It ' a nice feature to be included in the grid compute module , in my opinion . But I have a question : what difference we have with ComputeTask class in behavioral terms ?
https://apacheignite.readme.io/docs/compute-tasks 2015-09-23 16:45 GMT+02:00 Lalit Kumar Jha <lalitj....@gmail.com>: > Summarizing this: > > *Goal*: All data flow/stream in system adhere to “Reactive Streams” API, > includes data ingestion, async computation, compute result/stream > aggregation. > > *Benefits*: > > > - Common interface for all data transfer/exchange/processing. > - Streaming data ingestion can compose complex stream processing in a > standard (interoperable dropin implementation replacement) fashion, like > filter or aggregate streaming data, merge multiple stream etc. > > > - Compose multiple async compute jobs results as streams. > > > *Design*: To overcome “Cons of Exposing Reactive Stream APIs instead of > RxJava (or other implementation)” mentioned at > https://github.com/ReactiveX/RxJava/wiki/Reactive-Streams > > #1. Follow recommended approach there. > > > #2. Rather than having streaming (data transfer) stream processing logic > (filter, merge, flatMap) in Consumer, treat them as a separate concern, and > wrap Publisher/Consumer in concrete classes. > > This is discouraged by RxJava, but will keep standard and non standard > implementation separate and will allow using multiple libraries for > different use-cases. Rather than having this behaviour tightly coupled in > Consumer. > > > > On Wed, Sep 23, 2015 at 6:29 PM, Raul Kripalani <ra...@apache.org> wrote: > > > Yeah, in my original post I already pointed out that Reactive Streams is > a > > proposed standard but not an implementation. > > > > My proposal was to go with RxJava 1.0 + the Reactive Streams adapter ( > > https://github.com/ReactiveX/RxJavaReactiveStreams) for now, as that > would > > allow users to benefit from things like Hystrix whose base is RxJava 1.0, > > while keeping their code future-proof for when RxJava 2.0 is released > with > > a native RS interface. > > > > Thanks for chiming in. > > > > Raúl. > > > > On Wed, Sep 23, 2015 at 1:35 PM, akarnokd <akarn...@gmail.com> wrote: > > > > > I'd definitely go on a Reactive-Streams path (although I don't know > what > > > Apache Ignite does), however, there seems to be a small confusion > around > > > RS. > > > RS is a set of 4 interfaces and practically unusable on its own. One > > needs > > > a > > > library built around it such as Project Reactor, Akka Streams and > RxJava > > > 2.0. Such library, likely, will also slip into the API itself due to > its > > > usefulness. > > > > > > For example, the following API is inconvenient to use but is definitely > > > cutting-edge: > > > > > > Publisher<Data> getData(); > > > > > > but you can only subscribe to it in its plain form or wrap it with one > of > > > the libraries: > > > > > > Observable.fromPublisher(getData()).map().filter().group()... > > > > > > I'm not sure about the others, but RxJava 2.0 Observable (which later > may > > > be > > > renamed to Flowable) implement Publisher directly, therefore, the > > following > > > getData() may return an RxJava Observable and still be consumable via > > > Reactor or Akka (after wrapping/fluent-conversion, of course): > > > > > > getData().to(Streams::from).dispatchOn(...) > > > > > > Although I'm pretty confident that RxJava 2.0 works right now, it is > > still > > > considered alpha and based on past experience with its gatekeepers, it > > may > > > take 6-12 months until an official release comes out. > > > > > > > > > > > > -- > > > View this message in context: > > > > > > http://apache-ignite-developers.2346864.n4.nabble.com/Brainstorming-about-Reactive-Streams-tp3346p3422.html > > > Sent from the Apache Ignite Developers mailing list archive at > > Nabble.com. > > > > > >
