I am open to remove the index API and also open to piggyback this change on this KIP... Not sure if this was the intention of your comment? Or should be have a separate KIP for this?
Not sure, what other think about removing the index API?
-Matthias
On 2/5/18 4:13 PM, Guozhang Wang wrote:
> This is not included in this KIP, but I'm wondering if we should still
> support forwarding to a child with index.
>
> Because 1) if users are using DSL, they would probably never use this API
> since the ordering of the child is hard to reason from a DSL anyways.
> 2) if users are using PAPI, they would just use the other
> overload with child name.
>
> In fact the index currently is only used at KStream.branch internally, and
> maybe we can just keep a map from predicate to child name inside Filter
> operator, so that we do not use it at all. Just as a hindsight, I feel this
> overloading API is not useful to users at all.
>
>
> Guozhang
>
>
> On Thu, Feb 1, 2018 at 5:05 PM, Matthias J. Sax <matth...@confluent.io>
> wrote:
>
>> Thanks.
>>
>> I updated the KIP accordingly and started work on the PR to see if this
>> `To` interface work nicely.
>>
>> -Matthias
>>
>> On 2/1/18 4:00 PM, Ted Yu wrote:
>>> Yeah.
>>> Cleaner in this formation.
>>>
>>> On Thu, Feb 1, 2018 at 3:59 PM, Bill Bejeck <bbej...@gmail.com> wrote:
>>>
>>>> `To` works for me.
>>>>
>>>> Thanks,
>>>> Bill
>>>>
>>>> On Thu, Feb 1, 2018 at 6:47 PM, Matthias J. Sax <matth...@confluent.io>
>>>> wrote:
>>>>
>>>>> @Paolo:
>>>>>
>>>>> The timestamp will be used to set the message/record metadata timestamp
>>>>> on `Producer.send(new ProducerRecord(...,timestamp,...))`.
>>>>>
>>>>> @Bill,Ted:
>>>>>
>>>>> Might be a good idea. I was thinking about the name, and came up with
>>>> `To`:
>>>>>
>>>>>> context.forward(key, value, To.child("processorX").withTimestamp(5));
>>>>>> context.forward(key, value, To.child(1).withTimestamp(10));
>>>>>
>>>>> Without specifying the downstream child processor:
>>>>>
>>>>>> context.forward(key, value, To.all().withTimestamp(5));
>>>>>
>>>>> WDYT?
>>>>>
>>>>>
>>>>> -Matthias
>>>>>
>>>>> On 2/1/18 8:45 AM, Ted Yu wrote:
>>>>>> I like Bill's idea (pending a better name for the Forwarded).
>>>>>>
>>>>>> Cheers
>>>>>>
>>>>>> On Thu, Feb 1, 2018 at 7:47 AM, Bill Bejeck <bbej...@gmail.com>
>> wrote:
>>>>>>
>>>>>>> Hi Matthias,
>>>>>>>
>>>>>>> Thanks for the KIP!
>>>>>>>
>>>>>>> Could we consider taking an approach similar to what was done in
>>>> KIP-182
>>>>>>> with regards to overloading?
>>>>>>>
>>>>>>> Meaning we could add a "Forwarded" object (horrible name I know) with
>>>>>>> methods withTimestamp, withChildName, and withChildIndex. To handle
>>>> the
>>>>>>> case when both a child-name and child-index is provided we could
>> throw
>>>>> an
>>>>>>> exception.
>>>>>>>
>>>>>>> Then we could reduce the overloaded {{forward}} methods from 6 to 2.
>>>>>>>
>>>>>>> Thanks,
>>>>>>> Bill
>>>>>>>
>>>>>>>
>>>>>>> On Thu, Feb 1, 2018 at 3:49 AM, Paolo Patierno <ppatie...@live.com>
>>>>> wrote:
>>>>>>>
>>>>>>>> Hi Matthias,
>>>>>>>>
>>>>>>>> just a question : what will be the timestamp "type" in the new
>>>> message
>>>>> on
>>>>>>>> the wire ?
>>>>>>>>
>>>>>>>> Thanks,
>>>>>>>> Paolo.
>>>>>>>> ________________________________
>>>>>>>> From: Matthias J. Sax <matth...@confluent.io>
>>>>>>>> Sent: Wednesday, January 31, 2018 2:06 AM
>>>>>>>> To: dev@kafka.apache.org
>>>>>>>> Subject: [DISCUSS] KIP-251: Allow timestamp manipulation in
>> Processor
>>>>> API
>>>>>>>>
>>>>>>>> Hi,
>>>>>>>>
>>>>>>>> I want to propose a new KIP for Kafka Streams that allows timestamp
>>>>>>>> manipulation at Processor API level.
>>>>>>>>
>>>>>>>> https://cwiki.apache.org/confluence/display/KAFKA/KIP-
>>>>>>>> 251%3A+Allow+timestamp+manipulation+in+Processor+API
>>>>>>>>
>>>>>>>> Looking forward to your feedback.
>>>>>>>>
>>>>>>>>
>>>>>>>> -Matthias
>>>>>>>>
>>>>>>>>
>>>>>>>
>>>>>>
>>>>>
>>>>>
>>>>
>>>
>>
>>
>
>
signature.asc
Description: OpenPGP digital signature
