Are these the blog posts? https://beam.apache.org/blog/2017/02/13/stateful-processing.html https://beam.apache.org/blog/2017/08/28/timely-processing.html
On Tue, Apr 9, 2019 at 3:41 PM Pablo Estrada <[email protected]> wrote: > soooounds good. Thanks guys <3 > > On Tue, Apr 9, 2019 at 3:19 PM Lukasz Cwik <[email protected]> wrote: > >> UnboundedSources and SplittableDoFns report watermarks which the runner >> uses to compute how much the watermark could advance if it processed some >> outstanding work. But it is always upto the runner to choose when the >> watermark advances. The runner could process each work item in watermark >> priority order and advance the watermark in small increments or could >> process many work items and then advance the watermark a lot. (Note that >> the BoundedSources API doesn't allow for reporting the watermark and it >> starts at Beam's concept of START OF TIME and advances in one step to >> Beam's concept of END OF TIME). >> >> You might be able to write what you want with an event based timer. Kenn >> wrote (2?) blog posts on state and timers that have some pretty good >> explanations and examples. >> >> On Tue, Apr 9, 2019 at 2:27 PM Pablo Estrada <[email protected]> wrote: >> >>> hi Luke, >>> thanks for the prompt reply: ) >>> >>> That makes sense. I think I'll go back to my cave to read a bunch about >>> streaming. : ) >>> >>> I was looking for this to try to write a sequence generator for Python >>> in streaming, and I was trying to debug what was going on. I was trying to >>> allow the DoFn to receive a watermark reported by the upstream source. (... >>> does that answer "which watermark?"... I am not sure that it does... but >>> maybe..). >>> >>> Do you think it's a reasonable use case for DoFns to know what the >>> upstream watermark is? >>> I hope that makes at least a some sense... : ) >>> >>> If it doesn't make sense, feel free to ignore, and I'll go do my >>> readings. >>> Thanks! >>> -P. >>> >>> On Tue, Apr 9, 2019 at 1:44 PM Lukasz Cwik <[email protected]> wrote: >>> >>>> WatermarkReporterParam is about reporting the watermark. The main >>>> usecase is for SplittableDoFns to be able to report the data watermark. >>>> >>>> The watermark is per input and output of a DoFn. Also each bundle being >>>> processed has its local watermarks while the runner computes the global >>>> watermark. The runners watermark could be per key, or key range or global >>>> across all keys. >>>> >>>> There is no runner agnostic way to read the watermark today. Is there a >>>> usecase you are targeting that would help from having access to the >>>> watermark (also, which watermark?)? >>>> >>>> >>>> On Tue, Apr 9, 2019 at 1:28 PM Pablo Estrada <[email protected]> >>>> wrote: >>>> >>>>> I am experimenting with state / timers in Python. As I look at the >>>>> DoFnProcessParams[1], I see that it's possible for a DoFn to receive >>>>> several arguments (e.g. Timers, Side Inputs, etc). Also the Watermark via >>>>> WatermarkReporterParam. >>>>> >>>>> I see that this parameter is not handled by runners when filling up >>>>> the arguments for a DoFn[2][3]. So, as far as I can tell, DoFns are not >>>>> currently able to get the watermark. >>>>> >>>>> Is this a bug, or is it intentional? Perhaps there's another way to >>>>> find out the watermark for a DoFn? >>>>> >>>>> Best >>>>> -P. >>>>> >>>>> [1] >>>>> https://github.com/apache/beam/blob/master/sdks/python/apache_beam/transforms/core.py#L381-L390 >>>>> >>>>> [2] >>>>> https://github.com/apache/beam/blob/master/sdks/python/apache_beam/runners/common.py#L477-L488 >>>>> [3] >>>>> https://github.com/apache/beam/blob/master/sdks/python/apache_beam/runners/common.py#L605-L620 >>>>> >>>>
