roryqi commented on PR #14675: URL: https://github.com/apache/gravitino/pull/14675#issuecomment-6675727216
> @roryqi For partitioned tables, `builtin-iceberg-update-stats` refreshes **partition** statistics only and does not rewrite the corresponding table-level keys in `statistic_meta`. > > Previously, partition trigger/score evaluation still seeded **table-level** statistics into the expression context. After an update-stats job, those table-level values can stay stale (e.g. a large `custom-data-file-mse`) while every partition is already below the threshold — so `submit-strategy-jobs` could still recommend work incorrectly. > > So for partition-aware strategies on partitioned tables we only evaluate with: > > * partition statistics > * table metadata + policy rules > > Non-partitioned tables are unchanged and still use table-level statistics. I can't still understand this point. I will use codex to analyze this PR later to help me understand. -- 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]
