luwei16 commented on code in PR #65810:
URL: https://github.com/apache/doris/pull/65810#discussion_r3690231158
##########
be/src/storage/compaction/compaction.cpp:
##########
@@ -603,15 +605,18 @@ bool CompactionMixin::handle_ordered_data_compaction() {
}
if (is_binlog_compaction) {
+ DCHECK(!_input_rowsets.empty()) << "tablet=" << _tablet->tablet_id();
+ auto compaction_level =
_input_rowsets.front()->rowset_meta()->compaction_level();
bool can_quick_merge_binlog =
- compaction_level() ==
BinlogCompactionPolicy::kBinlogCompactionMaxLevel - 1 &&
+ compaction_level ==
+
BinlogCumulativeCompactionPolicy::kBinlogCompactionMaxLevel - 1 &&
_input_rowsets.size() >= 2 &&
_input_rowsets[0]->start_version() == 0;
if (!can_quick_merge_binlog) {
return false;
}
// Binlog quick merge at LMax is a special meta/link compaction path
selected by
- // BinlogCompactionPolicy. It does not require the whole input to
satisfy the normal
+ // BinlogCumulativeCompactionPolicy. It does not require the whole
input to satisfy the normal
// ordered-data tidy check, but the output rowset built by
do_compact_ordered_rowsets()
// is still NONOVERLAPPING.
auto st = do_compact_ordered_rowsets();
Review Comment:
> Could this remain NONOVERLAPPING for row-binlog quick compaction? Linked
L2 rowsets may have overlapping key ranges because row-binlog uses DUP_KEYS and
TSO/LSN are not key columns. The previous
> BinlogCompaction::modify_rowsets() explicitly changed quick-compaction
output to OVERLAPPING, while the new binlog policy still documents that
requirement. It looks like this behavior was lost when the dedicated
implementation
> was removed. Please preserve OVERLAPPING for this path and add an
execution test with repeated keys across versions. We may also need to ensure
RowsetMeta::is_segments_overlapping() recognizes non-singleton row-binlog quick-
> compaction outputs.
--
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]
