zhangyue19921010 opened a new issue, #11231: URL: https://github.com/apache/paimon/issues/11231
### Search before asking - [x] I searched in the issues and found no matching report. ### Paimon version master at `33dbcfcb17eae5a9c89fd3c7e087df7e6a3957a0`. ### Compute Engine Java API. The affected planner is also called by Flink and Spark incremental clustering. ### Minimal reproduce step Use a partitioned, bucket-unaware append table with incremental clustering, an explicit partition predicate, and `clustering.history-partition.idle-to-full-sort=2h`. Run the planner with JVM timezone `America/New_York`. At `2026-03-08T07:30:00Z` (03:30 after the spring transition), a partition last written at `06:00:00Z` has been idle for 90 minutes. It is nevertheless selected for historical full clustering outside the requested partition predicate. At `2026-11-01T07:30:00Z` (02:30 after the autumn transition), a partition last written at `04:45:00Z` has been idle for 165 minutes. It is incorrectly omitted. The regression writes real files with controlled creation times, verifies persisted and aggregated timestamps, then checks the compaction units produced by `IncrementalClusterManager`. ### What doesn't meet your expectations? A two-hour idle duration should represent two elapsed hours. `HistoryPartitionCluster` subtracts it from a local date-time before applying the timezone, moving the cutoff by an hour when the interval crosses a DST change. This can trigger premature work or delay eligible historical clustering. ### Anything else? Compute the cutoff from the current instant minus the duration. Keep the existing strict comparison and file metadata representation. ### Are you willing to submit a PR? - [x] I'm willing to submit a PR! -- 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]
