Re: [DISCUSSION] IgniteFuture class future in Ignite-3.0.
Hi Andrey, I agree that public API based on CompletableFuture might be very useful. The thing I worry about is that with CompletableFuture used internally we can miss some important capabilities (e.g. mentioned cancellation). ReactiveStreams specification on the other hand defines a lot of interesting and important things and it can serve as guidance for Ignite developers. So, my vision here is that it is very important to have well-defined semantics internally. Introducing public API based on CompletableFuture is fine as well because enough users can be quite happy with it. 2021-04-05 14:49 GMT+03:00, Andrey Mashenkov : > Hi Ivan, > > JDK flow API looks interesting. I think we should use it abroad. > Though, Java 14 is not LTS (long-term support) release. I guess many users > will prefer Java 15, > but actually, we have no agreement about switching to Java 15 which may > require some additional efforts. > For now, we could import the required classes into Ignite module as we've > done with JSR-166 (concurrent collections). > > I think we should use Flow-like API for Queries, Cursors, Compute results > and even Transactions. > E.g. reactive transaction API is must have and it can be looked like: > Transaction.compose(tx -> table1.getAsync(a)).thenApply(tx -> > table2.putAsync()).thenApply(tx -> tx.commit()); > > Agree, CompletableFuture looks totally unusable for cases that supposed > cancellation capabilities, due to broken cancellation semantic [1]. > However, cancellation support for simple table operations doesn't make any > sense and we still can use CompletableFuture for e.g. table operations > unless we found it unusable for Transaction API that is not designed yet. > > I've looked at reactive API as CompletableFuture replacement for simple > cases. > At first glance, it is very complex and may require too much effort on our > user side, and/or hard to understand for the user. > > Maybe it make sense to create an IEP for future replacement with all the > examples for the table, transaction, compute APIs? > WDYT? > > [1] > https://stackoverflow.com/questions/43389894/recursively-cancel-an-allof-completablefuture > > On Sat, Apr 3, 2021 at 1:40 AM Ivan Pavlukhin wrote: > >> Andrey, >> >> As you might know Java has it is own Reactive abstractions since 9 >> [1]. Moreover, an original Reactive Streams library has a bridge with >> Java [2]. >> >> Personally I do not love CompletableFuture because it's API seems >> questionable to me in various aspects, e.g. mentioned cancellation. >> Currently I think that the cleanest way is custom abstractions >> (possibly implementing CompletionStage) at first. And providing >> bridges to other APIs (including CompletableFuture) as a next step. >> Various API integrations can come in form of extensions. >> >> > JDK classes are well-known and reactive frameworks usually have >> converters >> to/from CompletableFuture [1] [2]. >> >> Here I have doubts that it is feasible to achieve it fairly. E.g. once >> again cancellation. While reactive API supports cancellation it will >> not be possible to cancel computation for anything built from >> CompletableFuture. >> >> [1] >> https://docs.oracle.com/en/java/javase/14/docs/api/java.base/java/util/concurrent/Flow.Publisher.html >> [2] >> https://www.reactive-streams.org/reactive-streams-1.0.3-javadoc/org/reactivestreams/FlowAdapters.html >> >> 2021-04-02 12:00 GMT+03:00, Andrey Mashenkov >> : >> > Ivan, >> > thanks for the link. I think, now I've got what you mean. >> > >> > I don't think we want to have any framework as a dependency and rely on >> > their lifecycle, roadmaps goals and >> > bother about compatibility. >> > JDK classes are well-known and reactive frameworks usually have >> converters >> > to/from CompletableFuture [1] [2]. >> > So, users are free to choose any reactive framework. >> > >> > I think will need reactive abstractions in near future for Queries API >> and >> > maybe Transaction API design. >> > These projects are good enough where we can get inspiration. >> > >> > [1] >> > >> https://projectreactor.io/docs/core/release/api/reactor/core/publisher/Mono.html#fromFuture-java.util.concurrent.CompletableFuture- >> > [2] >> > >> https://github.com/ReactiveX/RxJava/blob/3.x/src/main/java/io/reactivex/rxjava3/internal/jdk8/CompletableFromCompletionStage.java >> > >> > On Thu, Apr 1, 2021 at 1:29 PM Ivan Pavlukhin >> wrote: >> > >> >> Andrey, >> >> >> >> > Reactive abstractions look more suitable for Queries rather than >> >> > cache/table async operations and don't offer the power of chaining >> like >> >> > CompletableStage. >> >> >> >> Could you please elaborate what capabilities do you mean? AFAIK there >> >> are quite powerful APIs for singular reactive abstractions as well. >> >> E.g. [1]. >> >> >> >> [1] >> >> >> https://projectreactor.io/docs/core/release/api/reactor/core/publisher/Mono.html >> >> >> >> 2021-04-01 12:33 GMT+03:00, Andrey Mashenkov >> >> : >> >> > Val, >> >> > I just complain JDK
Re: [DISCUSSION] IgniteFuture class future in Ignite-3.0.
Hi Ivan, JDK flow API looks interesting. I think we should use it abroad. Though, Java 14 is not LTS (long-term support) release. I guess many users will prefer Java 15, but actually, we have no agreement about switching to Java 15 which may require some additional efforts. For now, we could import the required classes into Ignite module as we've done with JSR-166 (concurrent collections). I think we should use Flow-like API for Queries, Cursors, Compute results and even Transactions. E.g. reactive transaction API is must have and it can be looked like: Transaction.compose(tx -> table1.getAsync(a)).thenApply(tx -> table2.putAsync()).thenApply(tx -> tx.commit()); Agree, CompletableFuture looks totally unusable for cases that supposed cancellation capabilities, due to broken cancellation semantic [1]. However, cancellation support for simple table operations doesn't make any sense and we still can use CompletableFuture for e.g. table operations unless we found it unusable for Transaction API that is not designed yet. I've looked at reactive API as CompletableFuture replacement for simple cases. At first glance, it is very complex and may require too much effort on our user side, and/or hard to understand for the user. Maybe it make sense to create an IEP for future replacement with all the examples for the table, transaction, compute APIs? WDYT? [1] https://stackoverflow.com/questions/43389894/recursively-cancel-an-allof-completablefuture On Sat, Apr 3, 2021 at 1:40 AM Ivan Pavlukhin wrote: > Andrey, > > As you might know Java has it is own Reactive abstractions since 9 > [1]. Moreover, an original Reactive Streams library has a bridge with > Java [2]. > > Personally I do not love CompletableFuture because it's API seems > questionable to me in various aspects, e.g. mentioned cancellation. > Currently I think that the cleanest way is custom abstractions > (possibly implementing CompletionStage) at first. And providing > bridges to other APIs (including CompletableFuture) as a next step. > Various API integrations can come in form of extensions. > > > JDK classes are well-known and reactive frameworks usually have > converters > to/from CompletableFuture [1] [2]. > > Here I have doubts that it is feasible to achieve it fairly. E.g. once > again cancellation. While reactive API supports cancellation it will > not be possible to cancel computation for anything built from > CompletableFuture. > > [1] > https://docs.oracle.com/en/java/javase/14/docs/api/java.base/java/util/concurrent/Flow.Publisher.html > [2] > https://www.reactive-streams.org/reactive-streams-1.0.3-javadoc/org/reactivestreams/FlowAdapters.html > > 2021-04-02 12:00 GMT+03:00, Andrey Mashenkov : > > Ivan, > > thanks for the link. I think, now I've got what you mean. > > > > I don't think we want to have any framework as a dependency and rely on > > their lifecycle, roadmaps goals and > > bother about compatibility. > > JDK classes are well-known and reactive frameworks usually have > converters > > to/from CompletableFuture [1] [2]. > > So, users are free to choose any reactive framework. > > > > I think will need reactive abstractions in near future for Queries API > and > > maybe Transaction API design. > > These projects are good enough where we can get inspiration. > > > > [1] > > > https://projectreactor.io/docs/core/release/api/reactor/core/publisher/Mono.html#fromFuture-java.util.concurrent.CompletableFuture- > > [2] > > > https://github.com/ReactiveX/RxJava/blob/3.x/src/main/java/io/reactivex/rxjava3/internal/jdk8/CompletableFromCompletionStage.java > > > > On Thu, Apr 1, 2021 at 1:29 PM Ivan Pavlukhin > wrote: > > > >> Andrey, > >> > >> > Reactive abstractions look more suitable for Queries rather than > >> > cache/table async operations and don't offer the power of chaining > like > >> > CompletableStage. > >> > >> Could you please elaborate what capabilities do you mean? AFAIK there > >> are quite powerful APIs for singular reactive abstractions as well. > >> E.g. [1]. > >> > >> [1] > >> > https://projectreactor.io/docs/core/release/api/reactor/core/publisher/Mono.html > >> > >> 2021-04-01 12:33 GMT+03:00, Andrey Mashenkov > >> : > >> > Val, > >> > I just complain JDK CompletableFuture itself is not ideal, but already > >> > implemented, well-known and tested. > >> > It still a good alternative compared to custom future implementation > to > >> me. > >> > > >> > Ok, I feel most of us agree with CompletableFuture as a replacement > for > >> > custom IgniteFuture. > >> > Let's try to use it, but keep in mind that we MUST return a defective > >> copy > >> > (via copy() or minimalStage()) to user to prevent misusage on the user > >> > side. > >> > It would be nice if we'll follow the same requirement in our internal > >> code, > >> > e.g. if a component returns a future that further can be used in other > >> > components, especially in custom plugins. > >> > > >> > Ivan, thanks for the example. > >> > Reactive abstractions look
Re: [DISCUSSION] IgniteFuture class future in Ignite-3.0.
Andrey, As you might know Java has it is own Reactive abstractions since 9 [1]. Moreover, an original Reactive Streams library has a bridge with Java [2]. Personally I do not love CompletableFuture because it's API seems questionable to me in various aspects, e.g. mentioned cancellation. Currently I think that the cleanest way is custom abstractions (possibly implementing CompletionStage) at first. And providing bridges to other APIs (including CompletableFuture) as a next step. Various API integrations can come in form of extensions. > JDK classes are well-known and reactive frameworks usually have converters to/from CompletableFuture [1] [2]. Here I have doubts that it is feasible to achieve it fairly. E.g. once again cancellation. While reactive API supports cancellation it will not be possible to cancel computation for anything built from CompletableFuture. [1] https://docs.oracle.com/en/java/javase/14/docs/api/java.base/java/util/concurrent/Flow.Publisher.html [2] https://www.reactive-streams.org/reactive-streams-1.0.3-javadoc/org/reactivestreams/FlowAdapters.html 2021-04-02 12:00 GMT+03:00, Andrey Mashenkov : > Ivan, > thanks for the link. I think, now I've got what you mean. > > I don't think we want to have any framework as a dependency and rely on > their lifecycle, roadmaps goals and > bother about compatibility. > JDK classes are well-known and reactive frameworks usually have converters > to/from CompletableFuture [1] [2]. > So, users are free to choose any reactive framework. > > I think will need reactive abstractions in near future for Queries API and > maybe Transaction API design. > These projects are good enough where we can get inspiration. > > [1] > https://projectreactor.io/docs/core/release/api/reactor/core/publisher/Mono.html#fromFuture-java.util.concurrent.CompletableFuture- > [2] > https://github.com/ReactiveX/RxJava/blob/3.x/src/main/java/io/reactivex/rxjava3/internal/jdk8/CompletableFromCompletionStage.java > > On Thu, Apr 1, 2021 at 1:29 PM Ivan Pavlukhin wrote: > >> Andrey, >> >> > Reactive abstractions look more suitable for Queries rather than >> > cache/table async operations and don't offer the power of chaining like >> > CompletableStage. >> >> Could you please elaborate what capabilities do you mean? AFAIK there >> are quite powerful APIs for singular reactive abstractions as well. >> E.g. [1]. >> >> [1] >> https://projectreactor.io/docs/core/release/api/reactor/core/publisher/Mono.html >> >> 2021-04-01 12:33 GMT+03:00, Andrey Mashenkov >> : >> > Val, >> > I just complain JDK CompletableFuture itself is not ideal, but already >> > implemented, well-known and tested. >> > It still a good alternative compared to custom future implementation to >> me. >> > >> > Ok, I feel most of us agree with CompletableFuture as a replacement for >> > custom IgniteFuture. >> > Let's try to use it, but keep in mind that we MUST return a defective >> copy >> > (via copy() or minimalStage()) to user to prevent misusage on the user >> > side. >> > It would be nice if we'll follow the same requirement in our internal >> code, >> > e.g. if a component returns a future that further can be used in other >> > components, especially in custom plugins. >> > >> > Ivan, thanks for the example. >> > Reactive abstractions look more suitable for Queries rather than >> > cache/table async operations and don't offer the power of chaining like >> > CompletableStage. >> > AFAIK, guys who involved in SQL engine development based on Calcite >> already >> > use reactive patterns. >> > >> > >> > On Thu, Apr 1, 2021 at 3:15 AM Ivan Pavlukhin >> wrote: >> > >> >> Folks, >> >> >> >> Regarding Reactive abstractions. While it might look too complex for >> >> simple KV cases it can be quite powerful for queries. Also there are >> >> known solutions for cancellation, backpressure and flow control. It >> >> can greatly simplify things for users familiar with Reactive >> >> programming rather than learning Ignite-specific Query/QueryCursor >> >> API. >> >> >> >> Also it looks like there are well-defined semantics [1]. E.g. >> >> cancellation seems to be defined much better than for >> >> CompletableFuture. >> >> >> >> [1] >> >> https://github.com/reactive-streams/reactive-streams-jvm#specification >> >> >> >> 2021-03-31 21:30 GMT+03:00, Valentin Kulichenko < >> >> valentin.kuliche...@gmail.com>: >> >> > Hi Andrey, >> >> > >> >> > Please see below. >> >> > >> >> > -Val >> >> > >> >> > On Wed, Mar 31, 2021 at 7:55 AM Andrey Mashenkov >> >> > >> >> > wrote: >> >> > >> >> >> CompletableFuture cancellation will not work as many users >> >> >> expected. >> >> >> Javadoc says: >> >> >> /* Since (unlike {@link FutureTask}) this class has no direct >> >> >> * control over the computation that causes it to be completed, >> >> >> * cancellation is treated as just another form of exceptional >> >> >> * completion. >> >> >> */ >> >> >> >> >> > >> >> > Let's not make assumptions about the expectations of the users. >> >> >
Re: [DISCUSSION] IgniteFuture class future in Ignite-3.0.
