If we were to go with a separate repo for the spouts how would we version it? Would it be consistent with the Heron repo? How would people know what version spout to use with the Heron version they are running?
On Thu, Jan 17, 2019 at 1:26 AM Ning Wang <wangnin...@gmail.com> wrote: > This is an option. I have a few concerns about it: > - There will be a lot of repos and it will be messy to manage and it might > be harder for users to find it. I am expecting at least more than ten > (different services times different languages). > - There will be some duplicated code such as build/release configs, > scripts. etc. > > I think we should be able to achieve the first reason with a single repo. > Different spouts should likely be in different folders and they can evolve > separately. > The second reason is valid, but duplicated code is a side effect. > The third reason depends on building tool I feel. Bazel is powerful, but it > is just changing time by time. :( > > Just my two cents. > > > > > > On Wed, Jan 16, 2019 at 8:09 PM Simon Weng <siming.w...@gmail.com> wrote: > > > Hi, all: > > > > Can it also be one of the options to even have separate repo for each > type > > of spouts? The reasons it is worth considering are: > > > > 1. Allow each spout to evolve and release in different pace because each > > is technically driven by external source software. For example, the > > community may need different versions of the Kafka Spout to be compatible > > with their deployed Kafka cluster in production > > 2. Allow each spout project to use the de facto build tool that suits the > > external SDK best. This will help to minimize the learning curve for > > constributors who specialize in different source software stack > > 3. Simply the maintainence of the build and CI > > > > I’m not familiar with the capability of Bazel, so certainly I’m not > > against it. If it can help to achieve some of the above, I guess one > single > > repo will also work then. > > > > SiMing > > > > On Wed, Jan 16, 2019 at 5:34 PM Ning Wang <wangnin...@gmail.com> wrote: > > > >> +Siming > >> > >> On Tue, Jan 15, 2019 at 11:35 PM Ning Wang <wangnin...@gmail.com> > wrote: > >> > >>> Hi, all, > >>> > >>> A few of us (Spencer, Saikat, Siming, Karthik, Josh, Sree) discussed > >>> today in our general slack channel that we should have spouts code > >>> somewhere so that people can reuse them (spouts are highly reusable in > >>> general) and contribute improvements. This is just a recap of the idea > and > >>> some updates. > >>> > >>> We have two options: > >>> 1. add a spouts/ dir in heron project. > >>> 2. create a new project in github. > >>> > >>> For option 1, it is easy to start. But the iteration and release will > be > >>> coupled with Heron project itself. It is likely there will be quite > some > >>> activities around spouts time by time when new spouts are added. Also, > >>> Heron itself is basically the engine itself plus APIs and tooling, > while > >>> there could be quite some spouts in future with many new dependencies > like > >>> Kafka, pubsub, neo4j and neptune, etc. It is debatable to have spout > >>> implementations in Heron project, and these extra dependencies could > add > >>> some unnecessary complexity. > >>> > >>> For option 2, there will be some work up front. but it will be much > >>> easier to manage and evolve. And here will be less concerns about new > >>> spouts (in different languages) and dependencies because spouts are > >>> relatively independent to each other and we may generate artifacts per > >>> spout. > >>> > >>> Overall most people prefer option 2 for its cleanness. > >>> > >>> I talked with Twitter OSS team. They are happy to support the > initiative > >>> and suggest us to check with Apache team and see what is the best > process. > >>> First question is that should this new side project be under Apache or > not? > >>> This might be a question to mentors. What do you think/suggest? > >>> > >>> Another topic being discussed is the build tool in case we decide to > >>> create a new side project. Maven is more mature for sure, but we will > >>> likely need multi language support so currently Bazel seems to be the > >>> winner (I personally vote for Bazel 1.0 because the backward > compatibility > >>> has been bad so far). > >>> > >>> Any ideas or suggestions, please feel free to reply. > >>> > >>> Regards, > >>> --ning > >>> > >> -- > > Sent from Gmail Mobile > > > -- Sent from A Mobile Device
