I like that. I also think it is important to have a process to remove provider if needed. +1
On 2023/10/27 09:00:25 Jarek Potiuk wrote: > > I think in the case of Qubole it is pretty easy to remove it from the > provider codebase. I'm pretty sure that almost no one even noticed this > removal. > > Yeah. Agree. This one is pretty "obvious" that's why I would like to create > a process for doing it along the way so that in the future if we have other > non-obvious cases we can just "follow the process". > > BTW.That would be really great to have as with it, we will have a complete > "life-cycle" of the providers. > > * we know how we approve the new ones > * we know how we release new versions (Elad is amazing to release them > every few weeks) > * we know how we maintain back-compatibility (bumping min-version of > Airflow that we regularly do ) > * we know how to involve stakeholders to system-test them and to make them > work in a stable way when they connect to external services (Amazon works, > Google in Progres, Open API and others promised by Astronomer) > * we know how to involve stakeholders with mixed-governance in case they > want older releases (never happened yet but we know how to do it) > * we know how to suspend and resume them when they prove to be problematic > and pass resolution of that to external stakeholders (happened with Yandex > - both suspend and resume) > * we (will finally) know how to retire them when we decide we do not want > to maintain them - except security fixes - any more > > That will pretty much complete our process of "life-cycle" management for > providers. > > J. > > > > > On Thu, Oct 26, 2023 at 10:00 PM Jarek Potiuk <ja...@potiuk.com> wrote: > > > > >> I suggest also removing it from pypi for security reasons. If there is a > >> security issue with it then the issue will remain with us. > >> > >> > > I am quite sure we still have to handle security issues if someone finds > > them. releasing such a provider will still be possible using the tag/branch > > and we will be obliged to release a new version IF it is still used and a > > security issue is found. > > Removal of packages from PyPI does not remove our obligation to fix a > > security problem. We have also source packages released via > > downloads.apache.org and archives.apache.org - and those we can't remove > > either. > > > > I think this is one of the obligations of the Foundation being the > > "steward" of software it releases - that's why there is also the attic PMC > > to manage projects that PMC is unable to support (projects are moved to > > attic when they fail a roll call from the board with less than 2 PMC > > members confirming that they are still there and ready to handle releases > > if needed. It's also being discussed to be more formal for the CRA > > regulations right now - "stewards" of the software put on the market should > > be responsible for handling security issues in a timely manner). The act of > > release with 3 +1s of PMC is a legal act of the Foundation placing software > > on the market and we can't make it "unhappen". > > > > > >> B. > >> > >> Sent from my iPhone > >> > >> > On 26 Oct 2023, at 20:20, Jarek Potiuk <ja...@potiuk.com> wrote: > >> > > >> > Hello Airflow community, > >> > > >> > How do we feel about removing the Qubole provider completely (leaving > >> only > >> > old releases in PyPI? > >> > > >> > On September 1 2023 ( > >> > https://lists.apache.org/thread/p394d7w7gc7lz61g7qdthl96bc9kprxh) the > >> > Qubole operator ws suspended. > >> > > >> > Due to the reasons described in the thread (Qubole got acquired and the > >> > service is generally abandoned) there is pretty much no chance for it > >> to be > >> > resumed. > >> > > >> > I'd love to remove it completely and introduce a process where we can do > >> > similar things in the future for other providers if we decide to do so. > >> > > >> > I checked in the Attic project in the ASF (this is where abandoned > >> project > >> > of the ASF get moved to) and it seems that just removing part of the > >> > project that has an active PMC is not going through attic > >> > https://issues.apache.org/jira/browse/ATTIC-218 . We are free to > >> define our > >> > rules for that and I would like to use the opportunity to hash it out > >> and > >> > propose a process (similarly to suspension) and criteria to remove > >> > providers from being maintained by us. > >> > > >> > It's more than suspension. We will completely stop updating the related > >> > code (right now some automated changes can still be applied and > >> suspended > >> > providers can be resumed with simple PR). I would like to have the > >> "next" > >> > step after "suspending" - removal. > >> > > >> > Roughly - we send PROPOSAL followed by VOTE (or immediately VOTE in > >> > obvious cases) with justification, PMC members only have the binding > >> votes > >> > (similar as for releases). > >> > > >> > Only git history will remain - all the rest will be removed (including > >> > extra) - no traces of the provider remain in the next MINOR release > >> (2.8.0 > >> > in the case of Quibole). The provider will still be in PyPI and > >> historical > >> > releases will be in https://archive.apache.org . If someone would like > >> to > >> > bring back such a provider, It should go through the same process as a > >> new > >> > provider (voting/consensus). And we might reject it. > >> > > >> > WDYT? Any comments for such an approach / process ? > >> > > >> > J. > >> > >> --------------------------------------------------------------------- > >> To unsubscribe, e-mail: dev-unsubscr...@airflow.apache.org > >> For additional commands, e-mail: dev-h...@airflow.apache.org > >> > >> > --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@airflow.apache.org For additional commands, e-mail: dev-h...@airflow.apache.org