Ivan, thanks for the link. I think, now I've got what you mean. I don't think we want to have any framework as a dependency and rely on their lifecycle, roadmaps goals and bother about compatibility. JDK classes are well-known and reactive frameworks usually have converters to/from CompletableFuture [1] [2]. So, users are free to choose any reactive framework. I think will need reactive abstractions in near future for Queries API and maybe Transaction API design. These projects are good enough where we can get inspiration. [1] https://projectreactor.io/docs/core/release/api/reactor/core/publisher/Mono.html#fromFuture-java.util.concurrent.CompletableFuture- [2] https://github.com/ReactiveX/RxJava/blob/3.x/src/main/java/io/reactivex/rxjava3/internal/jdk8/CompletableFromCompletionStage.java On Thu, Apr 1, 2021 at 1:29 PM Ivan Pavlukhin wrote: > Andrey, > > > Reactive abstractions look more suitable for Queries rather than > > cache/table async operations and don't offer the power of chaining like > > CompletableStage. > > Could you please elaborate what capabilities do you mean? AFAIK there > are quite powerful APIs for singular reactive abstractions as well. > E.g. [1]. > > [1] > https://projectreactor.io/docs/core/release/api/reactor/core/publisher/Mono.html > > 2021-04-01 12:33 GMT+03:00, Andrey Mashenkov : > > Val, > > I just complain JDK CompletableFuture itself is not ideal, but already > > implemented, well-known and tested. > > It still a good alternative compared to custom future implementation to > me. > > > > Ok, I feel most of us agree with CompletableFuture as a replacement for > > custom IgniteFuture. > > Let's try to use it, but keep in mind that we MUST return a defective > copy > > (via copy() or minimalStage()) to user to prevent misusage on the user > > side. > > It would be nice if we'll follow the same requirement in our internal > code, > > e.g. if a component returns a future that further can be used in other > > components, especially in custom plugins. > > > > Ivan, thanks for the example. > > Reactive abstractions look more suitable for Queries rather than > > cache/table async operations and don't offer the power of chaining like > > CompletableStage. > > AFAIK, guys who involved in SQL engine development based on Calcite > already > > use reactive patterns. > > > > > > On Thu, Apr 1, 2021 at 3:15 AM Ivan Pavlukhin > wrote: > > > >> Folks, > >> > >> Regarding Reactive abstractions. While it might look too complex for > >> simple KV cases it can be quite powerful for queries. Also there are > >> known solutions for cancellation, backpressure and flow control. It > >> can greatly simplify things for users familiar with Reactive > >> programming rather than learning Ignite-specific Query/QueryCursor > >> API. > >> > >> Also it looks like there are well-defined semantics [1]. E.g. > >> cancellation seems to be defined much better than for > >> CompletableFuture. > >> > >> [1] > >> https://github.com/reactive-streams/reactive-streams-jvm#specification > >> > >> 2021-03-31 21:30 GMT+03:00, Valentin Kulichenko < > >> valentin.kuliche...@gmail.com>: > >> > Hi Andrey, > >> > > >> > Please see below. > >> > > >> > -Val > >> > > >> > On Wed, Mar 31, 2021 at 7:55 AM Andrey Mashenkov > >> > > >> > wrote: > >> > > >> >> CompletableFuture cancellation will not work as many users expected. > >> >> Javadoc says: > >> >> /* Since (unlike {@link FutureTask}) this class has no direct > >> >> * control over the computation that causes it to be completed, > >> >> * cancellation is treated as just another form of exceptional > >> >> * completion. > >> >> */ > >> >> > >> > > >> > Let's not make assumptions about the expectations of the users. That's > >> > exactly why I initially leaned towards a custom interface as well, but > >> > that's a mistake. > >> > Indeed, this contract might look weird to us, because the current > >> > version > >> > of Ignite behaves differently. However, there are much more developers > >> > using CompletableFuture and other standard async frameworks, than > >> > developers using the async functionality of Ignite. Therefore, our > >> > intuitions can easily be wrong. > >> > > >> > > >> >> Completion of a child future doesn't trigger the completion of a > >> >> parent. > >> >> So, we will need to extend the future anyway. > >> >> > >> > > >> > First of all, as Pavel mentioned, you can attach your own listener > >> > before > >> > returning a CompletableFuture to the user. You don't need to extend. > >> > Second of all, there is still a discussion to be had on whether the > >> parent > >> > needs to be completed. I don't actually think it's obvious, and most > >> likely > >> > it's case by case. With CompletableFuture you have enough flexibility > >> > to > >> > control the behavior. > >> > > >> > > >> >> > >> >> Also, cancel() method contract differs from IgniteFuture in 2.0, > >> >> Completable method return true if the future was cancelled, > >> >> but IgniteFuture
Re: [DISCUSSION] IgniteFuture class future in Ignite-3.0.
Andrey, > Reactive abstractions look more suitable for Queries rather than > cache/table async operations and don't offer the power of chaining like > CompletableStage. Could you please elaborate what capabilities do you mean? AFAIK there are quite powerful APIs for singular reactive abstractions as well. E.g. [1]. [1] https://projectreactor.io/docs/core/release/api/reactor/core/publisher/Mono.html 2021-04-01 12:33 GMT+03:00, Andrey Mashenkov : > Val, > I just complain JDK CompletableFuture itself is not ideal, but already > implemented, well-known and tested. > It still a good alternative compared to custom future implementation to me. > > Ok, I feel most of us agree with CompletableFuture as a replacement for > custom IgniteFuture. > Let's try to use it, but keep in mind that we MUST return a defective copy > (via copy() or minimalStage()) to user to prevent misusage on the user > side. > It would be nice if we'll follow the same requirement in our internal code, > e.g. if a component returns a future that further can be used in other > components, especially in custom plugins. > > Ivan, thanks for the example. > Reactive abstractions look more suitable for Queries rather than > cache/table async operations and don't offer the power of chaining like > CompletableStage. > AFAIK, guys who involved in SQL engine development based on Calcite already > use reactive patterns. > > > On Thu, Apr 1, 2021 at 3:15 AM Ivan Pavlukhin wrote: > >> Folks, >> >> Regarding Reactive abstractions. While it might look too complex for >> simple KV cases it can be quite powerful for queries. Also there are >> known solutions for cancellation, backpressure and flow control. It >> can greatly simplify things for users familiar with Reactive >> programming rather than learning Ignite-specific Query/QueryCursor >> API. >> >> Also it looks like there are well-defined semantics [1]. E.g. >> cancellation seems to be defined much better than for >> CompletableFuture. >> >> [1] >> https://github.com/reactive-streams/reactive-streams-jvm#specification >> >> 2021-03-31 21:30 GMT+03:00, Valentin Kulichenko < >> valentin.kuliche...@gmail.com>: >> > Hi Andrey, >> > >> > Please see below. >> > >> > -Val >> > >> > On Wed, Mar 31, 2021 at 7:55 AM Andrey Mashenkov >> > >> > wrote: >> > >> >> CompletableFuture cancellation will not work as many users expected. >> >> Javadoc says: >> >> /* Since (unlike {@link FutureTask}) this class has no direct >> >> * control over the computation that causes it to be completed, >> >> * cancellation is treated as just another form of exceptional >> >> * completion. >> >> */ >> >> >> > >> > Let's not make assumptions about the expectations of the users. That's >> > exactly why I initially leaned towards a custom interface as well, but >> > that's a mistake. >> > Indeed, this contract might look weird to us, because the current >> > version >> > of Ignite behaves differently. However, there are much more developers >> > using CompletableFuture and other standard async frameworks, than >> > developers using the async functionality of Ignite. Therefore, our >> > intuitions can easily be wrong. >> > >> > >> >> Completion of a child future doesn't trigger the completion of a >> >> parent. >> >> So, we will need to extend the future anyway. >> >> >> > >> > First of all, as Pavel mentioned, you can attach your own listener >> > before >> > returning a CompletableFuture to the user. You don't need to extend. >> > Second of all, there is still a discussion to be had on whether the >> parent >> > needs to be completed. I don't actually think it's obvious, and most >> likely >> > it's case by case. With CompletableFuture you have enough flexibility >> > to >> > control the behavior. >> > >> > >> >> >> >> Also, cancel() method contract differs from IgniteFuture in 2.0, >> >> Completable method return true if the future was cancelled, >> >> but IgniteFuture return true only if it wasn't cancel prior the call. >> >> Thus, if cancel() was called twice we will have different results for >> >> CompletableFuture and IgniteFuture, >> >> that makes CompletableFuture barely usable for our internal purposes. >> >> >> > >> > It doesn't really matter how IgniteFuture in 2.0 behaves. It was >> > created >> > long before continuations and other async concepts became mainstream >> > (in >> > Java world at least). >> > Also, we don't have to use it for internal purposes, of course. I'm >> > only >> > talking about public APIs. >> > >> > >> >> >> >> BTW, CompletableFuture.anyOf() still can be used, see >> >> CompletionStage.toCompletableFuture() contract. >> >> >> > >> > In my view, this actually kills the idea of a custom future. Basically, >> > the proposal is to introduce a custom entity to restrict access to some >> of >> > the CompletableFuture methods, but then allow to convert this custom >> entity >> > to a CompletableFuture that has all the methods. This makes zero sense >> > to >> > me. >> > >> > >> >> >> >> The more I
Re: [DISCUSSION] IgniteFuture class future in Ignite-3.0.
Val, I just complain JDK CompletableFuture itself is not ideal, but already implemented, well-known and tested. It still a good alternative compared to custom future implementation to me. Ok, I feel most of us agree with CompletableFuture as a replacement for custom IgniteFuture. Let's try to use it, but keep in mind that we MUST return a defective copy (via copy() or minimalStage()) to user to prevent misusage on the user side. It would be nice if we'll follow the same requirement in our internal code, e.g. if a component returns a future that further can be used in other components, especially in custom plugins. Ivan, thanks for the example. Reactive abstractions look more suitable for Queries rather than cache/table async operations and don't offer the power of chaining like CompletableStage. AFAIK, guys who involved in SQL engine development based on Calcite already use reactive patterns. On Thu, Apr 1, 2021 at 3:15 AM Ivan Pavlukhin wrote: > Folks, > > Regarding Reactive abstractions. While it might look too complex for > simple KV cases it can be quite powerful for queries. Also there are > known solutions for cancellation, backpressure and flow control. It > can greatly simplify things for users familiar with Reactive > programming rather than learning Ignite-specific Query/QueryCursor > API. > > Also it looks like there are well-defined semantics [1]. E.g. > cancellation seems to be defined much better than for > CompletableFuture. > > [1] https://github.com/reactive-streams/reactive-streams-jvm#specification > > 2021-03-31 21:30 GMT+03:00, Valentin Kulichenko < > valentin.kuliche...@gmail.com>: > > Hi Andrey, > > > > Please see below. > > > > -Val > > > > On Wed, Mar 31, 2021 at 7:55 AM Andrey Mashenkov > > > > wrote: > > > >> CompletableFuture cancellation will not work as many users expected. > >> Javadoc says: > >> /* Since (unlike {@link FutureTask}) this class has no direct > >> * control over the computation that causes it to be completed, > >> * cancellation is treated as just another form of exceptional > >> * completion. > >> */ > >> > > > > Let's not make assumptions about the expectations of the users. That's > > exactly why I initially leaned towards a custom interface as well, but > > that's a mistake. > > Indeed, this contract might look weird to us, because the current version > > of Ignite behaves differently. However, there are much more developers > > using CompletableFuture and other standard async frameworks, than > > developers using the async functionality of Ignite. Therefore, our > > intuitions can easily be wrong. > > > > > >> Completion of a child future doesn't trigger the completion of a parent. > >> So, we will need to extend the future anyway. > >> > > > > First of all, as Pavel mentioned, you can attach your own listener before > > returning a CompletableFuture to the user. You don't need to extend. > > Second of all, there is still a discussion to be had on whether the > parent > > needs to be completed. I don't actually think it's obvious, and most > likely > > it's case by case. With CompletableFuture you have enough flexibility to > > control the behavior. > > > > > >> > >> Also, cancel() method contract differs from IgniteFuture in 2.0, > >> Completable method return true if the future was cancelled, > >> but IgniteFuture return true only if it wasn't cancel prior the call. > >> Thus, if cancel() was called twice we will have different results for > >> CompletableFuture and IgniteFuture, > >> that makes CompletableFuture barely usable for our internal purposes. > >> > > > > It doesn't really matter how IgniteFuture in 2.0 behaves. It was created > > long before continuations and other async concepts became mainstream (in > > Java world at least). > > Also, we don't have to use it for internal purposes, of course. I'm only > > talking about public APIs. > > > > > >> > >> BTW, CompletableFuture.anyOf() still can be used, see > >> CompletionStage.toCompletableFuture() contract. > >> > > > > In my view, this actually kills the idea of a custom future. Basically, > > the proposal is to introduce a custom entity to restrict access to some > of > > the CompletableFuture methods, but then allow to convert this custom > entity > > to a CompletableFuture that has all the methods. This makes zero sense to > > me. > > > > > >> > >> The more I learn about CompletableFuture the stronger my opinion about > >> overriding it. > >> > >> > >> On Wed, Mar 31, 2021 at 9:31 AM Denis Garus > wrote: > >> > >> > > Stripping them from such functionality, which they are used to, is > >> > > most > >> > likely a bad idea. > >> > > >> > Completely agree with this point of view. > >> > Moreover, a user can pass CompletableFuture to another library to do > >> > any > >> > manipulations. > >> > So if we want to introduce our class instead of the java class, we > >> > should > >> > have solid arguments; > >> > otherwise, it can be a reason for irritation. > >> > > >> > ср, 31 мар. 2021
Re: [DISCUSSION] IgniteFuture class future in Ignite-3.0.
Folks, Regarding Reactive abstractions. While it might look too complex for simple KV cases it can be quite powerful for queries. Also there are known solutions for cancellation, backpressure and flow control. It can greatly simplify things for users familiar with Reactive programming rather than learning Ignite-specific Query/QueryCursor API. Also it looks like there are well-defined semantics [1]. E.g. cancellation seems to be defined much better than for CompletableFuture. [1] https://github.com/reactive-streams/reactive-streams-jvm#specification 2021-03-31 21:30 GMT+03:00, Valentin Kulichenko : > Hi Andrey, > > Please see below. > > -Val > > On Wed, Mar 31, 2021 at 7:55 AM Andrey Mashenkov > > wrote: > >> CompletableFuture cancellation will not work as many users expected. >> Javadoc says: >> /* Since (unlike {@link FutureTask}) this class has no direct >> * control over the computation that causes it to be completed, >> * cancellation is treated as just another form of exceptional >> * completion. >> */ >> > > Let's not make assumptions about the expectations of the users. That's > exactly why I initially leaned towards a custom interface as well, but > that's a mistake. > Indeed, this contract might look weird to us, because the current version > of Ignite behaves differently. However, there are much more developers > using CompletableFuture and other standard async frameworks, than > developers using the async functionality of Ignite. Therefore, our > intuitions can easily be wrong. > > >> Completion of a child future doesn't trigger the completion of a parent. >> So, we will need to extend the future anyway. >> > > First of all, as Pavel mentioned, you can attach your own listener before > returning a CompletableFuture to the user. You don't need to extend. > Second of all, there is still a discussion to be had on whether the parent > needs to be completed. I don't actually think it's obvious, and most likely > it's case by case. With CompletableFuture you have enough flexibility to > control the behavior. > > >> >> Also, cancel() method contract differs from IgniteFuture in 2.0, >> Completable method return true if the future was cancelled, >> but IgniteFuture return true only if it wasn't cancel prior the call. >> Thus, if cancel() was called twice we will have different results for >> CompletableFuture and IgniteFuture, >> that makes CompletableFuture barely usable for our internal purposes. >> > > It doesn't really matter how IgniteFuture in 2.0 behaves. It was created > long before continuations and other async concepts became mainstream (in > Java world at least). > Also, we don't have to use it for internal purposes, of course. I'm only > talking about public APIs. > > >> >> BTW, CompletableFuture.anyOf() still can be used, see >> CompletionStage.toCompletableFuture() contract. >> > > In my view, this actually kills the idea of a custom future. Basically, > the proposal is to introduce a custom entity to restrict access to some of > the CompletableFuture methods, but then allow to convert this custom entity > to a CompletableFuture that has all the methods. This makes zero sense to > me. > > >> >> The more I learn about CompletableFuture the stronger my opinion about >> overriding it. >> >> >> On Wed, Mar 31, 2021 at 9:31 AM Denis Garus wrote: >> >> > > Stripping them from such functionality, which they are used to, is >> > > most >> > likely a bad idea. >> > >> > Completely agree with this point of view. >> > Moreover, a user can pass CompletableFuture to another library to do >> > any >> > manipulations. >> > So if we want to introduce our class instead of the java class, we >> > should >> > have solid arguments; >> > otherwise, it can be a reason for irritation. >> > >> > ср, 31 мар. 2021 г. в 09:06, Pavel Tupitsyn : >> > >> > > Val, >> > > >> > > > we can have an IgniteFuture that extends CompletableFuture. >> > > > This might be useful if want the cancel() operation to cancel the >> > > > underlying operation >> > > >> > > I think we can subscribe to the cancellation of the CompletableFuture >> > > and cancel the underlying operation without an extra class, >> > > something like >> > > >> > > fut.exceptionally(t -> { >> > > if (t instanceof CancellationException) { >> > > // Cancel Ignite operation >> > > } >> > > }); >> > > >> > > On Wed, Mar 31, 2021 at 7:45 AM Valentin Kulichenko < >> > > valentin.kuliche...@gmail.com> wrote: >> > > >> > > > These are actually some interesting points. As I'm thinking more >> about >> > > > this, I'm leaning towards changing my opinion and voting for the >> > > > CompletableFuture. Here is my reasoning. >> > > > >> > > > First, it's important to keep in mind that CompletableFuture is not >> an >> > > > interface that we will implement, it's an implemented class. >> Therefore, >> > > > some of the concerns around complete() and cancel() method are not >> > really >> > > > relevant -- it's
Re: [DISCUSSION] IgniteFuture class future in Ignite-3.0.
