[
https://issues.apache.org/jira/browse/FLINK-2870?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15125979#comment-15125979
]
Fabian Hueske commented on FLINK-2870:
--------------------------------------
[~shikhar], the late-arrival strategy for default time windows
({{stream.timeWindow(...)}}) is to invoke the window function on the single
late-arriving element and forward the result.
If you implement your own windowing logic, the behavior can be customized in
many ways. You can
1) keep the whole window for a specific time (like 1 day) and recompute the
function on the whole data (this requires enough memory),
2) discard late arrivers
3) compute delta's to the last emitted element
4) ...
Have a look at this [blog
post|http://flink.apache.org/news/2015/12/04/Introducing-windows.html] which
describes the window and trigger interfaces.
> Add support for accumulating/discarding for Event-Time Windows
> --------------------------------------------------------------
>
> Key: FLINK-2870
> URL: https://issues.apache.org/jira/browse/FLINK-2870
> Project: Flink
> Issue Type: Sub-task
> Components: Streaming
> Reporter: Aljoscha Krettek
> Assignee: Aljoscha Krettek
>
> This would allow to specify whether windows should be discarded after the
> trigger fires or kept in the operator if late elements arrive.
> When keeping elements, the user would also have to specify an allowed
> lateness time after which the window contents are discarded without emitting
> any further window evaluation result.
> If elements arrive after the allowed lateness they would trigger the window
> immediately with only the one single element.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
