HaoXuAI commented on issue #15660: URL: https://github.com/apache/iceberg/issues/15660#issuecomment-5326754543
We maintain a service that computes per-partition file deltas for every commit across a large table fleet, so we sit squarely in the multi-snapshot case here — and we ended up hand-rolling the manifest read, because no public API has the shape. Core has the same bypass: `PartitionStatsHandler.computeStatsDiff` flattens the manifests each snapshot wrote across an `ancestorsBetween` range and folds them in parallel with `Tasks.foreach`, never touching `SnapshotChanges`. Which suggests the gap is as much shape as multi-snapshot. For a caller that only aggregates, neither a materialized list nor an iterator fits — the dilemma you hit in #15656. A fold sidesteps it: fold each manifest's entries in the thread that read them, combine the partials, nothing cached and nothing streamed. We have that working for a single snapshot against main, with tests. Happy to extend it to a range and bring `PartitionStatsHandler` on as the in-core caller, if that is the direction you want. One signature question we would hold until then. -- 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]
