[
https://issues.apache.org/jira/browse/FLINK-21365?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17284667#comment-17284667
]
Piotr Nowojski commented on FLINK-21365:
----------------------------------------
This is some implementation detail of either the CPU caching and/or JVM
implementation. Based on the provided guarantees in Java, we programmers do not
have to care about those. It doesn't matter if two threads are updating two
different variables, or a different position in some array. Both of those cases
are just two different writes to some memory address, between we (Java
developers) need to establish happens-before relationship.
The minor difference with the array case is that you need to have happens
before relationship also to the array variable and array index, instead of just
to a variable.
And practically speaking on x86, JVM will just insert memory fences
before/after touching `numCompleted`, that would flush all of the cached
writes/sync reads that code has done so far.
(*as far as I understand it)
> Visibility issue in FutureUtils.ResultConjunctFuture.handleCompletedFuture
> ---------------------------------------------------------------------------
>
> Key: FLINK-21365
> URL: https://issues.apache.org/jira/browse/FLINK-21365
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination
> Affects Versions: 1.11.3, 1.12.1
> Reporter: Roman Khachatryan
> Assignee: Roman Khachatryan
> Priority: Major
> Fix For: 1.11.4, 1.12.2, 1.13.0
>
>
> * FutureUtils.ResultConjunctFuture.handleCompletedFuture can update *results*
> array from multiple threads
> * The array is declared as volatile but this only means the reference is
> volatile, not the contents
> * There are no other guards
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
