[infinispan-issues] [JBoss JIRA] (ISPN-6424) Distributed Executor get can propagate exception if interrupt returns before cancellation

[William Burns (JIRA)]

    [ https://issues.jboss.org/browse/ISPN-6424?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13184028#comment-13184028 ] 

William Burns commented on ISPN-6424:
-------------------------------------

[~dan.berindei] I was fully aware of that exception, and it is only part of the issue.

The core problem is when the following happens

# n1 userthread - distributed executor submits a task
# n1 userthread - task is interrupted
# n1 userthread - distributed executor sends cancellation command to remote node
# n2 remote thread - interrupts task and send back the exception you just mentioned
# n1 remote thread - receives exception response
# n1 remote thread - sets exception status on future
# n1 user thread tries to interrupt future but cannot since it was completed with exception

Unless you are saying to know somehow in the n2 remote thread that it was caused due solely to an interrupt and instead send back some kind of cancelled response?  But in that case we will mess up other exception call stacks if the cache was actually shut down.

Thus my change is to make sure to set the cancelled status on the future before we even send the remote cancellation which prevents this timing issue.

> Distributed Executor get can propagate exception if interrupt returns before cancellation
> -----------------------------------------------------------------------------------------
>
>                 Key: ISPN-6424
>                 URL: https://issues.jboss.org/browse/ISPN-6424
>             Project: Infinispan
>          Issue Type: Bug
>          Components: Distributed Execution and Map/Reduce
>            Reporter: William Burns
>            Assignee: William Burns
>             Fix For: 9.0.0.Alpha1
>
>
> This can cause get method to return an exception of the operation being interrupted instead of a CancellationException as it should.
> This causes a test failure http://ci.infinispan.org/viewLog.html?buildId=37756&tab=buildResultsDiv&buildTypeId=Infinispan_MasterHotspotJdk8#testNameId-3420236380418240325



--
This message was sent by Atlassian JIRA
(v6.4.11#64026)