I like (2) but without adding a new 'apache-airflow-providers-prestodb' for
Facebook's prestodb -- especially since I don't currently know if the
python client library will continue to work with it unless someone tests
and adds it.



On Wed, Jan 20, 2021 at 11:43 PM Philippe Gagnon <philgagn...@gmail.com>
wrote:

> Hello Airflow-dev,
>
> About a couple of years ago the Presto community was fractured when the
> three founders of the project, Dain, David and Martin, left Facebook due to
> irreconcilable differences with the company concerning the stewardship of
> the open-source project. In order to continue working on the open-source
> Presto software, they founded the Presto Software Foundation and forked the
> prestodb project on GitHub, renaming it to prestosql. Most active
> contributors followed them and started directing their PRs towards the
> prestosql repository, which has thrived since then, overtaking the prestodb
> project as the most active "Presto" flavor.
>
> Nevertheless, the community around the original Facebook prestodb project
> was diminished but carried on with their work. Over time incompatibilities
> started to appear between the two products which no longer shared a common
> vision and roadmap. Eventually Facebook leadership decided to formalize the
> community around their version of Presto by creating another Presto
> Foundation under the auspices of the Linux Foundation. The Presto
> Foundation then filed a trademark for the Presto name, forcing the
> prestosql project to change their name to trinodb late last year [1].
>
> This puts us here in the Airflow community in a bit of a conundrum. I
> believe that over the past two years or so we have primarily targeted
> support for the prestosql version of Presto (which has now been renamed).
> With that in mind we should probably rename
> our apache-airflow-providers-presto package
> to apache-airflow-providers-trino. However, this is a backwards
> incompatible move and risks causing a high level of confusion among our
> users as there is still the prestodb project which is more or less
> wire-compatible with trinodb, and there is merit to supporting both.
>
> I'm not sure what the best path forward is on our side as every option I
> can think of has pros and cons, as such I think it warrants a discussion
> with the dev community.
>
> Off the top of my head:
>
>    1. We create apache-airflow-providers-trino and start using
>    apache-airflow-providers-presto for prestodb. This has the disadvantage of
>    possibly introducing incompatibilities down the line for users of
>    prestosql/trinodb who fail to update to the new provider.
>    2. We create apache-airflow-providers-trino and alias
>    apache-airflow-providers-presto so that it points to the modules in
>    apache-airflow-providers-trino while providing a deprecation warning. We
>    also create an apache-airflow-providers-prestodb provider package for users
>    of Facebook's prestodb. Eventually in a subsequent release we
>    remove apache-airflow-providers-presto altogether. This is less disruptive
>    for users, but can also be confusing if we don't get our messaging right.
>
> Best,
>
> Philippe
>
> [1] https://trino.io/blog/2020/12/27/announcing-trino.html
>

Reply via email to