flink-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "ASF GitHub Bot (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (FLINK-3109) Join two streams with two different buffer time
Date Sun, 31 Jan 2016 20:10:39 GMT

    [ https://issues.apache.org/jira/browse/FLINK-3109?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15125480#comment-15125480

ASF GitHub Bot commented on FLINK-3109:

Github user StephanEwen commented on the pull request:

    Concerning the data management: @aljoscha and me are currently heavily reworking that.
    All window operations need to go onto the "state" interfaces. Before we merge this one,
we should also do that, so please do not spend much time on optimizing how the buffers for
the two inputs are implemented.
    The interfaces for that will go into the code in a few days (they are in this pull request:
    For now, I would focus on the API and we look into the buffers in a few days.
    BTW: how exactly the buffered data is held (managed memory, external databases, etc) depends
on the "state backend" of the job. Memory behavior can be changed that way and the operators
need not worry about that.

> Join two streams with two different buffer time
> -----------------------------------------------
>                 Key: FLINK-3109
>                 URL: https://issues.apache.org/jira/browse/FLINK-3109
>             Project: Flink
>          Issue Type: Improvement
>          Components: Streaming
>    Affects Versions: 0.10.1
>            Reporter: Wang Yangjun
>              Labels: easyfix, patch
>             Fix For: 0.10.2
>   Original Estimate: 48h
>  Remaining Estimate: 48h
> Current Flink streaming only supports join two streams on the same window. How to solve
this problem?
> For example, there are two streams. One is advertisements showed to users. The tuple
in which could be described as (id, showed timestamp). The other one is click stream -- (id,
clicked timestamp). We want get a joined stream, which includes all the advertisement that
is clicked by user in 20 minutes after showed.
> It is possible that after an advertisement is shown, some user click it immediately.
It is possible that "click" message arrives server earlier than "show" message because of
Internet delay. We assume that the maximum delay is one minute.
> Then the need is that we should alway keep a buffer(20 mins) of "show" stream and another
buffer(1 min) of "click" stream.
> It would be grate that there is such an API like.
> showStream.join(clickStream)
>             .where(keySelector)
>             .buffer(Time.of(20, TimeUnit.MINUTES))
>             .equalTo(keySelector)
>             .buffer(Time.of(1, TimeUnit.MINUTES))
>             .apply(JoinFunction)
> http://stackoverflow.com/questions/33849462/how-to-avoid-repeated-tuples-in-flink-slide-window-join/34024149#34024149

This message was sent by Atlassian JIRA

View raw message