Vamsi-klu commented on issue #69543:
URL: https://github.com/apache/airflow/issues/69543#issuecomment-5463684321

   The two approaches asked about in 
https://github.com/apache/airflow/issues/69543#issuecomment-5259850995 have now 
been discussed here. I think we can pick one.
   
   Keep and fix start_from_trigger 
(https://github.com/apache/airflow/pull/69746). Do not drop it 
(https://github.com/apache/airflow/pull/69925, also 
https://github.com/apache/airflow/pull/69610).
   
   What the bug actually is: TimeSensor.__init__ runs at parse time, combines 
datetime.now() with target_time, and with start_from_trigger=True writes that 
absolute moment into serialized start_trigger_args. The hash changes every 
parse. Main already isolates start_trigger_args per instance via 
dataclasses.replace. The remaining defect is the volatile moment.
   
   #69746 implements the lazy-moment design @potiuk asked for on that PR: 
TimeOfDayTrigger serializes only parse-stable target_time + tz + 
end_from_trigger, and resolves the UTC moment when the trigger starts. 
serialize() then persists moment so a reconstruct after midnight does not move 
the wait. start_from_trigger stays in the public signature. No major bump of 
apache-airflow-providers-standard. Docs are corrected: wall-clock date in the 
Dag timezone, not data_interval_end/run_after.
   
   #69925 (and #69610) stop the feature: ValueError originally, then a property 
that always reads False and warns. That is a functional regression for anyone 
using start_from_trigger=True to skip a worker slot. @potiuk called that out as 
needing a major provider bump if we went that way. @uranusjr later wrote here 
that dropping it is suboptimal and that we should try to fix the feature first 
(https://github.com/apache/airflow/issues/69543#issuecomment-5312368669). 
@RehanAhmad25 then said he would open a new lazy-resolution PR instead of 
continuing #69925; that new PR is not up, and #69925 still deprecates the flag.
   
   #69746 has been reworked since the Aug 11 pause (TimeOfDayTrigger, 
per-instance StartTriggerArgs, hash-stability tests with two frozen clocks, DST 
spring/fall, signature kept). @potiuk dismissed his older CHANGES_REQUESTED 
today to unblock. The remaining GitHub block is the Aug 11 CHANGES_REQUESTED on 
both PRs that asked us to decide here.
   
   Proposal: take #69746 as the fix, close #69925 and #69610 as superseded, and 
leave start_from_trigger working. Happy to adjust DST policy if maintainers 
want a different one. Not proposing to drop the API.
   
   
   ---
   Drafted-by: Cursor Grok 4.6; reviewed by @Vamsi-klu before posting


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to