[
https://issues.apache.org/jira/browse/FLINK-15607?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17016461#comment-17016461
]
Jark Wu commented on FLINK-15607:
---------------------------------
I discussed with [~lzljs3620320] about this yesterday and learned that the Hive
UDAF is supported in streaming mode (the accumulator will be stored in state).
But it seems that we dont' have tests for this.
> throw exception when users trying to use Hive aggregate functions in
> streaming mode
> -----------------------------------------------------------------------------------
>
> Key: FLINK-15607
> URL: https://issues.apache.org/jira/browse/FLINK-15607
> Project: Flink
> Issue Type: Improvement
> Components: Connectors / Hive, Table SQL / API
> Reporter: Bowen Li
> Priority: Major
>
> Due to that Hive functions are all built for batch processing, aggregate
> functions in Hive that implement UDAF and GenericUDAFResolver2 interfaces may
> have unpredictable behaviors when used in streaming mode in Flink. We advice
> users to only use Hive aggregate functions interfaces in batch mode.
>
> It will be better to throw exception when users trying to use Hive aggregate
> functions in streaming mode. Seems need to distinguish execution mode in
> FunctionCatalogOperatorTable, which is not achievable yet.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