Hi Andrey, Please see below. -Val On Wed, Mar 31, 2021 at 7:55 AM Andrey Mashenkov wrote: > CompletableFuture cancellation will not work as many users expected. > Javadoc says: > /* Since (unlike {@link FutureTask}) this class has no direct > * control over the computation that causes it to be completed, > * cancellation is treated as just another form of exceptional > * completion. > */ > Let's not make assumptions about the expectations of the users. That's exactly why I initially leaned towards a custom interface as well, but that's a mistake. Indeed, this contract might look weird to us, because the current version of Ignite behaves differently. However, there are much more developers using CompletableFuture and other standard async frameworks, than developers using the async functionality of Ignite. Therefore, our intuitions can easily be wrong. > Completion of a child future doesn't trigger the completion of a parent. > So, we will need to extend the future anyway. > First of all, as Pavel mentioned, you can attach your own listener before returning a CompletableFuture to the user. You don't need to extend. Second of all, there is still a discussion to be had on whether the parent needs to be completed. I don't actually think it's obvious, and most likely it's case by case. With CompletableFuture you have enough flexibility to control the behavior. > > Also, cancel() method contract differs from IgniteFuture in 2.0, > Completable method return true if the future was cancelled, > but IgniteFuture return true only if it wasn't cancel prior the call. > Thus, if cancel() was called twice we will have different results for > CompletableFuture and IgniteFuture, > that makes CompletableFuture barely usable for our internal purposes. > It doesn't really matter how IgniteFuture in 2.0 behaves. It was created long before continuations and other async concepts became mainstream (in Java world at least). Also, we don't have to use it for internal purposes, of course. I'm only talking about public APIs. > > BTW, CompletableFuture.anyOf() still can be used, see > CompletionStage.toCompletableFuture() contract. > In my view, this actually kills the idea of a custom future. Basically, the proposal is to introduce a custom entity to restrict access to some of the CompletableFuture methods, but then allow to convert this custom entity to a CompletableFuture that has all the methods. This makes zero sense to me. > > The more I learn about CompletableFuture the stronger my opinion about > overriding it. > > > On Wed, Mar 31, 2021 at 9:31 AM Denis Garus wrote: > > > > Stripping them from such functionality, which they are used to, is most > > likely a bad idea. > > > > Completely agree with this point of view. > > Moreover, a user can pass CompletableFuture to another library to do any > > manipulations. > > So if we want to introduce our class instead of the java class, we should > > have solid arguments; > > otherwise, it can be a reason for irritation. > > > > ср, 31 мар. 2021 г. в 09:06, Pavel Tupitsyn : > > > > > Val, > > > > > > > we can have an IgniteFuture that extends CompletableFuture. > > > > This might be useful if want the cancel() operation to cancel the > > > > underlying operation > > > > > > I think we can subscribe to the cancellation of the CompletableFuture > > > and cancel the underlying operation without an extra class, > > > something like > > > > > > fut.exceptionally(t -> { > > > if (t instanceof CancellationException) { > > > // Cancel Ignite operation > > > } > > > }); > > > > > > On Wed, Mar 31, 2021 at 7:45 AM Valentin Kulichenko < > > > valentin.kuliche...@gmail.com> wrote: > > > > > > > These are actually some interesting points. As I'm thinking more > about > > > > this, I'm leaning towards changing my opinion and voting for the > > > > CompletableFuture. Here is my reasoning. > > > > > > > > First, it's important to keep in mind that CompletableFuture is not > an > > > > interface that we will implement, it's an implemented class. > Therefore, > > > > some of the concerns around complete() and cancel() method are not > > really > > > > relevant -- it's not up to us how these methods behave, they're > already > > > > implemented. > > > > > > > > Second, CompletableFuture does provide some useful functionality > (anyOf > > > is > > > > one of the examples). I can even envision users wanting to complete > the > > > > future under certain circumstances, e.g. after a timeout, using > > > > the completeOnTimeout method. Stripping them from such functionality, > > > which > > > > they are used to, is most likely a bad idea. > > > > > > > > And finally, we can have an IgniteFuture that extends > > CompletableFuture. > > > > This might be useful if want the cancel() operation to cancel the > > > > underlying operation. This way we keep all the functionality of > > > > CompletableFuture while keeping a certain amount of
Re: [DISCUSSION] IgniteFuture class future in Ignite-3.0.
CompletableFuture cancellation will not work as many users expected. Javadoc says: /* Since (unlike {@link FutureTask}) this class has no direct * control over the computation that causes it to be completed, * cancellation is treated as just another form of exceptional * completion. */ Completion of a child future doesn't trigger the completion of a parent. So, we will need to extend the future anyway. Also, cancel() method contract differs from IgniteFuture in 2.0, Completable method return true if the future was cancelled, but IgniteFuture return true only if it wasn't cancel prior the call. Thus, if cancel() was called twice we will have different results for CompletableFuture and IgniteFuture, that makes CompletableFuture barely usable for our internal purposes. BTW, CompletableFuture.anyOf() still can be used, see CompletionStage.toCompletableFuture() contract. The more I learn about CompletableFuture the stronger my opinion about overriding it. On Wed, Mar 31, 2021 at 9:31 AM Denis Garus wrote: > > Stripping them from such functionality, which they are used to, is most > likely a bad idea. > > Completely agree with this point of view. > Moreover, a user can pass CompletableFuture to another library to do any > manipulations. > So if we want to introduce our class instead of the java class, we should > have solid arguments; > otherwise, it can be a reason for irritation. > > ср, 31 мар. 2021 г. в 09:06, Pavel Tupitsyn : > > > Val, > > > > > we can have an IgniteFuture that extends CompletableFuture. > > > This might be useful if want the cancel() operation to cancel the > > > underlying operation > > > > I think we can subscribe to the cancellation of the CompletableFuture > > and cancel the underlying operation without an extra class, > > something like > > > > fut.exceptionally(t -> { > > if (t instanceof CancellationException) { > > // Cancel Ignite operation > > } > > }); > > > > On Wed, Mar 31, 2021 at 7:45 AM Valentin Kulichenko < > > valentin.kuliche...@gmail.com> wrote: > > > > > These are actually some interesting points. As I'm thinking more about > > > this, I'm leaning towards changing my opinion and voting for the > > > CompletableFuture. Here is my reasoning. > > > > > > First, it's important to keep in mind that CompletableFuture is not an > > > interface that we will implement, it's an implemented class. Therefore, > > > some of the concerns around complete() and cancel() method are not > really > > > relevant -- it's not up to us how these methods behave, they're already > > > implemented. > > > > > > Second, CompletableFuture does provide some useful functionality (anyOf > > is > > > one of the examples). I can even envision users wanting to complete the > > > future under certain circumstances, e.g. after a timeout, using > > > the completeOnTimeout method. Stripping them from such functionality, > > which > > > they are used to, is most likely a bad idea. > > > > > > And finally, we can have an IgniteFuture that extends > CompletableFuture. > > > This might be useful if want the cancel() operation to cancel the > > > underlying operation. This way we keep all the functionality of > > > CompletableFuture while keeping a certain amount of flexibility for > > > specific cases. > > > > > > Thoughts? > > > > > > -Val > > > > > > > > > On Tue, Mar 30, 2021 at 5:36 AM Denis Garus > wrote: > > > > > > > > Completing future from outside will never respect other subscribers > > > that > > > > > may expect other guatantees. > > > > > > > > For example, if we talk about public API like IgniteCache, what > > > subscribers > > > > may expect other guatantees? > > > > IMHO, the best solution is to get the well-known standard interface > to > > a > > > > user, and he will be happy. > > > > > > > > But when we talk about internal classes like "exchange future" they > > could > > > > be custom futures if convenient. > > > > > > > > вт, 30 мар. 2021 г. в 15:25, Atri Sharma : > > > > > > > > > IMO the only way Ignite should cancel computations is iff cancel > > method > > > > is > > > > > invoked, not implicitly if complete is invoked. > > > > > > > > > > On Tue, 30 Mar 2021 at 4:58 PM, Denis Garus > > > wrote: > > > > > > > > > > > Hello! > > > > > > > > > > > > > Let's say a user started a compute with fut = > > > compute.runAsync(task); > > > > > > > and now calls fut.complete(someVal); Does this mean that Ignite > > no > > > > > longer > > > > > > needs to execute the task? > > > > > > > If the task is currently running, does it need to be canceled? > > > > > > > > > > > > Yes, this case looks like Ignite should cancel computations > > because a > > > > > user > > > > > > wants to complete the future. Why not? > > > > > > If there will be an opportunity to cancel a future, why is it a > bad > > > > > option > > > > > > to finish a future through a complete() method? > > > > > > > > > > > > > If you look at Ignite-2 code, you may
Re: [DISCUSSION] IgniteFuture class future in Ignite-3.0.
> Stripping them from such functionality, which they are used to, is most likely a bad idea. Completely agree with this point of view. Moreover, a user can pass CompletableFuture to another library to do any manipulations. So if we want to introduce our class instead of the java class, we should have solid arguments; otherwise, it can be a reason for irritation. ср, 31 мар. 2021 г. в 09:06, Pavel Tupitsyn : > Val, > > > we can have an IgniteFuture that extends CompletableFuture. > > This might be useful if want the cancel() operation to cancel the > > underlying operation > > I think we can subscribe to the cancellation of the CompletableFuture > and cancel the underlying operation without an extra class, > something like > > fut.exceptionally(t -> { > if (t instanceof CancellationException) { > // Cancel Ignite operation > } > }); > > On Wed, Mar 31, 2021 at 7:45 AM Valentin Kulichenko < > valentin.kuliche...@gmail.com> wrote: > > > These are actually some interesting points. As I'm thinking more about > > this, I'm leaning towards changing my opinion and voting for the > > CompletableFuture. Here is my reasoning. > > > > First, it's important to keep in mind that CompletableFuture is not an > > interface that we will implement, it's an implemented class. Therefore, > > some of the concerns around complete() and cancel() method are not really > > relevant -- it's not up to us how these methods behave, they're already > > implemented. > > > > Second, CompletableFuture does provide some useful functionality (anyOf > is > > one of the examples). I can even envision users wanting to complete the > > future under certain circumstances, e.g. after a timeout, using > > the completeOnTimeout method. Stripping them from such functionality, > which > > they are used to, is most likely a bad idea. > > > > And finally, we can have an IgniteFuture that extends CompletableFuture. > > This might be useful if want the cancel() operation to cancel the > > underlying operation. This way we keep all the functionality of > > CompletableFuture while keeping a certain amount of flexibility for > > specific cases. > > > > Thoughts? > > > > -Val > > > > > > On Tue, Mar 30, 2021 at 5:36 AM Denis Garus wrote: > > > > > > Completing future from outside will never respect other subscribers > > that > > > > may expect other guatantees. > > > > > > For example, if we talk about public API like IgniteCache, what > > subscribers > > > may expect other guatantees? > > > IMHO, the best solution is to get the well-known standard interface to > a > > > user, and he will be happy. > > > > > > But when we talk about internal classes like "exchange future" they > could > > > be custom futures if convenient. > > > > > > вт, 30 мар. 2021 г. в 15:25, Atri Sharma : > > > > > > > IMO the only way Ignite should cancel computations is iff cancel > method > > > is > > > > invoked, not implicitly if complete is invoked. > > > > > > > > On Tue, 30 Mar 2021 at 4:58 PM, Denis Garus > > wrote: > > > > > > > > > Hello! > > > > > > > > > > > Let's say a user started a compute with fut = > > compute.runAsync(task); > > > > > > and now calls fut.complete(someVal); Does this mean that Ignite > no > > > > longer > > > > > needs to execute the task? > > > > > > If the task is currently running, does it need to be canceled? > > > > > > > > > > Yes, this case looks like Ignite should cancel computations > because a > > > > user > > > > > wants to complete the future. Why not? > > > > > If there will be an opportunity to cancel a future, why is it a bad > > > > option > > > > > to finish a future through a complete() method? > > > > > > > > > > > If you look at Ignite-2 code, you may found a number of places > > where > > > we > > > > > return e.g. exchange futures or partition release futures. > > > > > > Assume the impact if we will return CompletableFuture instead, > > which > > > > can > > > > > be completed in 3-rd party plugin by mistake? > > > > > > > > > > If exchange futures or partition release futures can be returned to > > > 3-rd > > > > > party plugin by mistake, it is poor encapsulation. > > > > > And if it will be IgniteFuter rather than CompletedFuture, anyway, > > this > > > > can > > > > > harm. > > > > > > > > > > вт, 30 мар. 2021 г. в 13:14, Andrey Mashenkov < > > > > andrey.mashen...@gmail.com > > > > > >: > > > > > > > > > > > Guys, > > > > > > > > > > > > I want to remember there is one more point to pay attention to. > > > > > > Extending Future and CompletableStage is more than just prevents > > > > > unexpected > > > > > > behavior if a user completed the future. > > > > > > > > > > > > First of all, it helps us to write safer code as we won't a > method > > > > > contract > > > > > > exposed such methods as to a user as to a developer. > > > > > > If you look at Ignite-2 code, you may found a number of places > > where > > > we > > > > > > return e.g. exchange futures or
Re: [DISCUSSION] IgniteFuture class future in Ignite-3.0.
