[ 
https://issues.apache.org/jira/browse/EDGENT-231?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15437226#comment-15437226
 ] 

Dale LaBossiere commented on EDGENT-231:
----------------------------------------

Perhaps there's also value in tuple-event-time ordered count-based windows - 
e.g., with an analogous API like
```
<K> TWindow<T, K> last(int count, Function<T, K> keyFunction, ToLongFunction<T> 
timestampFn);
```

> Add event-time time-based windows
> ---------------------------------
>
>                 Key: EDGENT-231
>                 URL: https://issues.apache.org/jira/browse/EDGENT-231
>             Project: Edgent
>          Issue Type: New Feature
>            Reporter: Dale LaBossiere
>
> TStream.last(period,unit,keyFn) creates an insertion-time time-based TWindow 
> - a tuple's timestamp, hence its lifetime in the time-based window, is based 
> on the time it is inserted into the window.
> On a couple of occasions some folks have asked about event-time time-based 
> windows - a tuple's timestamp (event-time) is present as a property in the 
> tuple and that timestamp is used for insertion and eviction processing.  
> Tuples need not be added to the window in their event-time order though the 
> partition contents are maintained in event-time order.
> The TStream API for such a time-based window could be:
> ```
>     <K> TWindow<T, K> last(long time, TimeUnit unit, Function<T, K> 
> keyFunction, ToLongFunction<T> timestampFn);
> ```
> See also: QUARKS-230



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to