[
https://issues.apache.org/jira/browse/FLINK-29183?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17601316#comment-17601316
]
Weijie Guo commented on FLINK-29183:
------------------------------------
[~landlord] First of all, the phenomenon you observed is in line with
expectations, because this is the difference between batch and streaming
semantics. In batch mode, print needs to wait and sort all inputs, and then
output them together. In streaming mode, it is not required. Second,
BatchShuffleMode only affects the selection of your shuffle mode, which has
nothing to do with operator logic and semantics.
> flink job ended before window time up while using bounded stream source
> -----------------------------------------------------------------------
>
> Key: FLINK-29183
> URL: https://issues.apache.org/jira/browse/FLINK-29183
> Project: Flink
> Issue Type: Bug
> Components: API / DataStream
> Affects Versions: 1.15.2
> Reporter: Lu
> Priority: Major
> Attachments: image-2022-09-02-15-33-08-459.png,
> image-2022-09-02-15-35-18-590.png, image-2022-09-06-18-09-40-132.png,
> image-2022-09-06-18-09-43-745.png, image-2022-09-06-18-10-11-848.png,
> image-2022-09-06-18-11-02-915.png
>
>
> !image-2022-09-02-15-35-18-590.png|width=574,height=306!
>
> when i use a long time window while using bounded stream in STREAMING
> mode，the job shutdown before the window has been triggered and then i have
> nothing to print.
> if i use BATCH mode with the same code, it will work. But the batch shuffle
> is so expensive so i still prefer STREAMING.
> Is there any method could fix this problem? Or should i have mistake in my
> code?
> Looking forward to your reply. Thanks in advance.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