Val, > we can have an IgniteFuture that extends CompletableFuture. > This might be useful if want the cancel() operation to cancel the > underlying operation I think we can subscribe to the cancellation of the CompletableFuture and cancel the underlying operation without an extra class, something like fut.exceptionally(t -> { if (t instanceof CancellationException) { // Cancel Ignite operation } }); On Wed, Mar 31, 2021 at 7:45 AM Valentin Kulichenko < valentin.kuliche...@gmail.com> wrote: > These are actually some interesting points. As I'm thinking more about > this, I'm leaning towards changing my opinion and voting for the > CompletableFuture. Here is my reasoning. > > First, it's important to keep in mind that CompletableFuture is not an > interface that we will implement, it's an implemented class. Therefore, > some of the concerns around complete() and cancel() method are not really > relevant -- it's not up to us how these methods behave, they're already > implemented. > > Second, CompletableFuture does provide some useful functionality (anyOf is > one of the examples). I can even envision users wanting to complete the > future under certain circumstances, e.g. after a timeout, using > the completeOnTimeout method. Stripping them from such functionality, which > they are used to, is most likely a bad idea. > > And finally, we can have an IgniteFuture that extends CompletableFuture. > This might be useful if want the cancel() operation to cancel the > underlying operation. This way we keep all the functionality of > CompletableFuture while keeping a certain amount of flexibility for > specific cases. > > Thoughts? > > -Val > > > On Tue, Mar 30, 2021 at 5:36 AM Denis Garus wrote: > > > > Completing future from outside will never respect other subscribers > that > > > may expect other guatantees. > > > > For example, if we talk about public API like IgniteCache, what > subscribers > > may expect other guatantees? > > IMHO, the best solution is to get the well-known standard interface to a > > user, and he will be happy. > > > > But when we talk about internal classes like "exchange future" they could > > be custom futures if convenient. > > > > вт, 30 мар. 2021 г. в 15:25, Atri Sharma : > > > > > IMO the only way Ignite should cancel computations is iff cancel method > > is > > > invoked, not implicitly if complete is invoked. > > > > > > On Tue, 30 Mar 2021 at 4:58 PM, Denis Garus > wrote: > > > > > > > Hello! > > > > > > > > > Let's say a user started a compute with fut = > compute.runAsync(task); > > > > > and now calls fut.complete(someVal); Does this mean that Ignite no > > > longer > > > > needs to execute the task? > > > > > If the task is currently running, does it need to be canceled? > > > > > > > > Yes, this case looks like Ignite should cancel computations because a > > > user > > > > wants to complete the future. Why not? > > > > If there will be an opportunity to cancel a future, why is it a bad > > > option > > > > to finish a future through a complete() method? > > > > > > > > > If you look at Ignite-2 code, you may found a number of places > where > > we > > > > return e.g. exchange futures or partition release futures. > > > > > Assume the impact if we will return CompletableFuture instead, > which > > > can > > > > be completed in 3-rd party plugin by mistake? > > > > > > > > If exchange futures or partition release futures can be returned to > > 3-rd > > > > party plugin by mistake, it is poor encapsulation. > > > > And if it will be IgniteFuter rather than CompletedFuture, anyway, > this > > > can > > > > harm. > > > > > > > > вт, 30 мар. 2021 г. в 13:14, Andrey Mashenkov < > > > andrey.mashen...@gmail.com > > > > >: > > > > > > > > > Guys, > > > > > > > > > > I want to remember there is one more point to pay attention to. > > > > > Extending Future and CompletableStage is more than just prevents > > > > unexpected > > > > > behavior if a user completed the future. > > > > > > > > > > First of all, it helps us to write safer code as we won't a method > > > > contract > > > > > exposed such methods as to a user as to a developer. > > > > > If you look at Ignite-2 code, you may found a number of places > where > > we > > > > > return e.g. exchange futures or partition release futures. > > > > > Assume the impact if we will return CompletableFuture instead, > which > > > can > > > > be > > > > > completed in 3-rd party plugin by mistake? > > > > > > > > > > The suggested approach allows us to don't bother if a > > CompletableFuture > > > > has > > > > > to be wrapped or not. > > > > > > > > > > > > > > > On Tue, Mar 30, 2021 at 12:22 PM Alexey Goncharuk < > > > > > alexey.goncha...@gmail.com> wrote: > > > > > > > > > > > Ivan, > > > > > > > > > > > > My concern with the concept of a user completing the future > > returned > > > > from > > > > > > Ignite public API is that it is unclear how to interpret this > > action >
Re: [DISCUSSION] IgniteFuture class future in Ignite-3.0.
These are actually some interesting points. As I'm thinking more about this, I'm leaning towards changing my opinion and voting for the CompletableFuture. Here is my reasoning. First, it's important to keep in mind that CompletableFuture is not an interface that we will implement, it's an implemented class. Therefore, some of the concerns around complete() and cancel() method are not really relevant -- it's not up to us how these methods behave, they're already implemented. Second, CompletableFuture does provide some useful functionality (anyOf is one of the examples). I can even envision users wanting to complete the future under certain circumstances, e.g. after a timeout, using the completeOnTimeout method. Stripping them from such functionality, which they are used to, is most likely a bad idea. And finally, we can have an IgniteFuture that extends CompletableFuture. This might be useful if want the cancel() operation to cancel the underlying operation. This way we keep all the functionality of CompletableFuture while keeping a certain amount of flexibility for specific cases. Thoughts? -Val On Tue, Mar 30, 2021 at 5:36 AM Denis Garus wrote: > > Completing future from outside will never respect other subscribers that > > may expect other guatantees. > > For example, if we talk about public API like IgniteCache, what subscribers > may expect other guatantees? > IMHO, the best solution is to get the well-known standard interface to a > user, and he will be happy. > > But when we talk about internal classes like "exchange future" they could > be custom futures if convenient. > > вт, 30 мар. 2021 г. в 15:25, Atri Sharma : > > > IMO the only way Ignite should cancel computations is iff cancel method > is > > invoked, not implicitly if complete is invoked. > > > > On Tue, 30 Mar 2021 at 4:58 PM, Denis Garus wrote: > > > > > Hello! > > > > > > > Let's say a user started a compute with fut = compute.runAsync(task); > > > > and now calls fut.complete(someVal); Does this mean that Ignite no > > longer > > > needs to execute the task? > > > > If the task is currently running, does it need to be canceled? > > > > > > Yes, this case looks like Ignite should cancel computations because a > > user > > > wants to complete the future. Why not? > > > If there will be an opportunity to cancel a future, why is it a bad > > option > > > to finish a future through a complete() method? > > > > > > > If you look at Ignite-2 code, you may found a number of places where > we > > > return e.g. exchange futures or partition release futures. > > > > Assume the impact if we will return CompletableFuture instead, which > > can > > > be completed in 3-rd party plugin by mistake? > > > > > > If exchange futures or partition release futures can be returned to > 3-rd > > > party plugin by mistake, it is poor encapsulation. > > > And if it will be IgniteFuter rather than CompletedFuture, anyway, this > > can > > > harm. > > > > > > вт, 30 мар. 2021 г. в 13:14, Andrey Mashenkov < > > andrey.mashen...@gmail.com > > > >: > > > > > > > Guys, > > > > > > > > I want to remember there is one more point to pay attention to. > > > > Extending Future and CompletableStage is more than just prevents > > > unexpected > > > > behavior if a user completed the future. > > > > > > > > First of all, it helps us to write safer code as we won't a method > > > contract > > > > exposed such methods as to a user as to a developer. > > > > If you look at Ignite-2 code, you may found a number of places where > we > > > > return e.g. exchange futures or partition release futures. > > > > Assume the impact if we will return CompletableFuture instead, which > > can > > > be > > > > completed in 3-rd party plugin by mistake? > > > > > > > > The suggested approach allows us to don't bother if a > CompletableFuture > > > has > > > > to be wrapped or not. > > > > > > > > > > > > On Tue, Mar 30, 2021 at 12:22 PM Alexey Goncharuk < > > > > alexey.goncha...@gmail.com> wrote: > > > > > > > > > Ivan, > > > > > > > > > > My concern with the concept of a user completing the future > returned > > > from > > > > > Ignite public API is that it is unclear how to interpret this > action > > > > (this > > > > > backs Val's message). > > > > > Let's say a user started a compute with fut = > compute.runAsync(task); > > > and > > > > > now calls fut.complete(someVal); Does this mean that Ignite no > longer > > > > needs > > > > > to execute the task? If the task is currently running, does it need > > to > > > be > > > > > canceled? > > > > > > > > > > Using CompletableFuture.anyOf() is a good instrument in this case > > > because > > > > > it makes the 'first future wins' contract explicit in the code. > > Besides > > > > > that, the point regarding the cancel() method is valid, and we will > > > need > > > > > some custom mechanics to cancel a computation, so a custom > interface > > > that > > > > > simply extends both Future and CompletableStage seems reasonable to > >
Re: [DISCUSSION] IgniteFuture class future in Ignite-3.0.
> Completing future from outside will never respect other subscribers that > may expect other guatantees. For example, if we talk about public API like IgniteCache, what subscribers may expect other guatantees? IMHO, the best solution is to get the well-known standard interface to a user, and he will be happy. But when we talk about internal classes like "exchange future" they could be custom futures if convenient. вт, 30 мар. 2021 г. в 15:25, Atri Sharma : > IMO the only way Ignite should cancel computations is iff cancel method is > invoked, not implicitly if complete is invoked. > > On Tue, 30 Mar 2021 at 4:58 PM, Denis Garus wrote: > > > Hello! > > > > > Let's say a user started a compute with fut = compute.runAsync(task); > > > and now calls fut.complete(someVal); Does this mean that Ignite no > longer > > needs to execute the task? > > > If the task is currently running, does it need to be canceled? > > > > Yes, this case looks like Ignite should cancel computations because a > user > > wants to complete the future. Why not? > > If there will be an opportunity to cancel a future, why is it a bad > option > > to finish a future through a complete() method? > > > > > If you look at Ignite-2 code, you may found a number of places where we > > return e.g. exchange futures or partition release futures. > > > Assume the impact if we will return CompletableFuture instead, which > can > > be completed in 3-rd party plugin by mistake? > > > > If exchange futures or partition release futures can be returned to 3-rd > > party plugin by mistake, it is poor encapsulation. > > And if it will be IgniteFuter rather than CompletedFuture, anyway, this > can > > harm. > > > > вт, 30 мар. 2021 г. в 13:14, Andrey Mashenkov < > andrey.mashen...@gmail.com > > >: > > > > > Guys, > > > > > > I want to remember there is one more point to pay attention to. > > > Extending Future and CompletableStage is more than just prevents > > unexpected > > > behavior if a user completed the future. > > > > > > First of all, it helps us to write safer code as we won't a method > > contract > > > exposed such methods as to a user as to a developer. > > > If you look at Ignite-2 code, you may found a number of places where we > > > return e.g. exchange futures or partition release futures. > > > Assume the impact if we will return CompletableFuture instead, which > can > > be > > > completed in 3-rd party plugin by mistake? > > > > > > The suggested approach allows us to don't bother if a CompletableFuture > > has > > > to be wrapped or not. > > > > > > > > > On Tue, Mar 30, 2021 at 12:22 PM Alexey Goncharuk < > > > alexey.goncha...@gmail.com> wrote: > > > > > > > Ivan, > > > > > > > > My concern with the concept of a user completing the future returned > > from > > > > Ignite public API is that it is unclear how to interpret this action > > > (this > > > > backs Val's message). > > > > Let's say a user started a compute with fut = compute.runAsync(task); > > and > > > > now calls fut.complete(someVal); Does this mean that Ignite no longer > > > needs > > > > to execute the task? If the task is currently running, does it need > to > > be > > > > canceled? > > > > > > > > Using CompletableFuture.anyOf() is a good instrument in this case > > because > > > > it makes the 'first future wins' contract explicit in the code. > Besides > > > > that, the point regarding the cancel() method is valid, and we will > > need > > > > some custom mechanics to cancel a computation, so a custom interface > > that > > > > simply extends both Future and CompletableStage seems reasonable to > me. > > > > > > > > --AG > > > > > > > > пн, 29 мар. 2021 г. в 09:12, Ivan Pavlukhin : > > > > > > > > > Val, > > > > > > > > > > There were enough hype around Reactive programming past years. I > > > > > remind a lot of talks about RxJava. And I suppose it worth to > > consider > > > > > it. But it requires some time to study modern trends to make a > > choice. > > > > > So far I am not ready to facilitate Reactive API for Ignite 3. > > > > > > > > > > Regarding CompletableFuture. > > > > > > > > > > > The point is that currently a future returned from any of > Ignite's > > > > async > > > > > > operations is supposed to be completed with a value only by > Ignite > > > > > itself, > > > > > > not by the user. If we follow the same approach in Ignite 3, > > > returning > > > > > > CompletableFuture is surely wrong in my view. > > > > > > > > > > My first thoughts was similar. But later I thought what a user > would > > > > > like do with returned future. And one of cases I imagined was a > case > > > > > of alternative result. E.g. a user uses Ignite and another data > > source > > > > > in his application. He wants to use a value arrived faster. He > > > > > combines 2 futures like CompletableFuture.anyOf(...). Consequently > > > > > even if we prohibit CompletableFuture.complete(...) explicitly then > > it > > > > > will be possible to create a combination that will
Re: [DISCUSSION] IgniteFuture class future in Ignite-3.0.
IMO the only way Ignite should cancel computations is iff cancel method is invoked, not implicitly if complete is invoked. On Tue, 30 Mar 2021 at 4:58 PM, Denis Garus wrote: > Hello! > > > Let's say a user started a compute with fut = compute.runAsync(task); > > and now calls fut.complete(someVal); Does this mean that Ignite no longer > needs to execute the task? > > If the task is currently running, does it need to be canceled? > > Yes, this case looks like Ignite should cancel computations because a user > wants to complete the future. Why not? > If there will be an opportunity to cancel a future, why is it a bad option > to finish a future through a complete() method? > > > If you look at Ignite-2 code, you may found a number of places where we > return e.g. exchange futures or partition release futures. > > Assume the impact if we will return CompletableFuture instead, which can > be completed in 3-rd party plugin by mistake? > > If exchange futures or partition release futures can be returned to 3-rd > party plugin by mistake, it is poor encapsulation. > And if it will be IgniteFuter rather than CompletedFuture, anyway, this can > harm. > > вт, 30 мар. 2021 г. в 13:14, Andrey Mashenkov >: > > > Guys, > > > > I want to remember there is one more point to pay attention to. > > Extending Future and CompletableStage is more than just prevents > unexpected > > behavior if a user completed the future. > > > > First of all, it helps us to write safer code as we won't a method > contract > > exposed such methods as to a user as to a developer. > > If you look at Ignite-2 code, you may found a number of places where we > > return e.g. exchange futures or partition release futures. > > Assume the impact if we will return CompletableFuture instead, which can > be > > completed in 3-rd party plugin by mistake? > > > > The suggested approach allows us to don't bother if a CompletableFuture > has > > to be wrapped or not. > > > > > > On Tue, Mar 30, 2021 at 12:22 PM Alexey Goncharuk < > > alexey.goncha...@gmail.com> wrote: > > > > > Ivan, > > > > > > My concern with the concept of a user completing the future returned > from > > > Ignite public API is that it is unclear how to interpret this action > > (this > > > backs Val's message). > > > Let's say a user started a compute with fut = compute.runAsync(task); > and > > > now calls fut.complete(someVal); Does this mean that Ignite no longer > > needs > > > to execute the task? If the task is currently running, does it need to > be > > > canceled? > > > > > > Using CompletableFuture.anyOf() is a good instrument in this case > because > > > it makes the 'first future wins' contract explicit in the code. Besides > > > that, the point regarding the cancel() method is valid, and we will > need > > > some custom mechanics to cancel a computation, so a custom interface > that > > > simply extends both Future and CompletableStage seems reasonable to me. > > > > > > --AG > > > > > > пн, 29 мар. 2021 г. в 09:12, Ivan Pavlukhin : > > > > > > > Val, > > > > > > > > There were enough hype around Reactive programming past years. I > > > > remind a lot of talks about RxJava. And I suppose it worth to > consider > > > > it. But it requires some time to study modern trends to make a > choice. > > > > So far I am not ready to facilitate Reactive API for Ignite 3. > > > > > > > > Regarding CompletableFuture. > > > > > > > > > The point is that currently a future returned from any of Ignite's > > > async > > > > > operations is supposed to be completed with a value only by Ignite > > > > itself, > > > > > not by the user. If we follow the same approach in Ignite 3, > > returning > > > > > CompletableFuture is surely wrong in my view. > > > > > > > > My first thoughts was similar. But later I thought what a user would > > > > like do with returned future. And one of cases I imagined was a case > > > > of alternative result. E.g. a user uses Ignite and another data > source > > > > in his application. He wants to use a value arrived faster. He > > > > combines 2 futures like CompletableFuture.anyOf(...). Consequently > > > > even if we prohibit CompletableFuture.complete(...) explicitly then > it > > > > will be possible to create a combination that will allow premature > > > > future completion. After all generally CompletableFuture is a > > > > placeholder for async computaion result and if a user wants to > > > > substitute result returned from Ignite why should we disallow him to > > > > do it? > > > > > > > > Also I found one more suspicious thing with CompletableFuture. As it > > > > is a concrete class it implements a cancel() method. And as I see the > > > > implementation does not try to cancel underlying computations. Is not > > > > it a problem? > > > > > > > > 2021-03-29 7:30 GMT+03:00, Valentin Kulichenko < > > > > valentin.kuliche...@gmail.com>: > > > > > Ivan, > > > > > > > > > > It's not really about the "harm", but more about "what should we do > > if > > > > this
Re: [DISCUSSION] IgniteFuture class future in Ignite-3.0.
