github-actions[bot] commented on code in PR #65470:
URL: https://github.com/apache/doris/pull/65470#discussion_r3679611746
##########
be/src/cloud/cloud_cumulative_compaction_policy.cpp:
##########
@@ -134,11 +135,19 @@ int64_t
CloudSizeBasedCumulativeCompactionPolicy::pick_input_rowsets(
// Keep at least 1 rowset to avoid removing the only rowset
(consistent with fallback branch)
while (input_rowsets->size() > 1 &&
*compaction_score > static_cast<size_t>(max_compaction_score)) {
- auto& last_rowset = input_rowsets->back();
- *compaction_score -=
last_rowset->rowset_meta()->get_compaction_score();
- total_size -= last_rowset->rowset_meta()->total_disk_size();
+ last_popped = std::move(input_rowsets->back());
+ *compaction_score -=
last_popped->rowset_meta()->get_compaction_score();
+ total_size -= last_popped->rowset_meta()->total_disk_size();
input_rowsets->pop_back();
}
+ // A single non-overlapping rowset cannot be compacted by itself.
Restore the direct
+ // successor and accept a one-off max-score overshoot to keep the
input mergeable.
+ if (input_rowsets->size() == 1 && last_popped != nullptr &&
Review Comment:
[P1] Preserve mergeability after cloud transaction truncation
`CloudCumulativeCompaction::pick_rowsets_to_compact()` validates this
restored pair before calling `apply_txn_size_truncation_and_log()`. If the
pair's serialized metadata exceeds `compaction_txn_max_size_bytes`, that later
step keeps the first rowset and drops this successor, but still returns
success. With the new test's empty `[13-56]` predecessor, compaction then
recreates the same zero-byte version and `new_cumulative_point()` remains 13
(including MoW, because the 43-version span is below the default promotion
count), so the next run repeats while `[57-57]` stays unprocessed. Please
preserve or revalidate the cumulative mergeability invariant after
transaction-size limiting, and cover this at `CloudCumulativeCompaction` level
with a forced-small transaction threshold.
--
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]
