[
https://issues.apache.org/jira/browse/FLINK-3714?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15327400#comment-15327400
]
ASF GitHub Bot commented on FLINK-3714:
---------------------------------------
Github user aljoscha commented on the issue:
https://github.com/apache/flink/pull/2093
I had some inline comments.
One other thing I would like to see changed is moving the logic that checks
whether to do cleanup from `cleanup()` to the call-site. So in
`processElement()` it would be:
```
if (combinedTriggerResult.isPurge) {
cleanup(...);
} else {
registerCleanupTimer(...);
}
```
The reason for this is that the check in `cleanup()` for whether a timer is
a cleanup timer does not take into account the event-time/processing-time
split: if a processing-time timer fires on exactly the cleanup time of an
event-time window this would also trigger cleanup. The code in
`processTriggersFor` (which, by the way can be moved back into
`processWatermark()` now that you refactored this special event-time trigger
hack) would have this check:
```
if (assigner.isEventTime() && isCleanupTimer(...))
```
while the code in `trigger()` would be
```
if (!assigner.isEventTime() && isCleanupTimer(...))
```
> Add Support for "Allowed Lateness"
> ----------------------------------
>
> Key: FLINK-3714
> URL: https://issues.apache.org/jira/browse/FLINK-3714
> Project: Flink
> Issue Type: Sub-task
> Components: Streaming
> Reporter: Aljoscha Krettek
> Assignee: Kostas Kloudas
>
> As mentioned in
> https://docs.google.com/document/d/1Xp-YBf87vLTduYSivgqWVEMjYUmkA-hyb4muX3KRl08/edit#
> we should add support for an allowed lateness setting.
> This includes several things:
> - API for setting allowed lateness
> - Dropping of late elements
> - Garbage collection of windows state/timers
> Depending on whether the {{WindowAssigner}} assigns windows based on event
> time or processing time we have to adjust the GC behavior. For event-time
> windows "allowed lateness" makes sense and we should garbage collect after
> this expires. For processing-time windows "allowed lateness" does not make
> sense and we should always GC window state/timers at the end timestamp of a
> processing-time window. I think that we need a method for this on
> {{WindowAssigner}} that allows to differentiate between event-time windows
> and processing-time windows: {{boolean WindowAssigner.isEventTime()}}.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
