[
https://issues.apache.org/jira/browse/FLINK-4113?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15348356#comment-15348356
]
Greg Hogan commented on FLINK-4113:
-----------------------------------
I experienced the problem when verifying the
[HITS|https://github.com/apache/flink/pull/1967/files#diff-5af706cf3a08013de4ff97a440d2845bR194]
algorithm. With object reuse enabled the {{ChainedAllReduceDriver}} was saving
a reference to the first record emitted by my map function. My map then
overwrote the object and the same reference was passed as the second value to
my reduce function.
> Always copy first value in ChainedAllReduceDriver
> -------------------------------------------------
>
> Key: FLINK-4113
> URL: https://issues.apache.org/jira/browse/FLINK-4113
> Project: Flink
> Issue Type: Bug
> Components: Local Runtime
> Affects Versions: 1.1.0, 1.0.4
> Reporter: Greg Hogan
> Assignee: Greg Hogan
> Priority: Critical
> Fix For: 1.1.0, 1.0.4
>
>
> {{ChainedAllReduceDriver.collect}} must copy the first record even when
> object reuse is enabled or {{base}} may later point to the same object as
> {{record}}.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
