[
https://issues.apache.org/jira/browse/FLINK-14608?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16969703#comment-16969703
]
Kurt Young commented on FLINK-14608:
------------------------------------
Hi [~dwysakowicz], I didn't do any benchmark but did some searching instead.
It's seems every benchmark I saw has the same conclusion is java streams is
slower than native array operations. The only question is how much slower. I've
seen 3-5 times as well as 10-20%.
Since these codes are executed not only per record, but even per field, I think
it would be a good idea to avoid java streams.
> avoid using Java Streams in JsonRowDeserializationSchema
> --------------------------------------------------------
>
> Key: FLINK-14608
> URL: https://issues.apache.org/jira/browse/FLINK-14608
> Project: Flink
> Issue Type: Improvement
> Components: Formats (JSON, Avro, Parquet, ORC, SequenceFile)
> Affects Versions: 1.10.0
> Reporter: Kurt Young
> Assignee: Jiayi Liao
> Priority: Major
> Labels: pull-request-available
> Fix For: 1.10.0
>
> Time Spent: 20m
> Remaining Estimate: 0h
>
> According to
> [https://flink.apache.org/contributing/code-style-and-quality-java.html], we
> should avoid using Java Streams in any performance critical code. Since this
> `DeserializationRuntimeConverter` will be called per field of each coming
> record, we should provide a non Java Streams implementation.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
