[
https://issues.apache.org/jira/browse/FLINK-8482?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16649849#comment-16649849
]
ASF GitHub Bot commented on FLINK-8482:
---------------------------------------
florianschmidt1994 opened a new pull request #6843: [FLINK-8482][DataStream]
Allow users to choose from different timestamp strategies for interval join
URL: https://github.com/apache/flink/pull/6843
## What is the purpose of the change
This change will allow users to choose from different timestamp strategies
when using the IntervalJoin in the DataStream API. A timestamp strategy defines
which timestamp gets assigned to two elements that are joined together.
The usage in the API looks as follows
```
leftKeyedStream.intervalJoin(rightKeyedStream)
.between(<Time>, <Time>)
.assignLeftTimestamp()
.process(<ProcessJoinFunction>)
```
The possible options to pick from are
- `assignLeftTimestamp()`:
- `assignRightTimestamp()`
- `assignMinTimestamp()`
- `assignMaxTimestamp()`
In certain scenarios the watermark emitted by the IntervalJoinOperator needs
to be delayed, in order to not produce any late data. This is only necessary
when choosing `assignMinTimestamp()` or for certain combinations of upper /
lower bound and `assignRightTimestamp()` / `assignLeftTimestamp()`. It is never
necessary when using `assignMaxTimestamp()`.
Delaying watermarks is implemented by subtracting the necessary delay from
each incoming watermarks timestamp before emitting it. Note that this only
takes effect with respect to downstream operators. The internal timerservice
will still be advanced by the original watermark, so that timers (including
those defined by the user) will still be fired with respect to the original
watermark.
## Brief change log
- Implement timestamp strategies in IntervalJoinOperator
- Add to Scala DataStream API
- Add to Java DataStream API
- Add test cases to IntervalJoinITCase (both java & scala)
## Verifying this change
- Extends integration tests in IntervalJoinITCase
## Does this pull request potentially affect one of the following parts:
- Dependencies (does it add or upgrade a dependency): no
- The public API, i.e., is any changed class annotated with
`@Public(Evolving)`: yes
- The serializers: no
- The runtime per-record code paths (performance sensitive): yes
- Anything that affects deployment or recovery: JobManager (and its
components), Checkpointing, Yarn/Mesos, ZooKeeper: no
- The S3 file system connector: no
## Documentation
- Does this pull request introduce a new feature? yes
- If yes, how is the feature documented? JavaDocs
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
> Implement and expose option to use min / max / left / right timestamp for
> joined streamrecords
> ----------------------------------------------------------------------------------------------
>
> Key: FLINK-8482
> URL: https://issues.apache.org/jira/browse/FLINK-8482
> Project: Flink
> Issue Type: Sub-task
> Reporter: Florian Schmidt
> Assignee: Florian Schmidt
> Priority: Major
> Labels: pull-request-available
> Fix For: 1.7.0
>
>
> The idea: Expose the option of which timestamp to use for the result of a
> join. The idea that is currently the floating around includes the options
> * _left_: Use timestamp of the element in a join that came from the left
> stream
> * _right_: Use timestamp of the element in a join that came from the right
> stream
> * _max_: Use the max timestamp of both elements in a join
> * _min_: Use the max timestamp of both elements in a join
> All options but _max_ require to introduce delaying watermarks in the
> operator, which is something that we were hesitant to do until now. This
> should probably under go discussion once more in order to see if / how we
> want to add this now. We could even think of exposing this in a more general
> way by adding a base operator that allows delayed watermarks.
> This will also be groundwork for supporting outer joins (FLINK-8483) for
> which in any case we watermark delays to provide correctness.
> Also the API for this needs some feedback in order to expose this in a
> powerful, yet clear way. In my PoC at [1] I used the naming convention left /
> right to refer to specific streams with currently is not something the api
> exposes to the user, we should probably use something more clever here.
> Example
> {code:java}
> keyedStreamOne.
> .intervalJoin(keyedStreamTwo)
> .between(Time.milliseconds(0), Time.milliseconds(2))
> .assignMinTimestamp() // alternative .assignMaxTimestamp()
> .assignLeftTimestamp() .assignRightTimestamp()
> .process(new ProcessJoinFunction() { /* impl */ })
> {code}
>
> Any feedback is highly appreciated!
> [1]
> https://github.com/florianschmidt1994/flink/tree/flink-8482-add-option-for-different-timestamp-strategies-to-interval-join-operator
> cc [~StephanEwen] [~kkl0u]
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
