So my issue is 70791 is that it itself is (or will be) a regression on 3.3.0

By limiting Pandas to < 3 in this change in a point release, we will end up 
breaking any dag that was using Pandas 3.0 features (because 3.0.5 will happily 
install with 3.3.0), but with the proposed version limitation, it would get 
forcibly downgraded. Either this will result in unsatisfiable constraints if 
the users has `pandas>=3.0`, or perhaps worse it would silently downgrade.

And as for the worry about downgrade breaking things, yes, but Pandas 3.0 was 
released on 21st January 2026. By forcing a downgrade now we are re-breaking 
anything people may have already dealt with this. To me it's not really any 
different to "I changed my dag to push something new to xcom, if I roll back 
the dag I can't use those XComs anymore”. 

In this regard Airflow should behave as a library, not an application — and be 
lenient in the versions we work with.

If we’d applied this version constraint 6 months ago/just after pandas 3.0 was 
released it’s fine, but almost 7 months after the fact it seems actively 
harmful for compat/expectations.

I think we should revert 70791 and merge 
https://github.com/apache/airflow/pull/70558/ in instead. If everyone 
agrees/doesn’t disagree, then I can handle these changes as you are travelling 
Jarek. Perhaps we need to extend it with. “When attempting to load Pandas2 df 
when Pandas3 is the active version throw a nice error”. TBD? (Unless anyone 
already knows how this will behave?)