> > Yes, this case looks like Ignite should cancel computations because a user > wants to complete the future. Why not? > If there will be an opportunity to cancel a future, why is it a bad option > to finish a future through a complete() method? Future has cancel() method for this. Completing future from outside will never respect other subscribers that may expect other guatantees. > > If exchange futures or partition release futures can be returned to 3-rd > party plugin by mistake, it is poor encapsulation. > And if it will be IgniteFuter rather than CompletedFuture, anyway, this can > harm. > IgniteFuture contract will not have complete() or similar methods, but we a free to pass any implementation. Obviously, if a user casts the future to CompletableFuture or uses a reflection to access IgniteFutureImpl hidden methods - that means the user breaks the contract and CompletableFuture defensive copy will not help either. On Tue, Mar 30, 2021 at 2:28 PM Denis Garus wrote: > Hello! > > > Let's say a user started a compute with fut = compute.runAsync(task); > > and now calls fut.complete(someVal); Does this mean that Ignite no longer > needs to execute the task? > > If the task is currently running, does it need to be canceled? > > Yes, this case looks like Ignite should cancel computations because a user > wants to complete the future. Why not? > If there will be an opportunity to cancel a future, why is it a bad option > to finish a future through a complete() method? > > > If you look at Ignite-2 code, you may found a number of places where we > return e.g. exchange futures or partition release futures. > > Assume the impact if we will return CompletableFuture instead, which can > be completed in 3-rd party plugin by mistake? > > If exchange futures or partition release futures can be returned to 3-rd > party plugin by mistake, it is poor encapsulation. > And if it will be IgniteFuter rather than CompletedFuture, anyway, this can > harm. > > вт, 30 мар. 2021 г. в 13:14, Andrey Mashenkov >: > > > Guys, > > > > I want to remember there is one more point to pay attention to. > > Extending Future and CompletableStage is more than just prevents > unexpected > > behavior if a user completed the future. > > > > First of all, it helps us to write safer code as we won't a method > contract > > exposed such methods as to a user as to a developer. > > If you look at Ignite-2 code, you may found a number of places where we > > return e.g. exchange futures or partition release futures. > > Assume the impact if we will return CompletableFuture instead, which can > be > > completed in 3-rd party plugin by mistake? > > > > The suggested approach allows us to don't bother if a CompletableFuture > has > > to be wrapped or not. > > > > > > On Tue, Mar 30, 2021 at 12:22 PM Alexey Goncharuk < > > alexey.goncha...@gmail.com> wrote: > > > > > Ivan, > > > > > > My concern with the concept of a user completing the future returned > from > > > Ignite public API is that it is unclear how to interpret this action > > (this > > > backs Val's message). > > > Let's say a user started a compute with fut = compute.runAsync(task); > and > > > now calls fut.complete(someVal); Does this mean that Ignite no longer > > needs > > > to execute the task? If the task is currently running, does it need to > be > > > canceled? > > > > > > Using CompletableFuture.anyOf() is a good instrument in this case > because > > > it makes the 'first future wins' contract explicit in the code. Besides > > > that, the point regarding the cancel() method is valid, and we will > need > > > some custom mechanics to cancel a computation, so a custom interface > that > > > simply extends both Future and CompletableStage seems reasonable to me. > > > > > > --AG > > > > > > пн, 29 мар. 2021 г. в 09:12, Ivan Pavlukhin : > > > > > > > Val, > > > > > > > > There were enough hype around Reactive programming past years. I > > > > remind a lot of talks about RxJava. And I suppose it worth to > consider > > > > it. But it requires some time to study modern trends to make a > choice. > > > > So far I am not ready to facilitate Reactive API for Ignite 3. > > > > > > > > Regarding CompletableFuture. > > > > > > > > > The point is that currently a future returned from any of Ignite's > > > async > > > > > operations is supposed to be completed with a value only by Ignite > > > > itself, > > > > > not by the user. If we follow the same approach in Ignite 3, > > returning > > > > > CompletableFuture is surely wrong in my view. > > > > > > > > My first thoughts was similar. But later I thought what a user would > > > > like do with returned future. And one of cases I imagined was a case > > > > of alternative result. E.g. a user uses Ignite and another data > source > > > > in his application. He wants to use a value arrived faster. He > > > > combines 2 futures like CompletableFuture.anyOf(...). Consequently > > > > even if we prohibit CompletableFuture.complete(...)
Re: [DISCUSSION] IgniteFuture class future in Ignite-3.0.
Hello! > Let's say a user started a compute with fut = compute.runAsync(task); > and now calls fut.complete(someVal); Does this mean that Ignite no longer needs to execute the task? > If the task is currently running, does it need to be canceled? Yes, this case looks like Ignite should cancel computations because a user wants to complete the future. Why not? If there will be an opportunity to cancel a future, why is it a bad option to finish a future through a complete() method? > If you look at Ignite-2 code, you may found a number of places where we return e.g. exchange futures or partition release futures. > Assume the impact if we will return CompletableFuture instead, which can be completed in 3-rd party plugin by mistake? If exchange futures or partition release futures can be returned to 3-rd party plugin by mistake, it is poor encapsulation. And if it will be IgniteFuter rather than CompletedFuture, anyway, this can harm. вт, 30 мар. 2021 г. в 13:14, Andrey Mashenkov : > Guys, > > I want to remember there is one more point to pay attention to. > Extending Future and CompletableStage is more than just prevents unexpected > behavior if a user completed the future. > > First of all, it helps us to write safer code as we won't a method contract > exposed such methods as to a user as to a developer. > If you look at Ignite-2 code, you may found a number of places where we > return e.g. exchange futures or partition release futures. > Assume the impact if we will return CompletableFuture instead, which can be > completed in 3-rd party plugin by mistake? > > The suggested approach allows us to don't bother if a CompletableFuture has > to be wrapped or not. > > > On Tue, Mar 30, 2021 at 12:22 PM Alexey Goncharuk < > alexey.goncha...@gmail.com> wrote: > > > Ivan, > > > > My concern with the concept of a user completing the future returned from > > Ignite public API is that it is unclear how to interpret this action > (this > > backs Val's message). > > Let's say a user started a compute with fut = compute.runAsync(task); and > > now calls fut.complete(someVal); Does this mean that Ignite no longer > needs > > to execute the task? If the task is currently running, does it need to be > > canceled? > > > > Using CompletableFuture.anyOf() is a good instrument in this case because > > it makes the 'first future wins' contract explicit in the code. Besides > > that, the point regarding the cancel() method is valid, and we will need > > some custom mechanics to cancel a computation, so a custom interface that > > simply extends both Future and CompletableStage seems reasonable to me. > > > > --AG > > > > пн, 29 мар. 2021 г. в 09:12, Ivan Pavlukhin : > > > > > Val, > > > > > > There were enough hype around Reactive programming past years. I > > > remind a lot of talks about RxJava. And I suppose it worth to consider > > > it. But it requires some time to study modern trends to make a choice. > > > So far I am not ready to facilitate Reactive API for Ignite 3. > > > > > > Regarding CompletableFuture. > > > > > > > The point is that currently a future returned from any of Ignite's > > async > > > > operations is supposed to be completed with a value only by Ignite > > > itself, > > > > not by the user. If we follow the same approach in Ignite 3, > returning > > > > CompletableFuture is surely wrong in my view. > > > > > > My first thoughts was similar. But later I thought what a user would > > > like do with returned future. And one of cases I imagined was a case > > > of alternative result. E.g. a user uses Ignite and another data source > > > in his application. He wants to use a value arrived faster. He > > > combines 2 futures like CompletableFuture.anyOf(...). Consequently > > > even if we prohibit CompletableFuture.complete(...) explicitly then it > > > will be possible to create a combination that will allow premature > > > future completion. After all generally CompletableFuture is a > > > placeholder for async computaion result and if a user wants to > > > substitute result returned from Ignite why should we disallow him to > > > do it? > > > > > > Also I found one more suspicious thing with CompletableFuture. As it > > > is a concrete class it implements a cancel() method. And as I see the > > > implementation does not try to cancel underlying computations. Is not > > > it a problem? > > > > > > 2021-03-29 7:30 GMT+03:00, Valentin Kulichenko < > > > valentin.kuliche...@gmail.com>: > > > > Ivan, > > > > > > > > It's not really about the "harm", but more about "what should we do > if > > > this > > > > method is called?". Imagine the following code: > > > > > > > > CompletableFuture fut = cache.getAsync(key); > > > > fut.complete("something"); > > > > > > > > What should happen in this case? > > > > > > > > The point is that currently a future returned from any of Ignite's > > async > > > > operations is supposed to be completed with a value only by Ignite > > > itself, > > > > not by the user. If we
Re: [DISCUSSION] IgniteFuture class future in Ignite-3.0.
Guys, I want to remember there is one more point to pay attention to. Extending Future and CompletableStage is more than just prevents unexpected behavior if a user completed the future. First of all, it helps us to write safer code as we won't a method contract exposed such methods as to a user as to a developer. If you look at Ignite-2 code, you may found a number of places where we return e.g. exchange futures or partition release futures. Assume the impact if we will return CompletableFuture instead, which can be completed in 3-rd party plugin by mistake? The suggested approach allows us to don't bother if a CompletableFuture has to be wrapped or not. On Tue, Mar 30, 2021 at 12:22 PM Alexey Goncharuk < alexey.goncha...@gmail.com> wrote: > Ivan, > > My concern with the concept of a user completing the future returned from > Ignite public API is that it is unclear how to interpret this action (this > backs Val's message). > Let's say a user started a compute with fut = compute.runAsync(task); and > now calls fut.complete(someVal); Does this mean that Ignite no longer needs > to execute the task? If the task is currently running, does it need to be > canceled? > > Using CompletableFuture.anyOf() is a good instrument in this case because > it makes the 'first future wins' contract explicit in the code. Besides > that, the point regarding the cancel() method is valid, and we will need > some custom mechanics to cancel a computation, so a custom interface that > simply extends both Future and CompletableStage seems reasonable to me. > > --AG > > пн, 29 мар. 2021 г. в 09:12, Ivan Pavlukhin : > > > Val, > > > > There were enough hype around Reactive programming past years. I > > remind a lot of talks about RxJava. And I suppose it worth to consider > > it. But it requires some time to study modern trends to make a choice. > > So far I am not ready to facilitate Reactive API for Ignite 3. > > > > Regarding CompletableFuture. > > > > > The point is that currently a future returned from any of Ignite's > async > > > operations is supposed to be completed with a value only by Ignite > > itself, > > > not by the user. If we follow the same approach in Ignite 3, returning > > > CompletableFuture is surely wrong in my view. > > > > My first thoughts was similar. But later I thought what a user would > > like do with returned future. And one of cases I imagined was a case > > of alternative result. E.g. a user uses Ignite and another data source > > in his application. He wants to use a value arrived faster. He > > combines 2 futures like CompletableFuture.anyOf(...). Consequently > > even if we prohibit CompletableFuture.complete(...) explicitly then it > > will be possible to create a combination that will allow premature > > future completion. After all generally CompletableFuture is a > > placeholder for async computaion result and if a user wants to > > substitute result returned from Ignite why should we disallow him to > > do it? > > > > Also I found one more suspicious thing with CompletableFuture. As it > > is a concrete class it implements a cancel() method. And as I see the > > implementation does not try to cancel underlying computations. Is not > > it a problem? > > > > 2021-03-29 7:30 GMT+03:00, Valentin Kulichenko < > > valentin.kuliche...@gmail.com>: > > > Ivan, > > > > > > It's not really about the "harm", but more about "what should we do if > > this > > > method is called?". Imagine the following code: > > > > > > CompletableFuture fut = cache.getAsync(key); > > > fut.complete("something"); > > > > > > What should happen in this case? > > > > > > The point is that currently a future returned from any of Ignite's > async > > > operations is supposed to be completed with a value only by Ignite > > itself, > > > not by the user. If we follow the same approach in Ignite 3, returning > > > CompletableFuture is surely wrong in my view. > > > > > > At the same time, if we take a fundamentally different route with the > > async > > > APIs, this whole discussion might become irrelevant. For example, can > you > > > elaborate on your thinking around the reactive API? Do you have any > > > specifics in mind? > > > > > > -Val > > > > > > On Sat, Mar 27, 2021 at 9:18 PM Ivan Pavlukhin > > wrote: > > > > > >> > The methods below shouldn't be accessible for user: > > >> > complete() > > >> > completeExceptionaly() > > >> > > >> Folks, in case of user-facing API, do you think there is a real harm > > >> in allowing a user to manually "complete" a future? I suppose a user > > >> employs some post-processing for future results and potentially wants > > >> to have control of these results as well. E.g. premature completion in > > >> case when a result is no longer needed is possible usage. > > >> > > >> Also I thinkg it might be a good time to ponder about Future/Promise > > >> APIs in general. Why such API is our choice? Can we choose e.g. > > >> Reactive API style instead? > > >> > > >> 2021-03-27 0:33
Re: [DISCUSSION] IgniteFuture class future in Ignite-3.0.
Ivan, My concern with the concept of a user completing the future returned from Ignite public API is that it is unclear how to interpret this action (this backs Val's message). Let's say a user started a compute with fut = compute.runAsync(task); and now calls fut.complete(someVal); Does this mean that Ignite no longer needs to execute the task? If the task is currently running, does it need to be canceled? Using CompletableFuture.anyOf() is a good instrument in this case because it makes the 'first future wins' contract explicit in the code. Besides that, the point regarding the cancel() method is valid, and we will need some custom mechanics to cancel a computation, so a custom interface that simply extends both Future and CompletableStage seems reasonable to me. --AG пн, 29 мар. 2021 г. в 09:12, Ivan Pavlukhin : > Val, > > There were enough hype around Reactive programming past years. I > remind a lot of talks about RxJava. And I suppose it worth to consider > it. But it requires some time to study modern trends to make a choice. > So far I am not ready to facilitate Reactive API for Ignite 3. > > Regarding CompletableFuture. > > > The point is that currently a future returned from any of Ignite's async > > operations is supposed to be completed with a value only by Ignite > itself, > > not by the user. If we follow the same approach in Ignite 3, returning > > CompletableFuture is surely wrong in my view. > > My first thoughts was similar. But later I thought what a user would > like do with returned future. And one of cases I imagined was a case > of alternative result. E.g. a user uses Ignite and another data source > in his application. He wants to use a value arrived faster. He > combines 2 futures like CompletableFuture.anyOf(...). Consequently > even if we prohibit CompletableFuture.complete(...) explicitly then it > will be possible to create a combination that will allow premature > future completion. After all generally CompletableFuture is a > placeholder for async computaion result and if a user wants to > substitute result returned from Ignite why should we disallow him to > do it? > > Also I found one more suspicious thing with CompletableFuture. As it > is a concrete class it implements a cancel() method. And as I see the > implementation does not try to cancel underlying computations. Is not > it a problem? > > 2021-03-29 7:30 GMT+03:00, Valentin Kulichenko < > valentin.kuliche...@gmail.com>: > > Ivan, > > > > It's not really about the "harm", but more about "what should we do if > this > > method is called?". Imagine the following code: > > > > CompletableFuture fut = cache.getAsync(key); > > fut.complete("something"); > > > > What should happen in this case? > > > > The point is that currently a future returned from any of Ignite's async > > operations is supposed to be completed with a value only by Ignite > itself, > > not by the user. If we follow the same approach in Ignite 3, returning > > CompletableFuture is surely wrong in my view. > > > > At the same time, if we take a fundamentally different route with the > async > > APIs, this whole discussion might become irrelevant. For example, can you > > elaborate on your thinking around the reactive API? Do you have any > > specifics in mind? > > > > -Val > > > > On Sat, Mar 27, 2021 at 9:18 PM Ivan Pavlukhin > wrote: > > > >> > The methods below shouldn't be accessible for user: > >> > complete() > >> > completeExceptionaly() > >> > >> Folks, in case of user-facing API, do you think there is a real harm > >> in allowing a user to manually "complete" a future? I suppose a user > >> employs some post-processing for future results and potentially wants > >> to have control of these results as well. E.g. premature completion in > >> case when a result is no longer needed is possible usage. > >> > >> Also I thinkg it might be a good time to ponder about Future/Promise > >> APIs in general. Why such API is our choice? Can we choose e.g. > >> Reactive API style instead? > >> > >> 2021-03-27 0:33 GMT+03:00, Valentin Kulichenko < > >> valentin.kuliche...@gmail.com>: > >> > Andrey, > >> > > >> > I see. So in a nutshell, you're saying that we want to return a future > >> that > >> > the user's code is not allowed to complete. In this case, I think it's > >> > clear that CompletableFuture is not what we need. We actually need a > >> > NonCompletableFuture :) > >> > > >> > My vote is for the custom interface. > >> > > >> > -Val > >> > > >> > On Fri, Mar 26, 2021 at 2:25 AM Andrey Mashenkov > >> > > >> > wrote: > >> > > >> >> Val, > >> >> > >> >> The methods below shouldn't be accessible for user: > >> >> complete() > >> >> completeExceptionaly() > >> >> > >> >> Returning CompletableFuture we must always make a copy to prevent the > >> >> original future from being completed by mistake. > >> >> I think it will NOT be enough to do that returing the future to the > >> >> end-user, but from every critical module to the outside of the > module,
Re: [DISCUSSION] IgniteFuture class future in Ignite-3.0.
