Release 0.254

Warning

There is a backward compatibility issue in the DWRF writer that might cause other engines to be unable to read files written by this release.

Details

General Changes

  • Fix a bug where queries that have both remote functions and a local function with only constant arguments could throw an IndexOutOfBoundException during planning. The bug was introduced by #16039.

  • Fix a CPU regression for queries using element_at() for MAP. Introduced by #16027.

  • Add fragment result caching support for UNNEST queries.

  • Add poisson_cdf() and inverse_poisson_cdf() functions.

  • Add memory tracking in TableFinishOperator which can be enabled by setting the table-finish-operator-memory-tracking-enabled configuration property to true. Enabling this property can help with investigating GC issues on the coordinator by allowing us to debug whether stats collection uses a lot of memory.

  • Remove spilling strategy PER_QUERY_MEMORY_LIMIT and add configuration property experimental.query-limit-spill-enabled and session property query_limit_spill_enabled. When query_limit_spill_enabled is set to true and the spill strategy is not PER_TASK_MEMORY_THRESHOLD, then we will spill whenever a query uses more than the per-node total memory limit in combined revocable and non-revocable memory.

Hive Changes

  • Fix a bug where the files would not be sorted when inserting into bucketed sorted tables with Glue.

  • Add support for validating the values returned from the partition cache with the actual value from Metastore. This can be enabled by setting the configuration property hive.partition-cache-validation-percentage.

  • Add support for allowing to match columns between table and partition schemas by names when the configuration property hive.parquet.use-column-names or the hive catalog session property parquet_use_column_names is set to true. By default they are mapped by index.

  • Add support for configuring the Glue endpoint URL. Hive Connector.

  • Add support for accessing tables in Glue metastore that do not have a table type.

  • Add support for the S3 Intelligent-Tiering storage class writing data. This can be enabled by setting the configuration property hive.s3.storage-class to INTELLIGENT_TIERING.

  • Add configuration property hive.metastore.glue.max-error-retries for the maximum number of retries for glue client connections. The default value is 10. Hive Connector.

Presto On Spark Changes

  • Optimize Driver commit memory footprint.

  • Add session property spark_memory_revoking_threshold and configuration property spark.memory-revoking-threshold. Spilling is triggered when total memory is beyond this threshold.

SPI Changes

  • Add support for custom query prerequisites to be checked and satisfied through QueryPrerequisites interface. See #16073.

Contributors

Abhisek Gautam Saikia, Akhil Umesh Mehendale, Andrii Rosa, Arjun Gupta, Beinan, Bhavani Hari, Chunxu Tang, Jalpreet Singh Nanda (:imjalpreet), James Petty, James Sun, Ke Wang, Maria Basmanova, Mayank Garg, Nikhil Collooru, Rebecca Schlussel, Rohit Jain, Rongrong Zhong, Sergey Pershin, Sergii Druzkin, Shixuan Fan, Tal Galili, Tim Meehan, Vic Zhang, Zhenxiao Luo, guhanjie, linjunhua, v-jizhang