Agreed. It is clearly a balancing act. Let us know if you think we are off the reservation.
The two practical problems I want to solve: 1. We are too slow at reviewing patches, this makes it so that each patch has at least one other person who knows they need to look at it asap. 2. Cross cutting changes seem to lead to hacks in some areas. For example the producer code definitely got nastier during the replication work. I feel the cause is that when working on a feature you sometimes need to work on code you don't know that well, and consequently you don't really know the right thing to do. I am certainly guilty of this. Having someone who is specifically responsible for that will lead to greater incentive to catch this in review and increase the change that the right person will be doing the review. -Jay On Mon, Nov 19, 2012 at 11:24 AM, Jakob Homan <jghoman....@gmail.com> wrote: > One thing to keep in mind is that Apache dissuades against the @author > tag to try and ensure the whole community, particularly the > committers, feels ownership of the whole codebase. While committers > not familiar with a particular bit of code shouldn't necessarily > review a piece of code, neither should any one subset of the committer > community be the only ones able to move a section of the codebase > forward. > > Basically, we should just make sure that this system is not used to > fence people in or restrict proven contributors' ability to work in > one section or another. > > On 19 November 2012 11:08, Jay Kreps <jay.kr...@gmail.com> wrote: > > Okay so there seems to be a consensus that this is a good idea. I have > > added a wiki page with maintainers: > > https://cwiki.apache.org/confluence/display/KAFKA/Maintainers > > Could committers do the following: > > 1. Add themselves to areas they would like to act as a maintainer for? > Our > > goal will be two have two maintainers for each area, but if we have > either > > one or three that is fine too. Sign up for areas where you think you have > > expertise or can make a difference. > > 2. Add other areas I left out. I think it is fine if they are small. For > > example I think it would be fine to have a single maintainer for > > Utils.scala even though that is not a major area. > > > > -Jay > > > > > > On Mon, Nov 19, 2012 at 10:48 AM, Jay Kreps <jay.kr...@gmail.com> wrote: > > > >> Jun is just far and away the best reviewer. So maybe we should just > >> require that everybody get Jun to review their patches. :-) > >> > >> Or, more practically, maybe Jun should put together some guidelines on > >> what he does and we can try to emulate. > >> > >> -Jay > >> > >> > >> On Thu, Nov 8, 2012 at 2:40 PM, Prashanth Menon < > >> prashanth.men...@gmail.com> wrote: > >> > >>> +1 from me on the maintainer system and having a primary/secondary for > >>> components. > >>> > >>> I also think we should try as much as possible to get at least two > >>> reviewers for patches that come in. This is something I'm very guilty > of > >>> and am trying to correct. I get the feeling Jun is overwhelmed with > patch > >>> reviews :) > >>> > >>> - Prashanth > >>> > >>> On Fri, Nov 2, 2012 at 12:11 PM, Neha Narkhede < > neha.narkh...@gmail.com > >>> >wrote: > >>> > >>> > +1 on this, I like the idea of having a primary/secondary owner > system > >>> > for each component. > >>> > > >>> > Thanks, > >>> > Neha > >>> > > >>> > On Thu, Nov 1, 2012 at 10:15 PM, Jun Rao <jun...@gmail.com> wrote: > >>> > > I think that's a good idea. It will be good to have at least 2 > >>> > maintainers > >>> > > per component. > >>> > > > >>> > > I'd encourage more people to review patches. The more patches one > >>> > reviews, > >>> > > the more familiar he/she is with the components. > >>> > > > >>> > > Thanks, > >>> > > > >>> > > Jun > >>> > > > >>> > > On Thu, Oct 4, 2012 at 1:13 PM, Jay Kreps <jay.kr...@gmail.com> > >>> wrote: > >>> > > > >>> > >> Hey guys, > >>> > >> > >>> > >> The number of developers and code base size for Kafka is getting > >>> larger. > >>> > >> > >>> > >> One way to help scale things gracefully would be to have an > official > >>> > idea > >>> > >> of "subsystems" and have official maintainers for those. The > duties > >>> of a > >>> > >> maintainer would be > >>> > >> 1. Be the final word on API design in that area > >>> > >> 2. Ensure sufficient documentation and test coverage for that > >>> subsystem > >>> > >> 3. Review all code changes in that sub-system area > >>> > >> 4. Ensure that patches in that area get applied in a timely > fashion > >>> > >> > >>> > >> In particular I think we could do a better job of getting patches > in > >>> in > >>> > a > >>> > >> timely manner. > >>> > >> > >>> > >> Here are what I see as logically distinct systems or areas: > >>> > >> > >>> > >> - Producer (java and scala) > >>> > >> - Consumer (java and scala) > >>> > >> - Network layer (kafka.network.*) > >>> > >> - Log (kafka.log.*) > >>> > >> - Replication (controller, fetcher threads, hw mark stuff, etc) > >>> > >> - Kafka API impl (basically just KafkaApi.scala) > >>> > >> - Hadoop stuff > >>> > >> - Perf tools and system tests > >>> > >> - Misc other small things: metrics, utils, etc. > >>> > >> > >>> > >> Obviously many features will cut across these layers, but the > idea is > >>> > that > >>> > >> by having a real owner that is responsible for that area we will > get > >>> > higher > >>> > >> quality. > >>> > >> > >>> > >> I think we are doing this informally already, but making it formal > >>> would > >>> > >> help ensure you knew the right people to get input from. I think > it > >>> > >> probably wouldn't make sense to start doing this until post-0.8 > >>> since we > >>> > >> are in the middle of so many things right now, but I wanted to see > >>> what > >>> > >> people thought...? > >>> > >> > >>> > >> -Jay > >>> > >> > >>> > > >>> > >> > >> >