Val, There were enough hype around Reactive programming past years. I remind a lot of talks about RxJava. And I suppose it worth to consider it. But it requires some time to study modern trends to make a choice. So far I am not ready to facilitate Reactive API for Ignite 3. Regarding CompletableFuture. > The point is that currently a future returned from any of Ignite's async > operations is supposed to be completed with a value only by Ignite itself, > not by the user. If we follow the same approach in Ignite 3, returning > CompletableFuture is surely wrong in my view. My first thoughts was similar. But later I thought what a user would like do with returned future. And one of cases I imagined was a case of alternative result. E.g. a user uses Ignite and another data source in his application. He wants to use a value arrived faster. He combines 2 futures like CompletableFuture.anyOf(...). Consequently even if we prohibit CompletableFuture.complete(...) explicitly then it will be possible to create a combination that will allow premature future completion. After all generally CompletableFuture is a placeholder for async computaion result and if a user wants to substitute result returned from Ignite why should we disallow him to do it? Also I found one more suspicious thing with CompletableFuture. As it is a concrete class it implements a cancel() method. And as I see the implementation does not try to cancel underlying computations. Is not it a problem? 2021-03-29 7:30 GMT+03:00, Valentin Kulichenko : > Ivan, > > It's not really about the "harm", but more about "what should we do if this > method is called?". Imagine the following code: > > CompletableFuture fut = cache.getAsync(key); > fut.complete("something"); > > What should happen in this case? > > The point is that currently a future returned from any of Ignite's async > operations is supposed to be completed with a value only by Ignite itself, > not by the user. If we follow the same approach in Ignite 3, returning > CompletableFuture is surely wrong in my view. > > At the same time, if we take a fundamentally different route with the async > APIs, this whole discussion might become irrelevant. For example, can you > elaborate on your thinking around the reactive API? Do you have any > specifics in mind? > > -Val > > On Sat, Mar 27, 2021 at 9:18 PM Ivan Pavlukhin wrote: > >> > The methods below shouldn't be accessible for user: >> > complete() >> > completeExceptionaly() >> >> Folks, in case of user-facing API, do you think there is a real harm >> in allowing a user to manually "complete" a future? I suppose a user >> employs some post-processing for future results and potentially wants >> to have control of these results as well. E.g. premature completion in >> case when a result is no longer needed is possible usage. >> >> Also I thinkg it might be a good time to ponder about Future/Promise >> APIs in general. Why such API is our choice? Can we choose e.g. >> Reactive API style instead? >> >> 2021-03-27 0:33 GMT+03:00, Valentin Kulichenko < >> valentin.kuliche...@gmail.com>: >> > Andrey, >> > >> > I see. So in a nutshell, you're saying that we want to return a future >> that >> > the user's code is not allowed to complete. In this case, I think it's >> > clear that CompletableFuture is not what we need. We actually need a >> > NonCompletableFuture :) >> > >> > My vote is for the custom interface. >> > >> > -Val >> > >> > On Fri, Mar 26, 2021 at 2:25 AM Andrey Mashenkov >> > >> > wrote: >> > >> >> Val, >> >> >> >> The methods below shouldn't be accessible for user: >> >> complete() >> >> completeExceptionaly() >> >> >> >> Returning CompletableFuture we must always make a copy to prevent the >> >> original future from being completed by mistake. >> >> I think it will NOT be enough to do that returing the future to the >> >> end-user, but from every critical module to the outside of the module, >> >> e.g. to plugins. The impact of disclosing ExchangeFuture, >> >> PartitionReleaseFuture to plugins may be serious. >> >> >> >> IgniteFuture extends Future, CompletionStage which >> >> implementation >> >> will just wrap CompletableFuture these issues will be resolved in >> natural >> >> way. >> >> In addition we can force toCompletableFuture() method to return a >> >> defensive >> >> copy(), that resolves the last concern. >> >> >> >> >> >> On Fri, Mar 26, 2021 at 11:38 AM Konstantin Orlov >> >> >> >> wrote: >> >> >> >> > CompletableFuture seems a better option to me. >> >> > >> >> > -- >> >> > Regards, >> >> > Konstantin Orlov >> >> > >> >> > >> >> > >> >> > >> >> > > On 26 Mar 2021, at 11:07, Pavel Tupitsyn >> >> > > wrote: >> >> > > >> >> > > On the one hand, I agree with Alexey. >> >> > > CompletableFuture has complete* methods which should not be >> available >> >> to >> >> > > the user code. >> >> > > This can be solved with a simple interface like we do in Thin >> Client: >> >> > > IgniteClientFuture extends Future, CompletionStage >> >>
Re: [DISCUSSION] IgniteFuture class future in Ignite-3.0.
Ivan, It's not really about the "harm", but more about "what should we do if this method is called?". Imagine the following code: CompletableFuture fut = cache.getAsync(key); fut.complete("something"); What should happen in this case? The point is that currently a future returned from any of Ignite's async operations is supposed to be completed with a value only by Ignite itself, not by the user. If we follow the same approach in Ignite 3, returning CompletableFuture is surely wrong in my view. At the same time, if we take a fundamentally different route with the async APIs, this whole discussion might become irrelevant. For example, can you elaborate on your thinking around the reactive API? Do you have any specifics in mind? -Val On Sat, Mar 27, 2021 at 9:18 PM Ivan Pavlukhin wrote: > > The methods below shouldn't be accessible for user: > > complete() > > completeExceptionaly() > > Folks, in case of user-facing API, do you think there is a real harm > in allowing a user to manually "complete" a future? I suppose a user > employs some post-processing for future results and potentially wants > to have control of these results as well. E.g. premature completion in > case when a result is no longer needed is possible usage. > > Also I thinkg it might be a good time to ponder about Future/Promise > APIs in general. Why such API is our choice? Can we choose e.g. > Reactive API style instead? > > 2021-03-27 0:33 GMT+03:00, Valentin Kulichenko < > valentin.kuliche...@gmail.com>: > > Andrey, > > > > I see. So in a nutshell, you're saying that we want to return a future > that > > the user's code is not allowed to complete. In this case, I think it's > > clear that CompletableFuture is not what we need. We actually need a > > NonCompletableFuture :) > > > > My vote is for the custom interface. > > > > -Val > > > > On Fri, Mar 26, 2021 at 2:25 AM Andrey Mashenkov > > > > wrote: > > > >> Val, > >> > >> The methods below shouldn't be accessible for user: > >> complete() > >> completeExceptionaly() > >> > >> Returning CompletableFuture we must always make a copy to prevent the > >> original future from being completed by mistake. > >> I think it will NOT be enough to do that returing the future to the > >> end-user, but from every critical module to the outside of the module, > >> e.g. to plugins. The impact of disclosing ExchangeFuture, > >> PartitionReleaseFuture to plugins may be serious. > >> > >> IgniteFuture extends Future, CompletionStage which > >> implementation > >> will just wrap CompletableFuture these issues will be resolved in > natural > >> way. > >> In addition we can force toCompletableFuture() method to return a > >> defensive > >> copy(), that resolves the last concern. > >> > >> > >> On Fri, Mar 26, 2021 at 11:38 AM Konstantin Orlov > >> wrote: > >> > >> > CompletableFuture seems a better option to me. > >> > > >> > -- > >> > Regards, > >> > Konstantin Orlov > >> > > >> > > >> > > >> > > >> > > On 26 Mar 2021, at 11:07, Pavel Tupitsyn > >> > > wrote: > >> > > > >> > > On the one hand, I agree with Alexey. > >> > > CompletableFuture has complete* methods which should not be > available > >> to > >> > > the user code. > >> > > This can be solved with a simple interface like we do in Thin > Client: > >> > > IgniteClientFuture extends Future, CompletionStage > >> > > > >> > > > >> > > On the other hand: > >> > > - CompletionStage has toCompletableFuture anyway (rather weird) > >> > > - Other libraries use CompletableFuture and it seems to be fine > >> > > - Using CompletableFuture is the simplest approach > >> > > > >> > > > >> > > So I lean towards CompletableFuture too. > >> > > > >> > > On Fri, Mar 26, 2021 at 10:46 AM Alexey Kukushkin < > >> > kukushkinale...@gmail.com> > >> > > wrote: > >> > > > >> > >> I do not like Java's CompletableFuture and prefer our own Future > >> > (revised > >> > >> IgniteFuture). > >> > >> > >> > >> My understanding of the Future (or Promise) pattern in general is > >> having > >> > >> two separate APIs: > >> > >> > >> > >> 1. Server-side: create, set result, raise error, cancel from > >> > >> server. > >> > >> 2. Client-side: get result, handle error, cancel from client > >> > >> > >> > >> Java's CompletableFuture looks like both the client-side and > >> > >> server-side API. The "Completeable" prefix in the name is already > >> > confusing > >> > >> for a client since it cannot "complete" an operation, only a server > >> can. > >> > >> > >> > >> I would create our own IgniteFuture adding client-side > functionality > >> we > >> > >> currently miss (like client-side cancellation). > >> > >> > >> > >> > >> > >> пт, 26 мар. 2021 г. в 01:08, Valentin Kulichenko < > >> > >> valentin.kuliche...@gmail.com>: > >> > >> > >> > >>> Andrey, > >> > >>> > >> > >>> Can you compile a full list of these risky methods, and elaborate > >> > >>> on > >> > what > >> > >>> the risks are? > >> > >>> > >> > >>> Generally, CompletableFuture is a much better option, because
Re: [DISCUSSION] IgniteFuture class future in Ignite-3.0.
> The methods below shouldn't be accessible for user: > complete() > completeExceptionaly() Folks, in case of user-facing API, do you think there is a real harm in allowing a user to manually "complete" a future? I suppose a user employs some post-processing for future results and potentially wants to have control of these results as well. E.g. premature completion in case when a result is no longer needed is possible usage. Also I thinkg it might be a good time to ponder about Future/Promise APIs in general. Why such API is our choice? Can we choose e.g. Reactive API style instead? 2021-03-27 0:33 GMT+03:00, Valentin Kulichenko : > Andrey, > > I see. So in a nutshell, you're saying that we want to return a future that > the user's code is not allowed to complete. In this case, I think it's > clear that CompletableFuture is not what we need. We actually need a > NonCompletableFuture :) > > My vote is for the custom interface. > > -Val > > On Fri, Mar 26, 2021 at 2:25 AM Andrey Mashenkov > > wrote: > >> Val, >> >> The methods below shouldn't be accessible for user: >> complete() >> completeExceptionaly() >> >> Returning CompletableFuture we must always make a copy to prevent the >> original future from being completed by mistake. >> I think it will NOT be enough to do that returing the future to the >> end-user, but from every critical module to the outside of the module, >> e.g. to plugins. The impact of disclosing ExchangeFuture, >> PartitionReleaseFuture to plugins may be serious. >> >> IgniteFuture extends Future, CompletionStage which >> implementation >> will just wrap CompletableFuture these issues will be resolved in natural >> way. >> In addition we can force toCompletableFuture() method to return a >> defensive >> copy(), that resolves the last concern. >> >> >> On Fri, Mar 26, 2021 at 11:38 AM Konstantin Orlov >> wrote: >> >> > CompletableFuture seems a better option to me. >> > >> > -- >> > Regards, >> > Konstantin Orlov >> > >> > >> > >> > >> > > On 26 Mar 2021, at 11:07, Pavel Tupitsyn >> > > wrote: >> > > >> > > On the one hand, I agree with Alexey. >> > > CompletableFuture has complete* methods which should not be available >> to >> > > the user code. >> > > This can be solved with a simple interface like we do in Thin Client: >> > > IgniteClientFuture extends Future, CompletionStage >> > > >> > > >> > > On the other hand: >> > > - CompletionStage has toCompletableFuture anyway (rather weird) >> > > - Other libraries use CompletableFuture and it seems to be fine >> > > - Using CompletableFuture is the simplest approach >> > > >> > > >> > > So I lean towards CompletableFuture too. >> > > >> > > On Fri, Mar 26, 2021 at 10:46 AM Alexey Kukushkin < >> > kukushkinale...@gmail.com> >> > > wrote: >> > > >> > >> I do not like Java's CompletableFuture and prefer our own Future >> > (revised >> > >> IgniteFuture). >> > >> >> > >> My understanding of the Future (or Promise) pattern in general is >> having >> > >> two separate APIs: >> > >> >> > >> 1. Server-side: create, set result, raise error, cancel from >> > >> server. >> > >> 2. Client-side: get result, handle error, cancel from client >> > >> >> > >> Java's CompletableFuture looks like both the client-side and >> > >> server-side API. The "Completeable" prefix in the name is already >> > confusing >> > >> for a client since it cannot "complete" an operation, only a server >> can. >> > >> >> > >> I would create our own IgniteFuture adding client-side functionality >> we >> > >> currently miss (like client-side cancellation). >> > >> >> > >> >> > >> пт, 26 мар. 2021 г. в 01:08, Valentin Kulichenko < >> > >> valentin.kuliche...@gmail.com>: >> > >> >> > >>> Andrey, >> > >>> >> > >>> Can you compile a full list of these risky methods, and elaborate >> > >>> on >> > what >> > >>> the risks are? >> > >>> >> > >>> Generally, CompletableFuture is a much better option, because it's >> > >>> standard. But we need to make sure it actually fits our needs and >> > doesn't >> > >>> do more harm than good. >> > >>> >> > >>> -Val >> > >>> >> > >>> On Thu, Mar 25, 2021 at 12:23 PM Alexei Scherbakov < >> > >>> alexey.scherbak...@gmail.com> wrote: >> > >>> >> > I think both options are fine, but personally lean toward >> > CompletableFuture. >> > >> > чт, 25 мар. 2021 г. в 17:56, Atri Sharma : >> > >> > > I would suggest using CompletableFuture -- I don't see a need for >> > > a >> > custom >> > > interface that is unique to us. >> > > >> > > It also allows a lower barrier for new contributors for >> understanding >> > > existing code >> > > >> > > On Thu, 25 Mar 2021, 20:18 Andrey Mashenkov, < >> > >>> andrey.mashen...@gmail.com >> > > >> > > wrote: >> > > >> > >> Hi Igniters, >> > >> >> > >> I'd like to start a discussion about replacing our custom >> > >>> IgniteFuture >> > >> class with CompletableFuture - existed JDK class >> > >> or rework it's
Re: [DISCUSSION] IgniteFuture class future in Ignite-3.0.
