Hello Jim, Thanks for the proposed KIP. I have some meta questions about it:
1) Would an instance always pause/resume all of its current owned topologies (i.e. the named topologies), or are there any scenarios where we only want to pause/resume a subset of them? 2) From a user's perspective, do we want to always issue a `pause/resume` to all the instances or not? For example, we can define the semantics of the function as "you only need to call this function on any of the application's instances, and all instances would then pause (via the rebalance error codes)", or as "you would call this function for all the instances of an application". Which one are you referring to? 3) With EOS, there's a transaction timeout which would determine how long a transaction can stay idle before it's force-aborted on the broker side. I think when a pause is issued, that means we'd need to immediately commit the current transaction for EOS since we do not know how long we could pause for. Is that right? If yes could you please clarify that in the doc as well. Guozhang On Wed, May 4, 2022 at 10:51 AM Jim Hughes <jhug...@confluent.io.invalid> wrote: > Hi all, > > I have written up a KIP for adding the ability to pause and resume the > processing of a topology in AK Streams. The KIP is here: > https://cwiki.apache.org/confluence/pages/viewpage.action?pageId=211882832 > > Thanks in advance for your feedback! > > Cheers, > > Jim > -- -- Guozhang