Thanks for the KIP, +1 (binding). A few comments: 1. I agree with Jun about LEADER_NOT_AVAILABLE for the error code for older versions. 2. OffsetNotAvailableException seems clear enough (i.e. we don't need the "ForPartition" part) 3. The KIP seems to be missing the compatibility section. 4. It would be good to mention that it's now possible for a fetch to succeed while list offsets will not for a period of time. And for older versions, the latter will return LeaderNotAvailable while the former would work fine, which is a bit unexpected. Not much we can do about it, but worth mentioning it in my opinion.
Ismael On Tue, Oct 17, 2017 at 9:26 PM, Jun Rao <j...@confluent.io> wrote: > Hi, Colin, > > Thanks for the KIP. +1. Just a minor comment. For the old client requests, > would it be better to return a LEADER_NOT_AVAILABLE error instead? > > Jun > > On Tue, Oct 17, 2017 at 11:11 AM, Colin McCabe <cmcc...@apache.org> wrote: > > > Hi all, > > > > I'd like to start the voting process for KIP-207:The Offsets which > > ListOffsetsResponse returns should monotonically increase even during a > > partition leader change. > > > > See > > https://cwiki.apache.org/confluence/display/KAFKA/KIP- > > 207%3A+Offsets+returned+by+ListOffsetsResponse+should+be+ > > monotonically+increasing+even+during+a+partition+leader+change > > for details. > > > > The voting process will run for at least 72 hours. > > > > regards, > > Colin > > >
