Github user aljoscha commented on the pull request:
https://github.com/apache/flink/pull/1562#issuecomment-190649095
Hi @wenlonglwl, the reason for this were mostly practical concerns. We want
to use our partitioned state abstraction for the window state because that
makes it easy to change the state backend depending on the use case. So, when
using a ReduceFunction this is actually put into a ReducingState that keeps an
aggregated value using that ReduceFunction. Having RichFunctions in inside
state would require a lot of overhead. First the function would have to be
copied for every different key because the function can keep state internally,
second, it would require wiring all the additional RichFunction calls (such as
open(), close(), the RuntimeContext) through to the state abstraction.
Could you maybe use the `WindowedStream.apply(ReduceFunction,
WindowFunction)` method, this allows you to incrementally aggregate the
elements in the window and then get the final result in the WindowFunction once
the window is emitted. This WindowFunction can be a RichFunction and can also
keep state.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
