+1 (non-binding) I installed Airflow 3.4.0rc2 and Task SDK 1.4.0rc2 from PyPI in a clean Python 3.12 virtualenv with the 3.4.0rc2 constraints and compared them with 3.3.2 using dag.test() on SQLite. The fixes in #73960, #74283 (with #73959) and #69075 work as described, and each bug reproduces on 3.3.2. Details are on the testing issue: https://github.com/apache/airflow/issues/75865
Bingqin Wang On Thu, Oct 29, 2026 at 2:11 PM Rahul Vats <[email protected]> wrote: > Hi all, > > I am cancelling this vote. Due to issue > <https://github.com/apache/airflow/issues/75865#issuecomment-6483614606> > reported by Andrew above. The fix is already merged (#75943 > <https://github.com/apache/airflow/pull/75943>). I plan to cut 3.4.0rc2 > tomorrow. In the meantime, please keep testing rc1 the more we test now, > the more fixes we can fold into rc2. > > Testing status is tracked in > https://github.com/apache/airflow/issues/75865. > I have also started testing thread in the OSS slack > <https://apache-airflow.slack.com/archives/C06K9Q5G2UA/p1793289024229599>. > Please > check it for details on testing the significant changes in 3.4.0 > > Regards, > Rahul Vats > > On Thu, 29 Oct 2026 at 06:01, Jarek Potiuk <[email protected]> wrote: > > > Also a small comment from resolving a failed CI: we need to ensure we > merge > > the outstanding RCs of providers before releasing the final version of > > 3.4.0 (or other final versions) - because for example rc1 of 3.4.0 is > > tested with 1.1.0rc1 of common-ai - so we need to make sure either to > > release the final providers before or yank rcs when we gave up on > releasing > > the provider's rcs (otherwise release of 3.4.0 will fail). Fix in > > https://github.com/apache/airflow/pull/75919 > > > > J. > > > > On Thu, Oct 29, 2026 at 12:54 AM Andrew Chang <[email protected]> > > wrote: > > > > > -1 (non-binding) > > > > > > I found a scheduler bug that seems related to #70972. > > > > > > Long story short, when users use asset events to schedule their tasks > and > > > the events accumulate to a specific amount (about 7000~9000), the > > scheduler > > > could crash, and users wouldn't be able to restart their scheduler > again. > > > > > > I've reported it in the 3.4.0rc1 test tracking issue: > > > https://github.com/apache/airflow/issues/75865#issuecomment-6483614606 > > > > > > TBH, reaching this condition, 7000 to 9000 accumulated asset events, is > > not > > > very common. > > > But the consequence is quite serious -- I think it is worth another > look > > > before we release. > > > > > > Thanks, > > > Andrew > > > > > >