-ash

> On 4 Aug 2026, at 08:04, Jarek Potiuk <[email protected]> wrote:
> 
>> A (temporary) -1 on anything including
> https://github.com/apache/airflow/pull/70791 please — I  want to catch up
> with the logic and thinking there (I understand this was discussed on the
> dev call)
> 
> FYI Ash - I think this is important to mention that it also impacts Airlfow
> -> not only providers, so if we have other proposals - I think it would be
> great to get it sorted before RC. I am flying in about few hours, and then
> i will also be available "sometimes" so it would be great to know if we
> need to do some follow-ups - might delay 3.1.1 as currently there is a
> runtime check in Airflow preventing pandas3 to be used **when**
> serialization is used.
> 
> 
> J.
> 
> 
> On Tue, Aug 4, 2026 at 6:16 AM Steve Ahn <[email protected]> wrote:
> 
>> +1 (non-binding) on airbyte 6.0.1rc1.
>> 
>> Tested against the RC wheel on 3.3.0: a cancelled Airbyte sync now yields a
>> "cancelled" event rather than falling through the success branch (#69786).
>> 
>> I also verified my changes in amazon 9.34.0rc1 (#69676, deferred Redshift
>> statement
>> is cancelled on kill) and common.sql 2.1.0rc1 (#69736, value-check
>> tolerance no
>> longer inverts for negative expected values) — both behave correctly
>> against their
>> RC wheels. Not voting on those two since they carry #70791, which Ash has
>> asked to
>> hold.
>> 
>> --
>> Steve Ahn
>> 
>> On Mon, Aug 3, 2026 at 4:17 AM Jarek Potiuk <[email protected]> wrote:
>> 
>>> Correction: (We discussed these at the dev call and in the PRs/issues—not
>>> on the dev list, though.) Perhaps it's a good idea that we are doing this
>>> now. But if you have any thoughts, perhaps it's better to discuss them
>> in a
>>> separate thread :).
>>> 
>>> On Mon, Aug 3, 2026 at 1:15 PM Jarek Potiuk <[email protected]> wrote:
>>> 
>>>> Sure. Take your time Ash.
>>>> 
>>>> Just to give you more context (we can discuss this on Slack if you have
>>>> questions or doubts, rather than here).
>>>> 
>>>> I think there are no "perfect" solutions here due to how "pandas" is
>>> built
>>>> into our XCom processing and potential migration issues. There are
>>> **only**
>>>> trade-offs. We discussed a few of those at the devlist. I'm also not
>>>> entirely happy with what we came up with, but I think the issue here is
>>>> choosing the "lesser evil".
>>>> 
>>>> Unless of course we can find a solution that covers all potential
>> issues
>>>> here:
>>>> 
>>>> a) risk of accidental migration issues to Pandas 3 by people without
>>>> realising that it breaks both XCom **and** incompatibilities with
>> Pandas
>>> 3
>>>> b) risk of "no easy downgrade" path when people - even knowingly -
>>> migrate
>>>> to Pandas 3, only to realize later they need to downgrade.
>>>> c) risk of not supporting Pandas 3 "today" for those who need it
>>>> 
>>>> We must weigh all these risks. The trade-off we developed considers
>> risks
>>>> (a) and (b) more impactful than (c). Our approach is to deliberately
>> add
>>>> support for Pandas 3 in Airflow 3.4 while temporarily mitigating (a)
>> and
>>>> (b).
>>>> But other trade-offs here are possible of course.
>>>> 
>>>> J,
>>>> 
>>>> 
>>>> On Mon, Aug 3, 2026 at 12:33 PM Ash Berlin-Taylor <[email protected]>
>>> wrote:
>>>> 
>>>>> A (temporary) -1 on anything including
>>>>> https://github.com/apache/airflow/pull/70791 please — I  want to
>> catch
>>>>> up with the logic and thinking there (I understand this was discussed
>> on
>>>>> the dev call)
>>>>> 
>>>>> -ash
>>>>> 
>>>>> 
>>>>>> On 3 Aug 2026, at 08:43, Andrew Chang <[email protected]>
>> wrote:
>>>>>> 
>>>>>> Thanks for preparing the release, Jarek!
>>>>>> 
>>>>>> +1 (non-binding)
>>>>>> 
>>>>>> Tested apache-airflow-providers-microsoft-azure 14.1.0rc1 on
>> released
>>>>>> Airflow 3.3.0 with breeze --use-airflow-version. I focused on my
>>> change
>>>>>> #70301, which registers the wasb remote logging scheme.
>>>>>> 
>>>>>> The wasb scheme resolves to WasbRemoteLogIO through provider
>> dispatch,
