David, We never backport new features to old releases. This new feature will be > only available from 3.6 (or 3.7) onwards for both client and server.
Good to know. I think that makes the argument for bumping the version even stronger. On Mon, Jul 24, 2023 at 5:01 PM David Jacot <dja...@confluent.io.invalid> wrote: > Hi Mayank, > > We never backport new features to old releases. This new feature will be > only available from 3.6 (or 3.7) onwards for both client and server. > > Best, > David > > On Mon, Jul 24, 2023 at 5:20 PM Mayank Shekhar Narula < > mayanks.nar...@gmail.com> wrote: > > > Thanks Jose/David/Ismael for your inputs. > > > > Not bumping the version, would require both broker & client to backport > > changes. Especially for FetchResponse, as backporting would have to be > done > > all the way back to 3.1, so this effort isn't trivial, and was originally > > underestimated. > > > > Considering backporting effort, and given it does make it easier to > reason > > about the client-server interaction, I will bump the version. For > Produce & > > Fetch, new bumped versions are going to be 10 & 16 respectively. > > > > Do note that FetchResponse has currentLeader introduced at version 12, > but > > only Java client at version 16 will support it. But that's the best we > can > > do. > > > > > > > > > > > > > > On Mon, Jul 24, 2023 at 3:32 PM David Jacot <dja...@confluent.io.invalid > > > > wrote: > > > > > Hi Mayank, > > > > > > 01. Hum... I understand your reasoning. I think that this is mainly > > > beneficial for clients lagging behind in terms of supported versions. > > > However, it is the opposite for the java client which is up to date. > > > Personally, I would rather prefer to bump both versions and to add the > > > tagged fields. This would allow us to better reason about what the > client > > > is supposed to do when we see the version on the server side. > Otherwise, > > we > > > will never know if the client uses this or not. > > > > > > 03. Yeah, ignorable is not needed if the field has the same version. > > > > > > Thanks, > > > David > > > > > > On Mon, Jul 24, 2023 at 4:09 PM José Armando García Sancio > > > <jsan...@confluent.io.invalid> wrote: > > > > > > > Hey Mayank, > > > > > > > > It is probably binary compatible to have the NodeEndponts fielld at > > > > taggedVersion 12+ but I think it is misleading as a code reviewer. > The > > > > Java Kafka client at version 12 will never be able to handle those > > > > fields. Or are you planning to backport these improvements to those > > > > clients and brokers? If not, can we set the version and taggedVersion > > > > to 15+? > > > > > > > > Thanks, > > > > -- > > > > -José > > > > > > > > > > > > > -- > > Regards, > > Mayank Shekhar Narula > > > -- Regards, Mayank Shekhar Narula
