Hi Sachin,

The 2nd case can likely handled with the committed offset, which is covered by 
UNCOMMITTED_EARLIEST or UNCOMMITTED_LATEST.

The 1st case it may make sense but even if you give the start offset, since 
Kafka polls a certain number of bytes, and not specifically a number of 
records, it may not be trivial to guarantee that the same exact dataset is 
polled each time.

However, If we as a community agree that it is useful to support your proposed 
feature, I have no particular argument to do so.

Best,
Hugo
PS. We usually have the practice to initiate discussion threads with email 
subject prefixed with [DISCUSS]



> On Jan 18, 2017, at 6:33 AM, Sachin Pasalkar <[email protected]> 
> wrote:
> 
> Hi,
> 
> I was looking at code of current KafkaTridentSpoutEmitter & KafkaSpout class. 
> Can we add functionality based on user provided offset to start from 
> particular offset? This would be useful incase user wants to reprocess 
> particular data set. Another example user has changed the group id & aware 
> where old offset committed & he wants to start processing from same position.
> 
> Does this make sense? OR its explicit that it will not be supported?
> 
> Thanks,
> Sachin

Reply via email to