andygrove opened a new issue, #5938:
URL: https://github.com/apache/datafusion-comet/issues/5938

   ## What is the problem the feature request solves?
   
   The PR tier still runs Spark 4.1 (catalyst and `sql_core`) and Iceberg 1.11 
on every push to every pull request. That made sense when a PR was pushed a 
handful of times before merging. Two things changed:
   
   - Pull request review is now largely agent-driven, and so is responding to 
review feedback. A PR goes through several more push-review-push rounds before 
it is queued, and each round re-runs both suites.
   - There are more contributors, so there are more PRs in flight at once.
   
   A sample of the last 100 `pull_request` runs of `ci.yml` (2026-09-13 20:35 
UTC to 2026-09-15 06:15 UTC, about 34 hours):
   
   | Suite                          | Ran in | Runner-minutes | Share of PR 
total | Sole failing suite |
   | ------------------------------ | ------ | -------------- | 
----------------- | ------------------ |
   | Spark SQL (4.1, core)          | 36     | 5,679          | 16%             
  | 0 runs             |
   | Iceberg 1.11                   | 35     | 5,736          | 17%             
  | 0 runs             |
   | Linux build                    | 42     | 20,197         | 58%             
  | 1 run              |
   | all jobs                       |        | 34,597         |                 
  |                    |
   
   That is roughly 100 PR runs and 136 runner-hours a day spent on the two 
suites, and in this window neither one was the job that turned a PR run red. 
The merge queue runs both against the merge result before anything lands 
regardless.
   
   ## Describe the potential solution
   
   Move every Spark SQL and Iceberg suite behind the merge queue by default, 
leaving the Linux build as the whole PR tier:
   
   - `spark_4_1`: `["queue", "label:run-spark-4.1-tests"]`
   - `spark_4_1_hive`: also gated by `run-spark-4.1-tests`, so that label runs 
every module in the one 4.1 build while `run-spark-4.1-hive-tests` still 
selects only the hive shards
   - `iceberg_1_11`: `["queue", "label:run-iceberg-tests"]`, so one label opts 
a PR into all four Iceberg versions
   
   Create the `run-spark-4.1-tests` label (it did not exist, despite the 
workflows README saying it "gates nothing"), pin the new routing in 
`dev/ci/check-ci-config.py`, and update the workflows README and the 
contributor guide.
   
   The trade-off is that a Spark SQL or Iceberg regression now surfaces in the 
queue rather than on the PR, where a red run evicts the entry. Reviewers should 
apply the labels before queuing anything that touches the serde, planner, 
native operators, or the Iceberg paths. This continues #5870 and #5929.
   
   ## Additional context
   
   Follows the tiering in #5843, #5871 and #5930.
   


-- 
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]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to