xiangfu0 opened a new pull request, #19536: URL: https://github.com/apache/pinot/pull/19536
Match-all and sorted-range queries currently fill document blocks through one `next()` call per row. Add a compatible batch API with a scalar default and contiguous fast paths, then use it from `DocIdSetOperator` with the existing scratch array and EOF/checkpoint/accounting behavior. On a warm, 3,000,017-row fixture with two separately executed server responses, three alternating JMH pairs show 1.65–1.80× sorted-range, 1.13× match-all SUM/AVG, and 1.12× GROUP BY speedups. Bitmap, predicate-scan, metadata COUNT and LIMIT 1 controls are retained; the largest individual pair slowdown is 1.508%. Timing includes SQL preparation, server plans/execution, serialization/deserialization and broker reduction, and excludes HTTP/admission/concurrent-server capacity. Allocation is mostly unchanged. This is not a production QPS or universal all-query speedup claim. Validation: 148 tests across 19 classes; Spotless, checkstyle, license format/check and Apache RAT; standard compilation/package with compiler warnings reviewed; independent source and benchmark review. The benchmark owns and closes its executor; an earlier leaking-harness pilot was excluded and retained separately. All other runtime bytes and the fixture match across arms. Reproduction and tutorial: `pinot-perf/README.md`, “Document-ID batch query benchmark,” includes build, fixture generation and JMH commands. Full protocol, controls, all raw measurements and source/runtime manifests are in `pinot-perf/benchmark-results/docid-batching-20260912/`. -- 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]
