I have updated the KIP. I will start a vote on Monday if there are no further comments by then.
Best, David On Wed, Feb 19, 2020 at 5:59 PM David Jacot <dja...@confluent.io> wrote: > Hey Jason, > > You're right. The leader epoch is not bumped when a topic is deleted. Using > a sentinel (e.g. -1) which would override any existing epoch sounds like a > good approach in this case. Let me update the KIP to reflect that. > > Thanks, > David > > On Fri, Feb 14, 2020 at 8:11 PM Jason Gustafson <ja...@confluent.io> > wrote: > >> Hey David, >> >> Thanks, it makes sense to prevent reordering, especially for the case of >> reassignment. When a topic is deleted, however, I am not sure we will have >> a bumped epoch to send. I guess for that case, we could send a sentinel >> which would take the existing semantics of overriding any existing epoch? >> >> -Jason >> >> On Tue, Feb 11, 2020 at 12:48 PM David Jacot <dja...@confluent.io> wrote: >> >> > Hi all, >> > >> > I've put together a very small KIP which proposes to add the leader >> epoch >> > in the >> > StopReplicaRequest in order to make it robust to reordering: >> > >> > >> https://cwiki.apache.org/confluence/display/KAFKA/KIP-570%3A+Add+leader+epoch+in+StopReplicaRequest >> > >> > Please take a look at the KIP and let me know what you think. >> > >> > Best, >> > David >> > >> >
