[infinispan-issues] [JBoss JIRA] (ISPN-10373) Store/Loader Non blocking SPI

[Will Burns (Jira)]

    [ https://issues.redhat.com/browse/ISPN-10373?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14095753#comment-14095753 ] 

Will Burns commented on ISPN-10373:
-----------------------------------

Just so it is not lost this is how we can implement RemoteStore bulk operations

//   @Override
//   public CompletionStage<Void> bulkSegmentedWrite(int publisherCount,
//         Publisher<NonBlockingLoadWriteStore.SegmentedPublisher<MarshallableEntry<K, V>>> publisher) {
//      return Flowable.fromPublisher(publisher)
//            .flatMap(Functions.identity(), false, publisherCount)
//            .buffer(configuration.maxBatchSize())
//            .concatMapCompletable(mes -> {
//               CompletableFuture<Void> putAllFuture = remoteCache.putAllAsync(
//                     mes.stream().collect(Collectors.toMap(this::getKey, this::getValue)));
//               return RxJavaInterop.completionStageToCompletable(putAllFuture);
//            })
//            .doOnError(PersistenceException::new)
//            .to(RxJavaInterop.completableToCompletionStage());
//   }

//   @Override
//   public CompletionStage<Void> bulkSegmentedWrite(int publisherCount,
//         Publisher<NonBlockingStore.SegmentedPublisher<MarshallableEntry<K, V>>> publisher) {
//      return Flowable.fromPublisher(publisher)
//            .flatMap(sp ->
//               Flowable.fromPublisher(sp)
//                     .buffer(configuration.maxBatchSize())
//            , false, publisherCount)
//            .concatMapCompletable(mes -> {
//               CompletableFuture<Void> putAllFuture = remoteCache.putAllAsync(
//                     mes.stream().collect(Collectors.toMap(this::getKey, this::getValue)));
//               return RxJavaInterop.completionStageToCompletable(putAllFuture);
//            })
//            .doOnError(PersistenceException::new)
//            .to(RxJavaInterop.completableToCompletionStage());
//   }

> Store/Loader Non blocking SPI
> -----------------------------
>
>                 Key: ISPN-10373
>                 URL: https://issues.redhat.com/browse/ISPN-10373
>             Project: Infinispan
>          Issue Type: Feature Request
>          Components: Loaders and Stores
>            Reporter: Will Burns
>            Assignee: Will Burns
>            Priority: Major
>
> We need to add and use a non blocking SPI internally for our stores/loaders. We added ISPN-9722, which is a good step and refactored all of our internal code to use "non blocking" stores. However the stores themselves are all inherently sync even if the store itself could be non blocking. We would have to add a new SPI interface to allow for such non blocking operations. We would then remove all the explicit threading added in ISPN-9722 and move it to a wrapper around a currently sync loader instead. This way an invoking thread doesn't need to do a context switch or anything if invoking just a non blocking store operation.



--
This message was sent by Atlassian Jira
(v7.13.8#713008)