You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by "Jingsong Lee (Jira)" <ji...@apache.org> on 2019/11/04 12:40:00 UTC
[jira] [Comment Edited] (FLINK-14243) flink hiveudf needs some
check when it is using cache
[ https://issues.apache.org/jira/browse/FLINK-14243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16966632#comment-16966632 ]
Jingsong Lee edited comment on FLINK-14243 at 11/4/19 12:39 PM:
----------------------------------------------------------------
Hi [~jackylau] Can you share the code of GetJsonObject? I think UDFJson is is thread-safe too. What I know it is not thread-safe in the past until:
https://issues.apache.org/jira/browse/HIVE-16196
[https://github.com/apache/hive/commit/87be4b31ce5abbe03ee8461a437c901b5ee9ed05]
So it has been fixed to be thread-safe.
Hive has Tez runner and spark runner too. I think every hive functions should be thread-safe.
Can you check again?
was (Author: lzljs3620320):
Hi [~jackylau] Can you share the code of GetJsonObject? I think UDFJson is is thread-safe too. What I know it is not thread-safe in the past until:
https://issues.apache.org/jira/browse/HIVE-16196
[https://github.com/apache/hive/commit/87be4b31ce5abbe03ee8461a437c901b5ee9ed05]
So it has been fixed because of thread-not-safe.
Hive has Tez runner and spark runner too. I think every hive functions should be thread-safe.
Can you check again?
> flink hiveudf needs some check when it is using cache
> -----------------------------------------------------
>
> Key: FLINK-14243
> URL: https://issues.apache.org/jira/browse/FLINK-14243
> Project: Flink
> Issue Type: Bug
> Components: Connectors / Hive, Table SQL / Planner
> Affects Versions: 1.9.0
> Reporter: jackylau
> Priority: Major
> Fix For: 1.10.0
>
> Attachments: Snipaste_2019-10-30_15-34-09.png
>
>
> Flink1.9 brings in hive connector, but it will have some problem when the original hive udf using cache. We konw that hive is processed level parallel based on jvm, while flink/spark is task level parallel. If flink just calls the hive udf, it wll exists thread-safe problem when using cache.
> So it may need check the hive udf code and if it is not thread-safe, and set the flink parallize=1
--
This message was sent by Atlassian Jira
(v8.3.4#803005)