I think you're misunderstanding the idea of "process" here. The point of process is to make sure something happens automatically, which is useful to ensure a certain level of quality. For example, all our patches go through Jenkins, and nobody will make the mistake of merging them if they fail tests, or RAT checks, or API compatibility checks. The idea is to get the same kind of automation for design on these components. This is a very common process for large software projects, and it's essentially what we had already, but formalizing it will make clear that this is the process we want. It's important to do it early in order to be able to refine the process as the project grows.
In terms of scope, again, the maintainers are *not* going to be the only reviewers for that component, they are just a second level of sign-off required for architecture and API. Being a maintainer is also not a "promotion", it's a responsibility. Since we don't have much experience yet with this model, I didn't propose automatic rules beyond that the PMC can add / remove maintainers -- presumably the PMC is in the best position to know what the project needs. I think automatic rules are exactly the kind of "process" you're arguing against. The "process" here is about ensuring certain checks are made for every code change, not about automating personnel and development decisions. In any case, I appreciate your input on this, and we're going to evaluate the model to see how it goes. It might be that we decide we don't want it at all. However, from what I've seen of other projects (not Hadoop but projects with an order of magnitude more contributors, like Python or Linux), this is one of the best ways to have consistently great releases with a large contributor base and little room for error. With all due respect to what Hadoop's accomplished, I wouldn't use Hadoop as the best example to strive for; in my experience there I've seen patches reverted because of architectural disagreements, new APIs released and abandoned, and generally an experience that's been painful for users. A lot of the decisions we've made in Spark (e.g. time-based release cycle, built-in libraries, API stability rules, etc) were based on lessons learned there, in an attempt to define a better model. Matei > On Nov 6, 2014, at 2:18 PM, bc Wong <bcwal...@cloudera.com> wrote: > > On Thu, Nov 6, 2014 at 11:25 AM, Matei Zaharia <matei.zaha...@gmail.com > <mailto:matei.zaha...@gmail.com>> wrote: > <snip> > Ultimately, the core motivation is that the project has grown to the point > where it's hard to expect every committer to have full understanding of every > component. Some committers know a ton about systems but little about machine > learning, some are algorithmic whizzes but may not realize the implications > of changing something on the Python API, etc. This is just a way to make sure > that a domain expert has looked at the areas where it is most likely for > something to go wrong. > > Hi Matei, > > I understand where you're coming from. My suggestion is to solve this without > adding a new process. In the example above, those "algo whizzes" committers > should realize that they're touching the Python API, and loop in some Python > maintainers. Those Python maintainers would then respond and help move the > PR along. This is good hygiene and should already be happening. For example, > HDFS committers have commit rights to all of Hadoop. But none of them would > check in YARN code without getting agreement from the YARN folks. > > I think the majority of the effort here will be education and building the > convention. We have to ask committers to watch out for API changes, know > their own limits, and involve the component domain experts. We need that > anyways, which btw also seems to solve the problem. It's not clear what the > new process would add. > > It'd be good to know the details, too. What are the exact criteria for a > committer to get promoted to be a maintainer? How often does the PMC > re-evaluate the list of maintainers? Is there an upper bound on the number of > maintainers for a component? Can we have an automatic rule for a maintainer > promotion after X patches or Y lines of code in that area? > > Cheers, > bc > >> On Nov 6, 2014, at 10:53 AM, bc Wong <bcwal...@cloudera.com >> <mailto:bcwal...@cloudera.com>> wrote: >> >> Hi Matei, >> >> Good call on scaling the project itself. Identifying domain experts in >> different areas is a good thing. But I have some questions about the >> implementation. Here's my understanding of the proposal: >> >> (1) The PMC votes on a list of components and their maintainers. Changes to >> that list requires PMC approval. >> (2) No committer shall commit changes to a component without a +1 from a >> maintainer of that component. >> >> I see good reasons for #1, to help people navigate the project and identify >> expertise. For #2, I'd like to understand what problem it's trying to solve. >> Do we have rogue committers committing to areas that they don't know much >> about? If that's the case, we should address it directly, instead of adding >> new processes. >> >> To point out the obvious, it completely changes what "committers" means in >> Spark. Do we have clear promotion criteria from "committer" to "maintainer"? >> Is there a max number of maintainers per area Currently, as committers gains >> expertise in new areas, they could start reviewing code in those areas and >> give +1. This encourages more contributions and cross-component knowledge >> sharing. Under the new proposal, they now have to be promoted to >> "maintainers" first. That reduces our review bandwidth. >> >> Again, if there is a quality issue with code reviews, let's talk to those >> committers and help them do better. There are non-process ways to solve the >> problem. >> >> So I think we shouldn't require "maintainer +1". I do like the idea of >> having explicit maintainers on a volunteer basis. These maintainers should >> watch their jira and PR traffic, and be very active in design & API >> discussions. That leads to better consistency and long-term design choices. >> >> Cheers, >> bc >> >> On Wed, Nov 5, 2014 at 5:31 PM, Matei Zaharia <matei.zaha...