Andrey, I see. So in a nutshell, you're saying that we want to return a future that the user's code is not allowed to complete. In this case, I think it's clear that CompletableFuture is not what we need. We actually need a NonCompletableFuture :) My vote is for the custom interface. -Val On Fri, Mar 26, 2021 at 2:25 AM Andrey Mashenkov wrote: > Val, > > The methods below shouldn't be accessible for user: > complete() > completeExceptionaly() > > Returning CompletableFuture we must always make a copy to prevent the > original future from being completed by mistake. > I think it will NOT be enough to do that returing the future to the > end-user, but from every critical module to the outside of the module, > e.g. to plugins. The impact of disclosing ExchangeFuture, > PartitionReleaseFuture to plugins may be serious. > > IgniteFuture extends Future, CompletionStage which implementation > will just wrap CompletableFuture these issues will be resolved in natural > way. > In addition we can force toCompletableFuture() method to return a defensive > copy(), that resolves the last concern. > > > On Fri, Mar 26, 2021 at 11:38 AM Konstantin Orlov > wrote: > > > CompletableFuture seems a better option to me. > > > > -- > > Regards, > > Konstantin Orlov > > > > > > > > > > > On 26 Mar 2021, at 11:07, Pavel Tupitsyn wrote: > > > > > > On the one hand, I agree with Alexey. > > > CompletableFuture has complete* methods which should not be available > to > > > the user code. > > > This can be solved with a simple interface like we do in Thin Client: > > > IgniteClientFuture extends Future, CompletionStage > > > > > > > > > On the other hand: > > > - CompletionStage has toCompletableFuture anyway (rather weird) > > > - Other libraries use CompletableFuture and it seems to be fine > > > - Using CompletableFuture is the simplest approach > > > > > > > > > So I lean towards CompletableFuture too. > > > > > > On Fri, Mar 26, 2021 at 10:46 AM Alexey Kukushkin < > > kukushkinale...@gmail.com> > > > wrote: > > > > > >> I do not like Java's CompletableFuture and prefer our own Future > > (revised > > >> IgniteFuture). > > >> > > >> My understanding of the Future (or Promise) pattern in general is > having > > >> two separate APIs: > > >> > > >> 1. Server-side: create, set result, raise error, cancel from server. > > >> 2. Client-side: get result, handle error, cancel from client > > >> > > >> Java's CompletableFuture looks like both the client-side and > > >> server-side API. The "Completeable" prefix in the name is already > > confusing > > >> for a client since it cannot "complete" an operation, only a server > can. > > >> > > >> I would create our own IgniteFuture adding client-side functionality > we > > >> currently miss (like client-side cancellation). > > >> > > >> > > >> пт, 26 мар. 2021 г. в 01:08, Valentin Kulichenko < > > >> valentin.kuliche...@gmail.com>: > > >> > > >>> Andrey, > > >>> > > >>> Can you compile a full list of these risky methods, and elaborate on > > what > > >>> the risks are? > > >>> > > >>> Generally, CompletableFuture is a much better option, because it's > > >>> standard. But we need to make sure it actually fits our needs and > > doesn't > > >>> do more harm than good. > > >>> > > >>> -Val > > >>> > > >>> On Thu, Mar 25, 2021 at 12:23 PM Alexei Scherbakov < > > >>> alexey.scherbak...@gmail.com> wrote: > > >>> > > I think both options are fine, but personally lean toward > > CompletableFuture. > > > > чт, 25 мар. 2021 г. в 17:56, Atri Sharma : > > > > > I would suggest using CompletableFuture -- I don't see a need for a > > custom > > > interface that is unique to us. > > > > > > It also allows a lower barrier for new contributors for > understanding > > > existing code > > > > > > On Thu, 25 Mar 2021, 20:18 Andrey Mashenkov, < > > >>> andrey.mashen...@gmail.com > > > > > > wrote: > > > > > >> Hi Igniters, > > >> > > >> I'd like to start a discussion about replacing our custom > > >>> IgniteFuture > > >> class with CompletableFuture - existed JDK class > > >> or rework it's implementation (like some other products done) to a > > >> composition of CompletionStage and Future interfaces. > > >> or maybe other option if you have any ideas. Do you? > > >> > > >> 1. The first approach pros and cons are > > >> + Well-known JDK class > > >> + Already implemented > > >> - It is a class, not an interface. > > >> - Expose some potentially harmful methods like "complete()". > > >> > > >> On the other side, it has copy() method to create defensive copy > > >> and > > >> minimalCompletionStage() to restrict harmful method usage. > > >> Thus, this look like an applicable solution, but we should be > > >> careful > > >> exposing internal future to the outside. > > >> > > >> 2. The second approach is to implement our own interface like the > > >>> next >
Re: [DISCUSSION] IgniteFuture class future in Ignite-3.0.
Val, The methods below shouldn't be accessible for user: complete() completeExceptionaly() Returning CompletableFuture we must always make a copy to prevent the original future from being completed by mistake. I think it will NOT be enough to do that returing the future to the end-user, but from every critical module to the outside of the module, e.g. to plugins. The impact of disclosing ExchangeFuture, PartitionReleaseFuture to plugins may be serious. IgniteFuture extends Future, CompletionStage which implementation will just wrap CompletableFuture these issues will be resolved in natural way. In addition we can force toCompletableFuture() method to return a defensive copy(), that resolves the last concern. On Fri, Mar 26, 2021 at 11:38 AM Konstantin Orlov wrote: > CompletableFuture seems a better option to me. > > -- > Regards, > Konstantin Orlov > > > > > > On 26 Mar 2021, at 11:07, Pavel Tupitsyn wrote: > > > > On the one hand, I agree with Alexey. > > CompletableFuture has complete* methods which should not be available to > > the user code. > > This can be solved with a simple interface like we do in Thin Client: > > IgniteClientFuture extends Future, CompletionStage > > > > > > On the other hand: > > - CompletionStage has toCompletableFuture anyway (rather weird) > > - Other libraries use CompletableFuture and it seems to be fine > > - Using CompletableFuture is the simplest approach > > > > > > So I lean towards CompletableFuture too. > > > > On Fri, Mar 26, 2021 at 10:46 AM Alexey Kukushkin < > kukushkinale...@gmail.com> > > wrote: > > > >> I do not like Java's CompletableFuture and prefer our own Future > (revised > >> IgniteFuture). > >> > >> My understanding of the Future (or Promise) pattern in general is having > >> two separate APIs: > >> > >> 1. Server-side: create, set result, raise error, cancel from server. > >> 2. Client-side: get result, handle error, cancel from client > >> > >> Java's CompletableFuture looks like both the client-side and > >> server-side API. The "Completeable" prefix in the name is already > confusing > >> for a client since it cannot "complete" an operation, only a server can. > >> > >> I would create our own IgniteFuture adding client-side functionality we > >> currently miss (like client-side cancellation). > >> > >> > >> пт, 26 мар. 2021 г. в 01:08, Valentin Kulichenko < > >> valentin.kuliche...@gmail.com>: > >> > >>> Andrey, > >>> > >>> Can you compile a full list of these risky methods, and elaborate on > what > >>> the risks are? > >>> > >>> Generally, CompletableFuture is a much better option, because it's > >>> standard. But we need to make sure it actually fits our needs and > doesn't > >>> do more harm than good. > >>> > >>> -Val > >>> > >>> On Thu, Mar 25, 2021 at 12:23 PM Alexei Scherbakov < > >>> alexey.scherbak...@gmail.com> wrote: > >>> > I think both options are fine, but personally lean toward > CompletableFuture. > > чт, 25 мар. 2021 г. в 17:56, Atri Sharma : > > > I would suggest using CompletableFuture -- I don't see a need for a > custom > > interface that is unique to us. > > > > It also allows a lower barrier for new contributors for understanding > > existing code > > > > On Thu, 25 Mar 2021, 20:18 Andrey Mashenkov, < > >>> andrey.mashen...@gmail.com > > > > wrote: > > > >> Hi Igniters, > >> > >> I'd like to start a discussion about replacing our custom > >>> IgniteFuture > >> class with CompletableFuture - existed JDK class > >> or rework it's implementation (like some other products done) to a > >> composition of CompletionStage and Future interfaces. > >> or maybe other option if you have any ideas. Do you? > >> > >> 1. The first approach pros and cons are > >> + Well-known JDK class > >> + Already implemented > >> - It is a class, not an interface. > >> - Expose some potentially harmful methods like "complete()". > >> > >> On the other side, it has copy() method to create defensive copy > >> and > >> minimalCompletionStage() to restrict harmful method usage. > >> Thus, this look like an applicable solution, but we should be > >> careful > >> exposing internal future to the outside. > >> > >> 2. The second approach is to implement our own interface like the > >>> next > > one: > >> > >> interface IgniteFuture extends CompletableStage, Future { > >> } > >> > >> Pros and cons are > >> + Our interfaces/classes contracts will expose an interface rather > >>> than > >> concrete implementation. > >> + All methods are safe. > >> - Some implementation is required. > >> - CompletableStage has a method toCompletableFuture() and can be > > converted > >> to CompletableFuture. This should be supported. > >> > >> However, we still could wrap CompletableFuture and don't bother > >> about > >> creating a defensive copy. > >> >
Re: [DISCUSSION] IgniteFuture class future in Ignite-3.0.
CompletableFuture seems a better option to me. -- Regards, Konstantin Orlov > On 26 Mar 2021, at 11:07, Pavel Tupitsyn wrote: > > On the one hand, I agree with Alexey. > CompletableFuture has complete* methods which should not be available to > the user code. > This can be solved with a simple interface like we do in Thin Client: > IgniteClientFuture extends Future, CompletionStage > > > On the other hand: > - CompletionStage has toCompletableFuture anyway (rather weird) > - Other libraries use CompletableFuture and it seems to be fine > - Using CompletableFuture is the simplest approach > > > So I lean towards CompletableFuture too. > > On Fri, Mar 26, 2021 at 10:46 AM Alexey Kukushkin > wrote: > >> I do not like Java's CompletableFuture and prefer our own Future (revised >> IgniteFuture). >> >> My understanding of the Future (or Promise) pattern in general is having >> two separate APIs: >> >> 1. Server-side: create, set result, raise error, cancel from server. >> 2. Client-side: get result, handle error, cancel from client >> >> Java's CompletableFuture looks like both the client-side and >> server-side API. The "Completeable" prefix in the name is already confusing >> for a client since it cannot "complete" an operation, only a server can. >> >> I would create our own IgniteFuture adding client-side functionality we >> currently miss (like client-side cancellation). >> >> >> пт, 26 мар. 2021 г. в 01:08, Valentin Kulichenko < >> valentin.kuliche...@gmail.com>: >> >>> Andrey, >>> >>> Can you compile a full list of these risky methods, and elaborate on what >>> the risks are? >>> >>> Generally, CompletableFuture is a much better option, because it's >>> standard. But we need to make sure it actually fits our needs and doesn't >>> do more harm than good. >>> >>> -Val >>> >>> On Thu, Mar 25, 2021 at 12:23 PM Alexei Scherbakov < >>> alexey.scherbak...@gmail.com> wrote: >>> I think both options are fine, but personally lean toward CompletableFuture. чт, 25 мар. 2021 г. в 17:56, Atri Sharma : > I would suggest using CompletableFuture -- I don't see a need for a custom > interface that is unique to us. > > It also allows a lower barrier for new contributors for understanding > existing code > > On Thu, 25 Mar 2021, 20:18 Andrey Mashenkov, < >>> andrey.mashen...@gmail.com > > wrote: > >> Hi Igniters, >> >> I'd like to start a discussion about replacing our custom >>> IgniteFuture >> class with CompletableFuture - existed JDK class >> or rework it's implementation (like some other products done) to a >> composition of CompletionStage and Future interfaces. >> or maybe other option if you have any ideas. Do you? >> >> 1. The first approach pros and cons are >> + Well-known JDK class >> + Already implemented >> - It is a class, not an interface. >> - Expose some potentially harmful methods like "complete()". >> >> On the other side, it has copy() method to create defensive copy >> and >> minimalCompletionStage() to restrict harmful method usage. >> Thus, this look like an applicable solution, but we should be >> careful >> exposing internal future to the outside. >> >> 2. The second approach is to implement our own interface like the >>> next > one: >> >> interface IgniteFuture extends CompletableStage, Future { >> } >> >> Pros and cons are >> + Our interfaces/classes contracts will expose an interface rather >>> than >> concrete implementation. >> + All methods are safe. >> - Some implementation is required. >> - CompletableStage has a method toCompletableFuture() and can be > converted >> to CompletableFuture. This should be supported. >> >> However, we still could wrap CompletableFuture and don't bother >> about >> creating a defensive copy. >> >> >> Other project experience: >> * Spotify uses CompletableFuture directly [1]. >> * Redis goes the second approach [2] >> * Vertx explicitly extends CompletableFuture [3]. However, they >> have > custom >> future classes and a number of helpers that could be replaced with >> CompletableStage. Maybe it is just a legacy.' >> >> Any thoughts? >> >> [1] >> >> > >>> >> https://spotify.github.io/completable-futures/apidocs/com/spotify/futures/ConcurrencyReducer.html >> [2] >> >> > >>> >> https://lettuce.io/lettuce-4/release/api/com/lambdaworks/redis/RedisFuture.html >> [3] >> >> > >>> >> https://javadoc.io/static/org.jspare.vertx/vertx-jspare/1.1.0-M03/org/jspare/vertx/concurrent/VertxCompletableFuture.html >> -- >> Best regards, >> Andrey V. Mashenkov >> > -- Best regards, Alexei Scherbakov >>> >> >> >> -- >> Best regards, >>
Re: [DISCUSSION] IgniteFuture class future in Ignite-3.0.
