[ 
https://issues.apache.org/jira/browse/FLINK-9857?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16545286#comment-16545286
 ] 

Aljoscha Krettek commented on FLINK-9857:
-----------------------------------------

On a side note: even if we change this off-by-one bug, I think there can still 
be races because current processing time is queried using 
System.currentTimeMillis() and we set timers using a 
ScheduledThreadPoolExecutor (currently). If there's any race between those two 
you can also get weird results.

For these reasons, I would always suggest to go with event time or ingestion 
time, but I think the latter is currently not possible with the Table API/SQL.

> Processing-time timers fire too early
> -------------------------------------
>
>                 Key: FLINK-9857
>                 URL: https://issues.apache.org/jira/browse/FLINK-9857
>             Project: Flink
>          Issue Type: Bug
>          Components: DataStream API
>    Affects Versions: 1.3.4, 1.4.2, 1.5.1, 1.6.0
>            Reporter: Aljoscha Krettek
>            Assignee: Aljoscha Krettek
>            Priority: Blocker
>             Fix For: 1.5.2, 1.6.0
>
>
> The firing of processing-time timers is off by one. This leads to problems in 
> edge cases, as discovered [here (mailing 
> list)|https://lists.apache.org/thread.html/e49748fa5fa1c9217b9dfb65eea7a37af1f2895c769528e77a1a93fa@%3Cuser.flink.apache.org%3E]
>  when elements arrive at the timestamp that is the end of the window.
> The problem is [here 
> (github)|https://github.com/apache/flink/blob/79b38f8f9a79b917d525842cf46087c5b8c40f3d/flink-streaming-java/src/main/java/org/apache/flink/streaming/api/operators/HeapInternalTimerService.java#L231].
>  For event-time, we fire timers when the watermark is >= the timestamp, this 
> is correct because a watermark T says that we will not see elements with a 
> timestamp smaller or equal to T. For processing time, a time of T does not 
> say that we won't see an element with timestamp T, which makes 
> processing-time timers fire one ms too early.
> I think we can fix it by turning that {{<=}} into a {{<}}.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Reply via email to