github-actions[bot] commented on code in PR #65854:
URL: https://github.com/apache/doris/pull/65854#discussion_r3628179279


##########
be/src/cloud/cloud_cumulative_compaction.cpp:
##########
@@ -539,6 +567,19 @@ Status 
CloudCumulativeCompaction::pick_rowsets_to_compact() {
                                  config::cumulative_compaction_min_deltas, 
&_input_rowsets,
                                  &_last_delete_version, &compaction_score);
 
+    if (config::enable_cloud_single_rowset_compaction) {
+        for (const auto& rowset : _input_rowsets) {

Review Comment:
   [P1] Keep grouped candidates selectable through promotion
   
   This loop only sees the size policy's final `_input_rowsets`, so the policy 
can discard the exact rowset this feature targets. For example, with the 
default max score 1000, let v2 be a 1 GiB overlapping rowset with 600 segments 
and v3/v4 be two 1 MiB rowsets. The size-level loop removes v2 because it is 
larger than the remaining suffix, then clears the two small rowsets because 
their score/size is below the normal thresholds. `_input_rowsets` is empty 
here, the cumulative point stays at v2, and every eligible scheduler pass 
repeats. A one-shot pre-pruning check is not sufficient: grouped commit retains 
the cumulative point, and a typical roughly 10-segment overlapping replacement 
is then below the 512-segment trigger and can be level-pruned again ahead of 
the same suffix, so it never reaches ordinary singleton compaction/promotion. 
Please preserve the policy's delete/NOTREADY gates while keeping v2 selectable 
through the grouped rewrite and follow-up promotion, with an end-to-end 
 large-head plus small-suffix test that advances the cumulative point.



-- 
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