>>>>> and
>>>>>> the legacy discovery path is not called. from_config() builds the IO
>>>>> from
>>>>>> the [logging] section as expected. I also ran a real upload and read
>>>>>> round-trip against Azurite, and the log content matched.
>>>>>> 
>>>>>> Best,
>>>>>> Andrew
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> Jarek Potiuk <[email protected]> 於 2026年8月3日週一 上午8:37寫道:
>>>>>> 
>>>>>>> Hey all,
>>>>>>> 
>>>>>>> I have just cut the new wave Airflow Providers packages with
>> release
>>>>>>> preparation date 2026-08-01. This email is calling a vote on the
>>>>>>> release,
>>>>>>> which will last for 72 hours - which means that it will end on
>>>>>>> 2026-08-06 00:40 UTC and until 3 binding +1 votes have been
>> received.
>>>>>>> 
>>>>>>> Consider this my (binding) +1.
>>>>>>> 
>>>>>>> This wave contains 48 providers. There are no breaking changes in
>> any
>>>>>>> of them - all bumps are minor or patch level.
>>>>>>> 
>>>>>>> Four of the providers are re-cuts of release candidates that were
>>>>>>> prepared in the 2026-07-22 wave but never published, so they are at
>>>>>>> rc2 rather than rc1:
>>>>>>> 
>>>>>>> - apache-airflow-providers-clickhousedb 1.0.1rc2
>>>>>>> - apache-airflow-providers-common-ai 0.7.0rc2
>>>>>>> - apache-airflow-providers-google 22.3.0rc2
>>>>>>> - apache-airflow-providers-openlineage 2.20.0rc2
>>>>>>> 
>>>>>>> For those four, the testing status issue carries over the check
>> marks
>>>>>>> from the previous wave, so only the genuinely new commits are left
>>>>>>> unchecked.
>>>>>>> 
>>>>>>> Airflow Providers are available at:
>>>>>>> 
>> https://dist.apache.org/repos/dist/dev/airflow/providers/2026-08-01
>>>>>>> 
>>>>>>> *apache_airflow_providers-2026-08-01-source.tar.gz* is the full
>>> source
>>>>>>> tarball of airflow repo - snapshot taken at the moment of
>> provider's
>>>>>>> release.
>>>>>>> 
>>>>>>> *apache-airflow-providers-<PROVIDER>-*.tar.gz* are the convenience
>>>>>>> python "sdist" distributions that we publish in PyPI
>>>>>>> 
>>>>>>> *apache_airflow_providers_<PROVIDER>-*.whl are the convenience
>> Python
>>>>>>> "wheel" distributions that we publish in PyPI.
>>>>>>> 
>>>>>>> The test procedure for PMC members is described in
>>>>>>> 
>>>>>>> 
>>>>> 
>>> 
>> https://github.com/apache/airflow/blob/main/dev/README_RELEASE_PROVIDERS.md#verify-the-release-candidate-by-pmc-members
>>>>>>> 
>>>>>>> The test procedure for and Contributors who would like to test this
>>> RC
>>>>>>> is described in:
>>>>>>> 
>>>>>>> 
>>>>> 
>>> 
>> https://github.com/apache/airflow/blob/main/dev/README_RELEASE_PROVIDERS.md#verify-the-release-candidate-by-contributors
>>>>>>> 
>>>>>>> 
>>>>>>> Public keys are available at:
>>>>>>> https://dist.apache.org/repos/dist/release/airflow/KEYS
>>>>>>> 
>>>>>>> Please vote accordingly:
>>>>>>> 
>>>>>>> [ ] +1 approve
>>>>>>> [ ] +0 no opinion
>>>>>>> [ ] -1 disapprove with the reason
>>>>>>> 
>>>>>>> Only votes from PMC members are binding, but members of the
>> community
>>>>> are
>>>>>>> encouraged to test the release and vote with "(non-binding)".
>>>>>>> 
>>>>>>> Please note that the version number excludes the 'rcX' string.
>>>>>>> This will allow us to rename the artifact without modifying
>>>>>>> the artifact checksums when we actually release it.
>>>>>>> 
>>>>>>> The status of testing the providers by the community is kept here:
>>>>>>> https://github.com/apache/airflow/issues/70953
>>>>>>> 
>>>>>>> The issue is also the easiest way to see important PRs included in
>>> the
>>>>>>> RC candidates.
>>>>>>> Detailed changelog for the providers will be published in the
>>>>>>> documentation after the
>>>>>>> RC candidates are released.
>>>>>>> 
>>>>>>> You can find the RC packages in PyPI following these links:
>>>>>>> 
>>>>>>> 
>> https://pypi.org/project/apache-airflow-providers-airbyte/6.0.1rc1/
>>>>>>> 
>> https://pypi.org/project/apache-airflow-providers-alibaba/3.4.0rc1/
>>>>>>> 
>> https://pypi.org/project/apache-airflow-providers-amazon/9.34.0rc1/
>>>>>>> 
