-1 (non-binding)

Sorry for raising this so late, with many binding +1 votes already in and
the vote ending today.

I found a regression in rc2, as I reported in the tracking issue [1]: the
API server stalls when it gets many write requests at the same time. 3.3.2
doesn't have this problem.
(Some requests fail with 500, and running tasks can fail because their
heartbeats time out)

Jason already confirmed it and proposed a fix in [2]. It is a small change,
so I think it is worth including in 3.4.0.
Though, it only shows up with many write requests at the same time, so I
have no objection if we think fixing it in 3.4.1 is the better choice.

Thanks a lot for running the release!

[1] https://github.com/apache/airflow/issues/75865#issuecomment-6622866800
[2] https://github.com/apache/airflow/pull/76416

Andrew

Jarek Potiuk <[email protected]> 於 2026年11月4日週三 上午7:28寫道:

> +1 (binding)
>
> Checked Airflow 3.4.0rc2 and Task SDK 1.4.0rc2:
>
> - SVN: all expected artifacts present
> - Signatures: all artifacts and both release tags
> - SHA-512 checksums
> - Licenses: Apache RAT reports no unknown or unapproved files
> - Reproducible build: rebuilt all seven artifacts from the tag, all
>   byte-for-byte identical to the staged ones
>
> I also installed the RC from PyPI with the 3.4.0rc2 constraints in Breeze
> (Python 3.11, Postgres, LocalExecutor) and ran a test Dag through the
> scheduler, Dag processor, triggerer and API server. It covered dynamic task
> mapping, branching, retries, a deferrable sensor and an asset-triggered
> Dag.
> All worked as expected.
>
> J.
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: [email protected]
> For additional commands, e-mail: [email protected]
>
>

Reply via email to