On the one hand, I agree with Alexey. CompletableFuture has complete* methods which should not be available to the user code. This can be solved with a simple interface like we do in Thin Client: IgniteClientFuture extends Future, CompletionStage On the other hand: - CompletionStage has toCompletableFuture anyway (rather weird) - Other libraries use CompletableFuture and it seems to be fine - Using CompletableFuture is the simplest approach So I lean towards CompletableFuture too. On Fri, Mar 26, 2021 at 10:46 AM Alexey Kukushkin wrote: > I do not like Java's CompletableFuture and prefer our own Future (revised > IgniteFuture). > > My understanding of the Future (or Promise) pattern in general is having > two separate APIs: > >1. Server-side: create, set result, raise error, cancel from server. >2. Client-side: get result, handle error, cancel from client > > Java's CompletableFuture looks like both the client-side and > server-side API. The "Completeable" prefix in the name is already confusing > for a client since it cannot "complete" an operation, only a server can. > > I would create our own IgniteFuture adding client-side functionality we > currently miss (like client-side cancellation). > > > пт, 26 мар. 2021 г. в 01:08, Valentin Kulichenko < > valentin.kuliche...@gmail.com>: > > > Andrey, > > > > Can you compile a full list of these risky methods, and elaborate on what > > the risks are? > > > > Generally, CompletableFuture is a much better option, because it's > > standard. But we need to make sure it actually fits our needs and doesn't > > do more harm than good. > > > > -Val > > > > On Thu, Mar 25, 2021 at 12:23 PM Alexei Scherbakov < > > alexey.scherbak...@gmail.com> wrote: > > > > > I think both options are fine, but personally lean toward > > > CompletableFuture. > > > > > > чт, 25 мар. 2021 г. в 17:56, Atri Sharma : > > > > > > > I would suggest using CompletableFuture -- I don't see a need for a > > > custom > > > > interface that is unique to us. > > > > > > > > It also allows a lower barrier for new contributors for understanding > > > > existing code > > > > > > > > On Thu, 25 Mar 2021, 20:18 Andrey Mashenkov, < > > andrey.mashen...@gmail.com > > > > > > > > wrote: > > > > > > > > > Hi Igniters, > > > > > > > > > > I'd like to start a discussion about replacing our custom > > IgniteFuture > > > > > class with CompletableFuture - existed JDK class > > > > > or rework it's implementation (like some other products done) to a > > > > > composition of CompletionStage and Future interfaces. > > > > > or maybe other option if you have any ideas. Do you? > > > > > > > > > > 1. The first approach pros and cons are > > > > > + Well-known JDK class > > > > > + Already implemented > > > > > - It is a class, not an interface. > > > > > - Expose some potentially harmful methods like "complete()". > > > > > > > > > > On the other side, it has copy() method to create defensive copy > and > > > > > minimalCompletionStage() to restrict harmful method usage. > > > > > Thus, this look like an applicable solution, but we should be > careful > > > > > exposing internal future to the outside. > > > > > > > > > > 2. The second approach is to implement our own interface like the > > next > > > > one: > > > > > > > > > > interface IgniteFuture extends CompletableStage, Future { > > > > > } > > > > > > > > > > Pros and cons are > > > > > + Our interfaces/classes contracts will expose an interface rather > > than > > > > > concrete implementation. > > > > > + All methods are safe. > > > > > - Some implementation is required. > > > > > - CompletableStage has a method toCompletableFuture() and can be > > > > converted > > > > > to CompletableFuture. This should be supported. > > > > > > > > > > However, we still could wrap CompletableFuture and don't bother > about > > > > > creating a defensive copy. > > > > > > > > > > > > > > > Other project experience: > > > > > * Spotify uses CompletableFuture directly [1]. > > > > > * Redis goes the second approach [2] > > > > > * Vertx explicitly extends CompletableFuture [3]. However, they > have > > > > custom > > > > > future classes and a number of helpers that could be replaced with > > > > > CompletableStage. Maybe it is just a legacy.' > > > > > > > > > > Any thoughts? > > > > > > > > > > [1] > > > > > > > > > > > > > > > > > > > > https://spotify.github.io/completable-futures/apidocs/com/spotify/futures/ConcurrencyReducer.html > > > > > [2] > > > > > > > > > > > > > > > > > > > > https://lettuce.io/lettuce-4/release/api/com/lambdaworks/redis/RedisFuture.html > > > > > [3] > > > > > > > > > > > > > > > > > > > > https://javadoc.io/static/org.jspare.vertx/vertx-jspare/1.1.0-M03/org/jspare/vertx/concurrent/VertxCompletableFuture.html > > > > > -- > > > > > Best regards, > > > > > Andrey V. Mashenkov > > > > > > > > > > > > > > > > > > -- > > > > > > Best regards, > > > Alexei Scherbakov > > > > > > > > -- > Best regards, > Alexey >
Re: [DISCUSSION] IgniteFuture class future in Ignite-3.0.
I do not like Java's CompletableFuture and prefer our own Future (revised IgniteFuture). My understanding of the Future (or Promise) pattern in general is having two separate APIs: 1. Server-side: create, set result, raise error, cancel from server. 2. Client-side: get result, handle error, cancel from client Java's CompletableFuture looks like both the client-side and server-side API. The "Completeable" prefix in the name is already confusing for a client since it cannot "complete" an operation, only a server can. I would create our own IgniteFuture adding client-side functionality we currently miss (like client-side cancellation). пт, 26 мар. 2021 г. в 01:08, Valentin Kulichenko < valentin.kuliche...@gmail.com>: > Andrey, > > Can you compile a full list of these risky methods, and elaborate on what > the risks are? > > Generally, CompletableFuture is a much better option, because it's > standard. But we need to make sure it actually fits our needs and doesn't > do more harm than good. > > -Val > > On Thu, Mar 25, 2021 at 12:23 PM Alexei Scherbakov < > alexey.scherbak...@gmail.com> wrote: > > > I think both options are fine, but personally lean toward > > CompletableFuture. > > > > чт, 25 мар. 2021 г. в 17:56, Atri Sharma : > > > > > I would suggest using CompletableFuture -- I don't see a need for a > > custom > > > interface that is unique to us. > > > > > > It also allows a lower barrier for new contributors for understanding > > > existing code > > > > > > On Thu, 25 Mar 2021, 20:18 Andrey Mashenkov, < > andrey.mashen...@gmail.com > > > > > > wrote: > > > > > > > Hi Igniters, > > > > > > > > I'd like to start a discussion about replacing our custom > IgniteFuture > > > > class with CompletableFuture - existed JDK class > > > > or rework it's implementation (like some other products done) to a > > > > composition of CompletionStage and Future interfaces. > > > > or maybe other option if you have any ideas. Do you? > > > > > > > > 1. The first approach pros and cons are > > > > + Well-known JDK class > > > > + Already implemented > > > > - It is a class, not an interface. > > > > - Expose some potentially harmful methods like "complete()". > > > > > > > > On the other side, it has copy() method to create defensive copy and > > > > minimalCompletionStage() to restrict harmful method usage. > > > > Thus, this look like an applicable solution, but we should be careful > > > > exposing internal future to the outside. > > > > > > > > 2. The second approach is to implement our own interface like the > next > > > one: > > > > > > > > interface IgniteFuture extends CompletableStage, Future { > > > > } > > > > > > > > Pros and cons are > > > > + Our interfaces/classes contracts will expose an interface rather > than > > > > concrete implementation. > > > > + All methods are safe. > > > > - Some implementation is required. > > > > - CompletableStage has a method toCompletableFuture() and can be > > > converted > > > > to CompletableFuture. This should be supported. > > > > > > > > However, we still could wrap CompletableFuture and don't bother about > > > > creating a defensive copy. > > > > > > > > > > > > Other project experience: > > > > * Spotify uses CompletableFuture directly [1]. > > > > * Redis goes the second approach [2] > > > > * Vertx explicitly extends CompletableFuture [3]. However, they have > > > custom > > > > future classes and a number of helpers that could be replaced with > > > > CompletableStage. Maybe it is just a legacy.' > > > > > > > > Any thoughts? > > > > > > > > [1] > > > > > > > > > > > > > > https://spotify.github.io/completable-futures/apidocs/com/spotify/futures/ConcurrencyReducer.html > > > > [2] > > > > > > > > > > > > > > https://lettuce.io/lettuce-4/release/api/com/lambdaworks/redis/RedisFuture.html > > > > [3] > > > > > > > > > > > > > > https://javadoc.io/static/org.jspare.vertx/vertx-jspare/1.1.0-M03/org/jspare/vertx/concurrent/VertxCompletableFuture.html > > > > -- > > > > Best regards, > > > > Andrey V. Mashenkov > > > > > > > > > > > > > -- > > > > Best regards, > > Alexei Scherbakov > > > -- Best regards, Alexey
Re: [DISCUSSION] IgniteFuture class future in Ignite-3.0.
Andrey, Can you compile a full list of these risky methods, and elaborate on what the risks are? Generally, CompletableFuture is a much better option, because it's standard. But we need to make sure it actually fits our needs and doesn't do more harm than good. -Val On Thu, Mar 25, 2021 at 12:23 PM Alexei Scherbakov < alexey.scherbak...@gmail.com> wrote: > I think both options are fine, but personally lean toward > CompletableFuture. > > чт, 25 мар. 2021 г. в 17:56, Atri Sharma : > > > I would suggest using CompletableFuture -- I don't see a need for a > custom > > interface that is unique to us. > > > > It also allows a lower barrier for new contributors for understanding > > existing code > > > > On Thu, 25 Mar 2021, 20:18 Andrey Mashenkov, > > > wrote: > > > > > Hi Igniters, > > > > > > I'd like to start a discussion about replacing our custom IgniteFuture > > > class with CompletableFuture - existed JDK class > > > or rework it's implementation (like some other products done) to a > > > composition of CompletionStage and Future interfaces. > > > or maybe other option if you have any ideas. Do you? > > > > > > 1. The first approach pros and cons are > > > + Well-known JDK class > > > + Already implemented > > > - It is a class, not an interface. > > > - Expose some potentially harmful methods like "complete()". > > > > > > On the other side, it has copy() method to create defensive copy and > > > minimalCompletionStage() to restrict harmful method usage. > > > Thus, this look like an applicable solution, but we should be careful > > > exposing internal future to the outside. > > > > > > 2. The second approach is to implement our own interface like the next > > one: > > > > > > interface IgniteFuture extends CompletableStage, Future { > > > } > > > > > > Pros and cons are > > > + Our interfaces/classes contracts will expose an interface rather than > > > concrete implementation. > > > + All methods are safe. > > > - Some implementation is required. > > > - CompletableStage has a method toCompletableFuture() and can be > > converted > > > to CompletableFuture. This should be supported. > > > > > > However, we still could wrap CompletableFuture and don't bother about > > > creating a defensive copy. > > > > > > > > > Other project experience: > > > * Spotify uses CompletableFuture directly [1]. > > > * Redis goes the second approach [2] > > > * Vertx explicitly extends CompletableFuture [3]. However, they have > > custom > > > future classes and a number of helpers that could be replaced with > > > CompletableStage. Maybe it is just a legacy.' > > > > > > Any thoughts? > > > > > > [1] > > > > > > > > > https://spotify.github.io/completable-futures/apidocs/com/spotify/futures/ConcurrencyReducer.html > > > [2] > > > > > > > > > https://lettuce.io/lettuce-4/release/api/com/lambdaworks/redis/RedisFuture.html > > > [3] > > > > > > > > > https://javadoc.io/static/org.jspare.vertx/vertx-jspare/1.1.0-M03/org/jspare/vertx/concurrent/VertxCompletableFuture.html > > > -- > > > Best regards, > > > Andrey V. Mashenkov > > > > > > > > -- > > Best regards, > Alexei Scherbakov >
Re: [DISCUSSION] IgniteFuture class future in Ignite-3.0.
I think both options are fine, but personally lean toward CompletableFuture. чт, 25 мар. 2021 г. в 17:56, Atri Sharma : > I would suggest using CompletableFuture -- I don't see a need for a custom > interface that is unique to us. > > It also allows a lower barrier for new contributors for understanding > existing code > > On Thu, 25 Mar 2021, 20:18 Andrey Mashenkov, > wrote: > > > Hi Igniters, > > > > I'd like to start a discussion about replacing our custom IgniteFuture > > class with CompletableFuture - existed JDK class > > or rework it's implementation (like some other products done) to a > > composition of CompletionStage and Future interfaces. > > or maybe other option if you have any ideas. Do you? > > > > 1. The first approach pros and cons are > > + Well-known JDK class > > + Already implemented > > - It is a class, not an interface. > > - Expose some potentially harmful methods like "complete()". > > > > On the other side, it has copy() method to create defensive copy and > > minimalCompletionStage() to restrict harmful method usage. > > Thus, this look like an applicable solution, but we should be careful > > exposing internal future to the outside. > > > > 2. The second approach is to implement our own interface like the next > one: > > > > interface IgniteFuture extends CompletableStage, Future { > > } > > > > Pros and cons are > > + Our interfaces/classes contracts will expose an interface rather than > > concrete implementation. > > + All methods are safe. > > - Some implementation is required. > > - CompletableStage has a method toCompletableFuture() and can be > converted > > to CompletableFuture. This should be supported. > > > > However, we still could wrap CompletableFuture and don't bother about > > creating a defensive copy. > > > > > > Other project experience: > > * Spotify uses CompletableFuture directly [1]. > > * Redis goes the second approach [2] > > * Vertx explicitly extends CompletableFuture [3]. However, they have > custom > > future classes and a number of helpers that could be replaced with > > CompletableStage. Maybe it is just a legacy.' > > > > Any thoughts? > > > > [1] > > > > > https://spotify.github.io/completable-futures/apidocs/com/spotify/futures/ConcurrencyReducer.html > > [2] > > > > > https://lettuce.io/lettuce-4/release/api/com/lambdaworks/redis/RedisFuture.html > > [3] > > > > > https://javadoc.io/static/org.jspare.vertx/vertx-jspare/1.1.0-M03/org/jspare/vertx/concurrent/VertxCompletableFuture.html > > -- > > Best regards, > > Andrey V. Mashenkov > > > -- Best regards, Alexei Scherbakov
Re: [DISCUSSION] IgniteFuture class future in Ignite-3.0.
I would suggest using CompletableFuture -- I don't see a need for a custom interface that is unique to us. It also allows a lower barrier for new contributors for understanding existing code On Thu, 25 Mar 2021, 20:18 Andrey Mashenkov, wrote: > Hi Igniters, > > I'd like to start a discussion about replacing our custom IgniteFuture > class with CompletableFuture - existed JDK class > or rework it's implementation (like some other products done) to a > composition of CompletionStage and Future interfaces. > or maybe other option if you have any ideas. Do you? > > 1. The first approach pros and cons are > + Well-known JDK class > + Already implemented > - It is a class, not an interface. > - Expose some potentially harmful methods like "complete()". > > On the other side, it has copy() method to create defensive copy and > minimalCompletionStage() to restrict harmful method usage. > Thus, this look like an applicable solution, but we should be careful > exposing internal future to the outside. > > 2. The second approach is to implement our own interface like the next one: > > interface IgniteFuture extends CompletableStage, Future { > } > > Pros and cons are > + Our interfaces/classes contracts will expose an interface rather than > concrete implementation. > + All methods are safe. > - Some implementation is required. > - CompletableStage has a method toCompletableFuture() and can be converted > to CompletableFuture. This should be supported. > > However, we still could wrap CompletableFuture and don't bother about > creating a defensive copy. > > > Other project experience: > * Spotify uses CompletableFuture directly [1]. > * Redis goes the second approach [2] > * Vertx explicitly extends CompletableFuture [3]. However, they have custom > future classes and a number of helpers that could be replaced with > CompletableStage. Maybe it is just a legacy.' > > Any thoughts? > > [1] > > https://spotify.github.io/completable-futures/apidocs/com/spotify/futures/ConcurrencyReducer.html > [2] > > https://lettuce.io/lettuce-4/release/api/com/lambdaworks/redis/RedisFuture.html > [3] > > https://javadoc.io/static/org.jspare.vertx/vertx-jspare/1.1.0-M03/org/jspare/vertx/concurrent/VertxCompletableFuture.html > -- > Best regards, > Andrey V. Mashenkov >
[DISCUSSION] IgniteFuture class future in Ignite-3.0.
Hi Igniters, I'd like to start a discussion about replacing our custom IgniteFuture class with CompletableFuture - existed JDK class or rework it's implementation (like some other products done) to a composition of CompletionStage and Future interfaces. or maybe other option if you have any ideas. Do you? 1. The first approach pros and cons are + Well-known JDK class + Already implemented - It is a class, not an interface. - Expose some potentially harmful methods like "complete()". On the other side, it has copy() method to create defensive copy and minimalCompletionStage() to restrict harmful method usage. Thus, this look like an applicable solution, but we should be careful exposing internal future to the outside. 2. The second approach is to implement our own interface like the next one: interface IgniteFuture extends CompletableStage, Future { } Pros and cons are + Our interfaces/classes contracts will expose an interface rather than concrete implementation. + All methods are safe. - Some implementation is required. - CompletableStage has a method toCompletableFuture() and can be converted to CompletableFuture. This should be supported. However, we still could wrap CompletableFuture and don't bother about creating a defensive copy. Other project experience: * Spotify uses CompletableFuture directly [1]. * Redis goes the second approach [2] * Vertx explicitly extends CompletableFuture [3]. However, they have custom future classes and a number of helpers that could be replaced with CompletableStage. Maybe it is just a legacy.' Any thoughts? [1] https://spotify.github.io/completable-futures/apidocs/com/spotify/futures/ConcurrencyReducer.html [2] https://lettuce.io/lettuce-4/release/api/com/lambdaworks/redis/RedisFuture.html [3] https://javadoc.io/static/org.jspare.vertx/vertx-jspare/1.1.0-M03/org/jspare/vertx/concurrent/VertxCompletableFuture.html -- Best regards, Andrey V. Mashenkov