > > > > > > Thanks for your response. Is it possible to implement the > following > > > > logic > > > > > > using existing triggers: always fire the trigger on a > GlobalWindow > > > > > whenever > > > > > > watermark advances? > > > > > > > > > > > > Shen > > > > > > > > > > > > On Wed, Jun 7, 2017 at 10:05 AM, Lukasz Cwik > > > <lc...@google.com.invalid > > > > > > > > > > > wrote: > > > > > > > > > > > > > Users are unable to create custom trigger implementations. If > you > > > > tell > > > > > us > > > > > > > what you want your trigger to do, we may be able to suggest an > > > > > > alternative. > > > > > > > > > > > > > > On Wed, Jun 7, 2017 at 7:41 AM, Shen Li <cs.she...@gmail.com> > > > wrote: > > > > > > > > > > > > > > > Hi, > > > > > > > > > > > > > > > > I created a custom trigger class (XYZ) by extending the > > > > OnceTrigger. > > > > > > > During > > > > > > > > execution, I got this error "Cannot translate trigger class > XYZ > > > to > > > > a > > > > > > > > runner-API proto." It seems that the Triggers.ProtoConverter > > > class > > > > > > needs > > > > > > > to > > > > > > > > declare a convertSpecific method for my trigger XYZ. How can > I > > > use > > > > my > > > > > > > > custom trigger without modifying Beam's Triggers class? > > > > > > > > > > > > > > > > Thanks, > > > > > > > > > > > > > > > > Shen > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > >
