Hi Mika, Thanks for bringing up this discussion.
Regarding dropping support of Python 3.8: +1 for this. Since we will add support for Python 3.12 in Flink 2.1[1], this means we need to maintain 5 Python versions if we still support Python 3.8. Considering that 3.8 is already EOL and the issue you encountered, it makes sense to me to drop support for it in Flink 2.1. Besides, I see that many other Python libraries have dropped the support of Python 3.8 in the latest releases, e.g. Pandas, Numpy, Apache Beam, PySpark, Ray, etc. > We have some sort of policy or intuition around deprecating/dropping support > for Python versions to make maintenance burden a little easier, and whether > this should be based on languages reaching eol, wider language version usage > or whether we have insight into what python versions those on the latest > Flink versions are using. There are still no such policies for now. In the past, we proposed to drop the EOL Python versions when adding support for new Python versions. This also make sure it doesn't bring too much burden for the testing (It needs to test against all Python versions in the nightly tests). From what I see in other Python projects, they usually maintain 3 to 4 Python versions in practice. This is the same case for PyFlink in practice. > Whether we should deprecate the version first (like in FLINK-28195[3]), and > how many minor versions this deprecation should exist for before dropping > altogether (just one?) Kind regards, I'm slightly inclined to just drop it considering the issue you encountered and Python 3.8 was already EOL half a year ago. Regards, Dian [1] https://issues.apache.org/jira/browse/FLINK-37823 On Mon, May 26, 2025 at 10:55 PM Mika Naylor <m...@autophagy.io.invalid> wrote: > > Hi all, > > I recently wanted to look into using the new dependency groups mechanism for > centralising testing/development requirements in one place in a consistent > format, rather than scattered in various places. I opened FLINK-37775 to > reflect this, but had to drop it as it's only a feature in new versions of > pip, and new pip versions are no longer being released for our minimum Python > version, 3.8. > > 3.8 is officially end of life as of last year[1], but is still used[2]. With > 3.8 at EOL, and 3.9 reaching it end of this year, I wanted to ask for > feedback on whether: > > • We have some sort of policy or intuition around deprecating/dropping > support for Python versions to make maintenance burden a little easier, and > whether this should be based on languages reaching eol, wider language > version usage or whether we have insight into what python versions those on > the latest Flink versions are using. > • Whether we should deprecate the version first (like in FLINK-28195[3]), > and how many minor versions this deprecation should exist for before dropping > altogether (just one?) > Kind regards, > Mika > > [1] https://devguide.python.org/versions/ > [2] https://w3techs.com/technologies/history_details/pl-python/3 > [3] https://issues.apache.org/jira/browse/FLINK-28195
