[
https://issues.apache.org/jira/browse/TINKERPOP-1674?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16014024#comment-16014024
]
SmedbergM commented on TINKERPOP-1674:
--------------------------------------
I did indeed have to do a lot to reliably make this happen. (The code is
abstracted and significantly simplified from how we originally observed the
problem in production, using an async HTTP server library.) I'm not sure how
much work it would be to reproduce using less machinery; as the logs show, it
can be hundreds of transactions before a stale reference is encountered.
> Traversals reference elements after deletion
> --------------------------------------------
>
> Key: TINKERPOP-1674
> URL: https://issues.apache.org/jira/browse/TINKERPOP-1674
> Project: TinkerPop
> Issue Type: Bug
> Components: neo4j
> Affects Versions: 3.2.3, 3.2.4
> Environment: Demonstrated on Ubuntu, OSX
> Reporter: SmedbergM
>
> In a multiprocessor environment, a traversal will sometimes reference
> vertices which another thread/processor has already deleted. This causes the
> entire traversal to fail in an unrecoverable fashion and throw an uncaught
> exception.
> MWE: https://github.com/SmedbergM/neo4j-deletion-error
> Gist containing logging output:
> https://gist.github.com/SmedbergM/5fcf0d98a255e7d346b85b98bcc1ec0d
> This error has cropped up persistently over several months/releases (dating
> back to 3.1.x or earlier). I have not tried to diff back/bisect to find
> introduction time of the behavior.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
