Given my experience with Postgres in the past, I've never noticed any regressions between versions, not for the kind of run-of-the-mill usage we are doing, so I would vote for testing oldest version we want to support (to make sure the features we use are in that version) and the latest stable.
So I'd go for just testing versions 9.6 and 13. -ash On Oct 23 2020, at 12:29 pm, Jarek Potiuk <jarek.pot...@polidea.com> wrote: > Following the discussions in Slack: > https://apache-airflow.slack.com/archives/CCQ7EGB1P/p1603448752476900 > > Which versions of Postgres we should support and use in Airlfow 2.0? > > According to: https://www.postgresql.org/support/versioning/ > > Version Current minor Supported First Release Final Release > 13 13.0 Yes September 24, 2020 November 13, 2025 > 12 12.4 Yes October 3, 2019 November 14, 2024 > 11 11.9 Yes October 18, 2018 November 9, 2023 > 10 10.14 Yes October 5, 2017 November 10, 2022 > 9.6 9.6.19 Yes September 29, 2016 November 11, 2021 > 9.5 9.5.23 Yes January 7, 2016 February 11, 2021 > > > Looks like all 9.5., 9.6, 10, 11. 12, 13 are still "live", thought 9.5 is > close to end of it. > > Ideally I'd opt for 9.6. 10. 11. 12. 13 - currently we support (and test) 9.6 > and 10. From the discussion - Astronomer run their tests on 12 and 13, so I > presume it's just a matter of running the tests in CI. If we implement the > "no-full-matrix" approach for not-approved PRs, this might even be a viable > option to run all versions (but only for approved/master PRs). We might also > decide to exclude some matrix combinations (like run each Postgres with a > different Python version). > > WDYT? Any opinions? > > J. > > > -- > > > Jarek Potiuk > Polidea (https://www.polidea.com/) | Principal Software Engineer > > > > > > > > M: +48 660 796 129 (tel:+48660796129) > > > > > > > > > > > > > > > > > > >
