rammmaa opened a new pull request, #77099:
URL: https://github.com/apache/airflow/pull/77099

   Explain in the scheduler tuning docs what the scheduler does with the 
`max_tis_per_query` batch.
   
   The current text only says that the value should not exceed 
`core.parallelism`. It does not say that task instances blocked by pool, Dag 
run concurrency or task concurrency limits still take up room in the batch. 
Operators who see runnable lower-priority task instances wait in the scheduled 
state behind blocked high-priority ones get no hint that this is the setting to 
raise.
   
   The new text describes the current behaviour:
   
   - each query selects at most this many scheduled task instances, highest 
`priority_weight` first
   - the batch is also capped by `core.parallelism` minus the executor slots 
already in use, so values above `core.parallelism` do not make it larger
   - blocked candidates take up room in the batch; the scheduler repeats the 
query without the pools, Dag runs and tasks it found blocked, but only a 
limited number of times per loop once it has found a task instance that can run
   
   Documentation only; no behaviour change.
   
   ---
   
   ##### Was generative AI tooling used to co-author this PR?
   
   - [X] Yes — Claude Code (Fable 5.2)
   
   Generated-by: Claude Code (Fable 5.2) following [the 
guidelines](https://github.com/apache/airflow/blob/main/contributing-docs/05_pull_requests.rst#gen-ai-assisted-contributions)
   


-- 
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