Copilot commented on code in PR #7539: URL: https://github.com/apache/texera/pull/7539#discussion_r3762485380
########## common/workflow-core/src/main/scala/org/apache/texera/amber/core/storage/result/iceberg/IcebergDocument.scala: ########## @@ -72,7 +72,10 @@ private[storage] class IcebergDocument[T >: Null <: AnyRef]( private val lock = new ReentrantReadWriteLock() - @transient lazy val catalog: Catalog = IcebergCatalogInstance.getInstance(warehouse) + // Resolved per use, never held: the catalog cache is bounded and closes evicted + // entries (#7290), so a pinned reference could outlive its catalog. A public def + // (not a lazy val) also means a replaced/rebuilt catalog is picked up immediately. + def catalog: Catalog = IcebergCatalogInstance.getInstance(warehouse) Review Comment: This still leaves long-lived document readers holding a catalog-derived `Table`. `getUsingFileSequenceOrder` initializes its `table` once (lines 173–174), and later `seekToUsableFile` calls `table.refresh()` without resolving this accessor again (lines 198–216). An actively polled reader therefore looks idle to this cache; after 60 minutes another cache operation can expire and close its REST catalog, and the next refresh uses the closed `RESTTableOperations`. Resolve a fresh table from the current catalog at each polling/refresh boundary (while retaining the snapshot ID), or keep a lease for the iterator's lifetime. -- 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]
