JingsongLi commented on code in PR #11232:
URL: https://github.com/apache/paimon/pull/11232#discussion_r4624047893


##########
paimon-core/src/main/java/org/apache/paimon/append/cluster/HistoryPartitionCluster.java:
##########
@@ -129,12 +128,7 @@ public Map<BinaryRow, CompactUnit> 
pickForHistoryPartitions() {
 
     @VisibleForTesting
     public Map<BinaryRow, List<LevelSortedRun>> 
constructLevelsForHistoryPartitions() {
-        long historyMilli =
-                LocalDateTime.now()
-                        .minus(historyPartitionIdleTime)
-                        .atZone(ZoneId.systemDefault())
-                        .toInstant()
-                        .toEpochMilli();
+        long historyMilli = 
Instant.now().minus(historyPartitionIdleTime).toEpochMilli();

Review Comment:
   [P2] Avoid early clustering for files created during the repeated hour
   
   The new cutoff is an instant, but `DataFileMeta` persists a local date-time, 
and `creationTimeEpochMillis()` resolves an ambiguous time to the earlier 
offset. In a real partition test for `America/New_York`, I wrote the history 
file at `2026-11-01T06:15:00Z` (the second 01:15) and planned at `07:45Z` with 
a two-hour threshold. The stored timestamp is read back as `05:15Z`. Master 
excluded the 90-minute-old partition. This head selected it because the new 
cutoff is `05:45Z`. Please handle the overlap when translating existing file 
timestamps or use unambiguous creation metadata, and add a case with file 
creation inside the second occurrence of the hour.



-- 
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