>>> https://pypi.org/project/apache-airflow-providers-anthropic/0.2.1rc1/
>>>>>>> 
>>>>> 
>>> https://pypi.org/project/apache-airflow-providers-apache-hdfs/4.12.2rc1/
>>>>>>> 
>>>>> 
>> https://pypi.org/project/apache-airflow-providers-apache-hive/9.6.1rc1/
>>>>>>> 
>>>>> 
>>> 
>> https://pypi.org/project/apache-airflow-providers-apache-kafka/1.15.1rc1/
>>>>>>> 
>>>>> 
>> https://pypi.org/project/apache-airflow-providers-apache-livy/4.6.0rc1/
>>>>>>> 
>>>>> 
>>> https://pypi.org/project/apache-airflow-providers-apache-spark/6.3.1rc1/
>>>>>>> 
>> https://pypi.org/project/apache-airflow-providers-arangodb/2.9.6rc1/
>>>>>>> 
>> https://pypi.org/project/apache-airflow-providers-celery/3.23.1rc1/
>>>>>>> 
>>>>> 
>>> https://pypi.org/project/apache-airflow-providers-clickhousedb/1.0.1rc2/
>>>>>>> 
>>>>>>> 
>>>>> 
>>> 
>> https://pypi.org/project/apache-airflow-providers-cncf-kubernetes/10.21.0rc1/
>>>>>>> https://pypi.org/project/apache-airflow-providers-cohere/1.6.7rc1/
>>>>>>> 
>>> https://pypi.org/project/apache-airflow-providers-common-ai/0.7.0rc2/
>>>>>>> 
>>>>> 
>>> 
>> https://pypi.org/project/apache-airflow-providers-common-compat/1.18.0rc1/
>>>>>>> 
>>> https://pypi.org/project/apache-airflow-providers-common-sql/2.1.0rc1/
>>>>>>> 
>>>>> 
>> https://pypi.org/project/apache-airflow-providers-databricks/7.18.1rc1/
>>>>>>> 
>>> https://pypi.org/project/apache-airflow-providers-dbt-cloud/4.9.3rc1/
>>>>>>> https://pypi.org/project/apache-airflow-providers-docker/4.5.9rc1/
>>>>>>> https://pypi.org/project/apache-airflow-providers-edge3/4.3.0rc1/
>>>>>>> 
>>>>> 
>>> 
>> https://pypi.org/project/apache-airflow-providers-elasticsearch/6.9.0rc1/
>>>>>>> 
>> https://pypi.org/project/apache-airflow-providers-exasol/4.10.5rc1/
>>>>>>> https://pypi.org/project/apache-airflow-providers-fab/3.8.0rc1/
>>>>>>> https://pypi.org/project/apache-airflow-providers-git/0.4.2rc1/
>>>>>>> 
>> https://pypi.org/project/apache-airflow-providers-google/22.3.0rc2/
>>>>>>> 
>>> https://pypi.org/project/apache-airflow-providers-hashicorp/4.8.0rc1/
>>>>>>> https://pypi.org/project/apache-airflow-providers-imap/3.12.1rc1/
>>>>>>> 
>> https://pypi.org/project/apache-airflow-providers-keycloak/0.9.0rc1/
>>>>>>> 
>>>>>>> 
>>>>> 
>>> 
>> https://pypi.org/project/apache-airflow-providers-microsoft-azure/14.1.0rc1/
>>>>>>> https://pypi.org/project/apache-airflow-providers-neo4j/3.12.1rc1/
>>>>>>> https://pypi.org/project/apache-airflow-providers-openai/1.8.2rc1/
>>>>>>> 
>>>>> 
>>> https://pypi.org/project/apache-airflow-providers-openlineage/2.20.0rc2/
>>>>>>> 
>>>>> 
>> https://pypi.org/project/apache-airflow-providers-opensearch/1.11.2rc1/
>>>>>>> 
>>> https://pypi.org/project/apache-airflow-providers-papermill/3.13.2rc1/
>>>>>>> 
>> https://pypi.org/project/apache-airflow-providers-postgres/7.0.1rc1/
>>>>>>> 
>> https://pypi.org/project/apache-airflow-providers-presto/5.12.1rc1/
>>>>>>> 
>>>>> 
>> https://pypi.org/project/apache-airflow-providers-salesforce/5.14.2rc1/
>>>>>>> https://pypi.org/project/apache-airflow-providers-sftp/6.0.1rc1/
>>>>>>> https://pypi.org/project/apache-airflow-providers-smtp/3.0.3rc1/
>>>>>>> 
>>> https://pypi.org/project/apache-airflow-providers-snowflake/6.16.0rc1/
>>>>>>> https://pypi.org/project/apache-airflow-providers-ssh/6.0.1rc1/
>>>>>>> 
>>> https://pypi.org/project/apache-airflow-providers-standard/1.17.0rc1/
>>>>>>> 
>> https://pypi.org/project/apache-airflow-providers-telegram/4.9.6rc1/
>>>>>>> 
>> https://pypi.org/project/apache-airflow-providers-teradata/3.6.2rc1/
>>>>>>> https://pypi.org/project/apache-airflow-providers-trino/6.6.1rc1/
>>>>>>> 
>> https://pypi.org/project/apache-airflow-providers-weaviate/3.4.1rc1/
>>>>>>> https://pypi.org/project/apache-airflow-providers-yandex/4.5.1rc1/
>>>>>>> 
>>>>>>> Cheers,
>>>>>>> Jarek Potiuk
>>>>>>> 
>>>>>>> 
>> ---------------------------------------------------------------------
>>>>>>> To unsubscribe, e-mail: [email protected]
>>>>>>> For additional commands, e-mail: [email protected]
>>>>>>> 
>>>>>>> 
>>>>> 
>>>>> 
>>> 
>> 

Reply via email to