atognolas opened a new pull request, #39155: URL: https://github.com/apache/beam/pull/39155
## Summary - Add a `Map<String, Row> keyCache` in `AssignDestinationsAndPartitions.AssignDoFn` to reuse Row key objects across millions of input rows - Only N distinct `(destination, partition)` combinations exist (e.g. 400 for a 400-bucket table), but every row currently allocates a new Row with schema validation overhead - After the first N rows, all subsequent lookups are cache hits ## Motivation Observed during production benchmarks at scale (39M users, 400 partitions, 99-column schema). Row allocation + schema validation dominated bundle processing time unnecessarily. ## Test plan - [ ] Existing `AssignDestinationsAndPartitionsTest` passes - [ ] Run IcebergIO integration tests - [ ] Verify via profiling that Row allocation drops after initial cache warm-up 🤖 Generated with [Claude Code](https://claude.com/claude-code) -- 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]
