Github user asfgit closed the pull request at:
https://github.com/apache/flink/pull/872
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is
Github user gyfora commented on the pull request:
https://github.com/apache/flink/pull/872#issuecomment-120506018
Looks good to merge. If no objections, I will merge it tomorrow. :+1:
---
If your project is set up for it, you can reply to this email and have your
reply appear on
Github user gaborhermann commented on the pull request:
https://github.com/apache/flink/pull/872#issuecomment-117644273
Sorry.
* updated the docs (custom partitioning was also missing in the Scala batch
API docs)
* added IT case tests (also for the other stream partitioning
Github user gaborhermann commented on the pull request:
https://github.com/apache/flink/pull/872#issuecomment-117141916
Okay then. These are the effects of changing I did not know about. Let's
stick to (2) and later on, we might reconsider this.
---
If your project is set up for it,
Github user StephanEwen commented on the pull request:
https://github.com/apache/flink/pull/872#issuecomment-117117294
How about we leave the batch API as it is for now and address that as a
separate issue? There are quite some subtleties in how the optimizer assesses
equality of
Github user gaborhermann commented on the pull request:
https://github.com/apache/flink/pull/872#issuecomment-116647912
By the way, in the Scala DataSet the user should specify the Java
`Partitioner[K]` class. Wouldn't it be more convenient to wrap a function like
`(K, Int) = Int`
Github user StephanEwen commented on the pull request:
https://github.com/apache/flink/pull/872#issuecomment-116649358
In the batch API, equality of the partitioners is used to determine
compatibility of the partitioning. This may at some point become interesting
for the streaming
Github user gaborhermann commented on the pull request:
https://github.com/apache/flink/pull/872#issuecomment-116671285
I'd prefer the function implementation (like `(K, Int) = Int`), but it
should stay consistent with the batch API. I don't see why the wrapping would
effect the
Github user gaborhermann commented on the pull request:
https://github.com/apache/flink/pull/872#issuecomment-116736041
Sorry for not making myself clear.
I would actually go for
4. Only the Scala function (both in the streaming and batch API)
I don't understand
Github user StephanEwen commented on the pull request:
https://github.com/apache/flink/pull/872#issuecomment-116710090
I am confused now, what is it going to be?
1. Overloading, such that it is Scala function and Partitioner, at the
cost of redundant APIs.
2. Only
Github user StephanEwen commented on the pull request:
https://github.com/apache/flink/pull/872#issuecomment-116745707
The partitioner function in Scala was simply added as a mirror of the Java
API.
The batch API is stable, that means at most we can add a Scala function and
Github user gaborhermann commented on the pull request:
https://github.com/apache/flink/pull/872#issuecomment-116755999
Okay, then I will
* deprecate the partitioner implementation in the batch API
* add the function implementation to the batch API
* add the function
Github user gyfora commented on the pull request:
https://github.com/apache/flink/pull/872#issuecomment-116106372
I think I could find several use cases if I wanted to :) For example I
would often like to broadcast some model information to many downstream
operators at once. (not
Github user gaborhermann commented on the pull request:
https://github.com/apache/flink/pull/872#issuecomment-116103719
I guess it is easier for the users to understand and partitioning to
multiple channels at a time is rarely needed. Is there a use-case where it is
needed?
Github user gyfora commented on the pull request:
https://github.com/apache/flink/pull/872#issuecomment-115832961
Wouldn't it make sense to implement custom partitioning in a way that it
allows to return a array of indexes like in the ChannelSelector interface?
Returning only 1 index
GitHub user gaborhermann opened a pull request:
https://github.com/apache/flink/pull/872
[FLINK-2138] Added custom partitioning to DataStream
Custom partitioning added to DataStream in order to be more consistent with
the batch API.
You can merge this pull request into a Git
16 matches
Mail list logo