[
https://issues.apache.org/jira/browse/FLINK-27204?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17711722#comment-17711722
]
Wencong Liu commented on FLINK-27204:
-------------------------------------
In the implementation, I found it difficult to get the ioExecutor in the
Dispatcher when creating a new object for FileSystemJobResultStore. Therefore,
I changed the io operations in FileSystemJobResultStore to be based on
CompletableFuture, which also allows for asynchronous capabilities. The
interface of JobResultStore has also been modified. Please have a look. Thanks
:) [~mapohl]
> FileSystemJobResultStore should operator on the ioExecutor
> ----------------------------------------------------------
>
> Key: FLINK-27204
> URL: https://issues.apache.org/jira/browse/FLINK-27204
> Project: Flink
> Issue Type: Improvement
> Components: Runtime / Coordination
> Affects Versions: 1.16.0
> Reporter: Matthias Pohl
> Assignee: Wencong Liu
> Priority: Major
> Labels: pull-request-available
>
> The {{JobResultStore}} interface is synchronous currently. For the
> {{FileSystemJobResultStore}} this means that (possibly) IO-heavy operations
> have to be explicitly called moved to the ioExecutor within the Dispatcher.
> Instead, we could change the {{JobResultStore}} interface in a way that it
> returns {{CompletableFuture}} instances instead. That would enable us to run
> the {{FileSystemJobResultStore}} operations in the ioExecutor which would be
> set when initializing the {{{}FileSystemJobResultStore{}}}. This would move
> the responsibility of where to run the operation from the {{Dispatcher}} into
> the {{JobResultStore.}}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
