+1 (non-binding). Coincidentally I was just discussing the need for something similar to this in a meeting earlier today. Would love to see it as part of the Airflow public interface.
On Mon, Oct 5, 2026 at 9:36 AM Jens Scheffler via dev < [email protected]> wrote: > +1 > > On 05.10.26 11:45, Shahar Epstein wrote: > > +1 (binding), thanks for leading this effort! > > > > > > Shahar > > > > On Thu, Oct 1, 2026, 11:00 Blain David <[email protected]> wrote: > > > >> Hi everyone, > >> > >> I would like to restart the vote on AIP-104, now titled Iterable Tasks > >> (IT) and Task Spreading (TS): > >> > >> > >> > https://cwiki.apache.org/confluence/spaces/AIRFLOW/pages/421954527/AIP-104+Iterable+Tasks+IT+and+Task+Spreading+TS > >> > >> The previous vote, under the name "Task Iteration and Dynamic Task > >> Batching", ran during the summer and did not get much engagement: > >> > >> https://lists.apache.org/thread/x5bd26zfqkkhp7sx7nb0crjlsctf2z5j > >> > >> The AIP has changed since then, so please vote on the current page > rather > >> than on what you may remember: > >> > >> - Renamed "Task Iteration (TI)" to "Iterable Tasks (IT)", as TI is > already > >> widely used for Task Instance (as remarked by Daniel). > >> - Renamed "Dynamic Task Batching (DTB)" to "Task Spreading (TS)". The > >> primitive is now .spread(across=N) instead of .batch(size=N), as "size" > >> read like a chunk length while it is the number of Task Instances. Items > >> are distributed round robin over those N Task Instances. The new name > was > >> discussed and voted on in separate threads (linked below). > >> - Iterable Tasks now checkpoint progress per item through AIP-103 Task > >> State. On a retry, or after a worker crash, items that already finished > are > >> not run again; only the items that failed or were in flight are. > >> - The page now documents task_concurrency, the setting that controls how > >> many items run at once inside one Task Instance. > >> - The page now lists the current limits: deferrable operators, > >> reschedule-mode sensors and TriggerDagRunOperator are not supported > inside > >> .iterate() yet (deferrable support is planned in a separate PR); > >> execution_timeout applies per item only to async operators; > >> multiple_outputs is ignored; and pools count the Task Instance rather > than > >> its items. > >> > >> The previous discussion threads can be found here: > >> > >> https://lists.apache.org/thread/ztnfsqolow4v1zsv4pkpnxc1fk0hbf2p > >> https://lists.apache.org/thread/tr3x4sx55k0fk2coq28xs2648m1p65cc > >> > >> The naming of the spread primitive: > >> > >> https://lists.apache.org/thread/6qlbvvnt76514rstzk447pvs4784wcj2 > >> (discussion) > >> https://lists.apache.org/thread/rrc9bmhh5dcpyj4rcpjr2lj1q8vjdoxp (vote) > >> > >> Thank you again to everyone who reviewed the AIP and took part in the > >> discussions. A special thank you to Kaxil for the many detailed reviews, > >> and to TP, Jens, Jarek and Ash for their comments. Thanks as well to > >> Daniel, Jedidiah and Amogh for their more recent reviews on the AIP > page, > >> and to everyone who voted on the naming of the spread primitive. > >> > >> The vote will remain open for 5 days and will close on Wednesday, 7 > >> October 2026, at 00:00 GMT. > >> > >> Everyone is encouraged to vote, although only votes from PMC members and > >> committers are considered binding. > >> > >> Please vote accordingly: > >> > >> [ ] +1 Approve > >> [ ] +0 no opinion > >> [ ] -1 disapprove with the reason > >> > >> Please consider this my +1 (binding). > >> > >> Regards, > >> David > >> > >> General (Internal Property) > >> > > --------------------------------------------------------------------- > To unsubscribe, e-mail: [email protected] > For additional commands, e-mail: [email protected] > >
