[
https://issues.apache.org/jira/browse/FLINK-5480?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15831990#comment-15831990
]
ASF GitHub Bot commented on FLINK-5480:
---------------------------------------
Github user uce commented on the issue:
https://github.com/apache/flink/pull/3117
Thanks for addressing the comments.
I had a quick offline chat with Stephan regarding the naming of
`setAdditionalNodeHash`. He also agrees that the name moves internal
implementation details to the user API. Furthermore he raised the point that
there is no way for the user to get the relation between `uid` and
`setAdditionalNodeHash`.
He proposed the following, which I liked.
- Rename `setAdditionalNodeHash` to `setUidHash`. Add a big fat comment
that this should be used as a work around or for trouble shooting.
- Add a comment to `uid` saying that the given String will be hashed to the
ID that corresponds to the JobVertexID in the logs/web UI so users now what's
happening.
- Make in both Javadocs clear how the methods relate to eachother.
What do you think?
> User-provided hashes for operators
> ----------------------------------
>
> Key: FLINK-5480
> URL: https://issues.apache.org/jira/browse/FLINK-5480
> Project: Flink
> Issue Type: Improvement
> Components: DataStream API
> Affects Versions: 1.2.0
> Reporter: Stefan Richter
> Assignee: Stefan Richter
>
> We could allow users to provided (alternative) hashes for operators in a
> StreamGraph. This can make migration between Flink versions easier, in case
> the automatically produced hashes between versions are incompatible. For
> example, users could just copy the old hashes from the web ui to their job.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
