luis4a0 commented on PR #12963: URL: https://github.com/apache/gluten/pull/12963#issuecomment-6492473349
@rui-mo, the Kryo failure was caused by my singleton `Seq(0)` seed. Spark's `ParallelCollectionRDD` converts that sequence into an `ArraySeq$ofInt` partition payload, which fails under required Kryo registration. Commit: https://github.com/apache/gluten/commit/1e7c30de986fd3f030095ce6e4837d17978c407c I changed only the seed to `0 until 1`, using Spark's range-preserving slicing path. It still supplies exactly one partition and one zero-column batch. No Kryo class registrations were added, and `spark.kryo.registrationRequired` remains true. The original SPARK-51777 test and a new explicit OneRow regression both reproduced the failure before the fix. The full `GlutenCacheTableInKryoSuite` now passes all four tests on Spark 4.1 and 4.2; the Spark 4.1 suite also passes on JDK 21. The new regression checks actual serializer settings, literal/NULL values, one partition, and native OneRow presence. Normal Spark 4.1 operator coverage (106 executed tests) and Spark 4.2 runtime/EXPLAIN coverage also pass. I also restored the original ValuesNode operator-map row verbatim, with OneRow as a separate row. Fresh upstream CI is still required. The ClickHouse status remains a separate unknown: its linked log server returned HTTP 403, and the public status only reports a generic x86 failure. I have not labeled that a flake or bypassed its check. -- 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]
