You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@ignite.apache.org by "Andrey Gura (JIRA)" <ji...@apache.org> on 2018/03/30 17:19:00 UTC

[jira] [Commented] (IGNITE-6186) Remove redundant parameter of GridFutureAdapter::unregisterWaiter()

    [ https://issues.apache.org/jira/browse/IGNITE-6186?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16420733#comment-16420733 ] 

Andrey Gura commented on IGNITE-6186:
-------------------------------------

[~andrey-kuznetsov] LGTM. But I renamed methid to {{unregisterWaiter}} because it is symmetrical to {{registerWaiter}}. Merged to master branch.

> Remove redundant parameter of GridFutureAdapter::unregisterWaiter()
> -------------------------------------------------------------------
>
>                 Key: IGNITE-6186
>                 URL: https://issues.apache.org/jira/browse/IGNITE-6186
>             Project: Ignite
>          Issue Type: Improvement
>    Affects Versions: 2.1
>            Reporter: Andrey Kuznetsov
>            Assignee: Andrey Kuznetsov
>            Priority: Minor
>             Fix For: 2.5
>
>
> The method is not thread-safe unless actual parameter is currentThread.
> Let future state is a list of listeners and two concurrent threads are removing two adjacent non-root listener nodes from list simultaneously by calling {{unregisterWaiter()}}. Then data race is possible: one of these listeners can survive its removal. If the listener is a thread waiting for completion in {{get0()}} then this race leads at worst case to 1 extra call to {{LockSupport.park()}}, and it's negligible. Otherwise we deal with an arbitrary listener, and its {{apply()}} will be called twice.
> To be precise, this Jira issue does not relate to any existing bug, but it eliminates fragile construct that can explode on future chages/refactorings.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)