Hi Everyone,

I would like to propose suspending the apache-beam provider, until the
grpcio version limitation in Apache Beam is resolved upstream.

The current apache-beam Python SDK imposes a restrictive upper bound on
grpcio (e.g., <1.66), which creates significant dependency resolution
conflicts. This becomes problematic when our CI uv sync, and it blocks
progress on several fronts.

I have observed few key issues:

Conflicts arise during dependency resolution when working on AIP-99 with
the latest pydantic-ai version, resolution fails entirely, though it
succeeds using the outdated pydantic-ai 0.0.5.

Newer versions of opentelemetry-exporter-otlp / opentelemetry-proto demand
updated protobuf and grpcio.
pydantic-ai-slim (utilized in the common-ai provider) requires
opentelemetry-api >= 1.28.0.

These constraints prevent dependency upgrades across the repository and
stall development efforts, for  AIP-99. For example, PR [1] is currently
failing due to these pinned limitations.

This grpcio upper-bound restriction in Apache Beam has been a known issue
for some time, and Jarek has already raised this long ago see the upstream
issue [2] . The root cause traces back to the gRPC project [3], where fixes
have already landed, with expectations that a release like grpcio 1.78.1
(or a targeted patch) will remove most of these pins and conflicts.

In line with the Airflow provider suspension guidelines [4], I would like
to suspend apache beam provider and I have a PR draft in progress [5].

Please let me know your thoughts on this.

[1]: https://github.com/apache/airflow/pull/61794
[2]: https://github.com/apache/beam/issues/34081
[3]: https://github.com/grpc/grpc/issues/37710
[4]:
https://github.com/apache/airflow/blob/main/PROVIDERS.rst#suspending-releases-for-providers
[5]: https://github.com/apache/airflow/pull/61926

Regards,
Pavan

Reply via email to