@gmail.com >> <mailto:matei.zaha...@gmail.com>> wrote: >> Hi all, >> >> I wanted to share a discussion we've been having on the PMC list, as well as >> call for an official vote on it on a public list. Basically, as the Spark >> project scales up, we need to define a model to make sure there is still >> great oversight of key components (in particular internal architecture and >> public APIs), and to this end I've proposed implementing a maintainer model >> for some of these components, similar to other large projects. >> >> As background on this, Spark has grown a lot since joining Apache. We've had >> over 80 contributors/month for the past 3 months, which I believe makes us >> the most active project in contributors/month at Apache, as well as over 500 >> patches/month. The codebase has also grown significantly, with new libraries >> for SQL, ML, graphs and more. >> >> In this kind of large project, one common way to scale development is to >> assign "maintainers" to oversee key components, where each patch to that >> component needs to get sign-off from at least one of its maintainers. Most >> existing large projects do this -- at Apache, some large ones with this >> model are CloudStack (the second-most active project overall), Subversion, >> and Kafka, and other examples include Linux and Python. This is also >> by-and-large how Spark operates today -- most components have a de-facto >> maintainer. >> >> IMO, adopting this model would have two benefits: >> >> 1) Consistent oversight of design for that component, especially regarding >> architecture and API. This process would ensure that the component's >> maintainers see all proposed changes and consider them to fit together in a >> good way. >> >> 2) More structure for new contributors and committers -- in particular, it >> would be easy to look up who’s responsible for each module and ask them for >> reviews, etc, rather than having patches slip between the cracks. >> >> We'd like to start with in a light-weight manner, where the model only >> applies to certain key components (e.g. scheduler, shuffle) and user-facing >> APIs (MLlib, GraphX, etc). Over time, as the project grows, we can expand it >> if we deem it useful. The specific mechanics would be as follows: >> >> - Some components in Spark will have maintainers assigned to them, where one >> of the maintainers needs to sign off on each patch to the component. >> - Each component with maintainers will have at least 2 maintainers. >> - Maintainers will be assigned from the most active and knowledgeable >> committers on that component by the PMC. The PMC can vote to add / remove >> maintainers, and maintained components, through consensus. >> - Maintainers are expected to be active in responding to patches for their >> components, though they do not need to be the main reviewers for them (e.g. >> they might just sign off on architecture / API). To prevent inactive >> maintainers from blocking the project, if a maintainer isn't responding in a >> reasonable time period (say 2 weeks), other committers can merge the patch, >> and the PMC will want to discuss adding another maintainer. >> >> If you'd like to see examples for this model, check out the following >> projects: >> - CloudStack: >> https://cwiki.apache.org/confluence/display/CLOUDSTACK/CloudStack+Maintainers+Guide >> >> <https://cwiki.apache.org/confluence/display/CLOUDSTACK/CloudStack+Maintainers+Guide><https://cwiki.apache.org/confluence/display/CLOUDSTACK/CloudStack+Maintainers+Guide >> >> <https://cwiki.apache.org/confluence/display/CLOUDSTACK/CloudStack+Maintainers+Guide>> >> - Subversion: https://subversion.apache.org/docs/community-guide/roles.html >> <https://subversion.apache.org/docs/community-guide/roles.html><https://subversion.apache.org/docs/community-guide/roles.html >> <https://subversion.apache.org/docs/community-guide/roles.html>> >> >> Finally, I wanted to list our current proposal for initial components and >> maintainers. It would be good to get feedback on other components we might >> add, but please note that personnel discussions (e.g. "I don't think Matei >> should maintain *that* component) should only happen on the private list. >> The initial components were chosen to include all public APIs and the main >> core components, and the maintainers were chosen from the most active >> contributors to those modules. >> >> - Spark core public API: Matei, Patrick, Reynold >> - Job scheduler: Matei, Kay, Patrick >> - Shuffle and network: Reynold, Aaron, Matei >> - Block manager: Reynold, Aaron >> - YARN: Tom, Andrew Or >> - Python: Josh, Matei >> - MLlib: Xiangrui, Matei >> - SQL: Michael, Reynold >> - Streaming: TD, Matei >> - GraphX: Ankur, Joey, Reynold >> >> I'd like to formally call a [VOTE] on this model, to last 72 hours. The >> [VOTE] will end on Nov 8, 2014 at 6 PM PST. >> >> Matei > >
