Release 0.262¶
Highlights¶
Fix a correctness bug which could be triggered for queries with aggregations on partitioning columns and filters on non-partitioning columns when both optimizing metadata queries and filter pushdown are enabled.
Add support for elastic workers by implementing time-to-live based node scheduling for queries with execution time estimates. See #16409
Details¶
General Changes¶
Fix
reduce_agg()to allowNULLas the result of input/combine functions, while also allowing only constant expressions and notNULLas the initial value.Fix a correctness bug which could be triggered for queries with aggregations on partitioning columns and filters on non-partitioning columns when both optimizing metadata queries and filter pushdown are enabled.
Add default size limit (100MB) to build side of broadcast join.
Add support for elastic workers by implementing time-to-live based node scheduling for queries with execution time estimates. This can be enabled by setting either the
resource_aware_scheduling_strategysession property or theexperimental.resource-aware-scheduling-strategyconfiguration property toTTL. See #16409.Add adjusted queue size information to the
v1/clusterendpoint.
SPI Changes¶
Add support for custom node time-to-live (TTL) fetchers (to fetch TTLs of nodes) and cluster TTL providers (to compute the TTL of a cluster) through the
NodeTtlFetcherandClusterTtlProviderinterfaces respectively. See #16409.
Hive Changes¶
Fix issue while reading Apache HUDI tables when
PrestoFileSystemCacheis refreshed. The issue occurs becauseHoodieROTablePathFilteris cached with defaultConfigurationobject (#16611).Add session property
hive.cache_enabledto allow turning on and off data cache per query.
Credits¶
Ajay George, Ariel Weisberg, Arunachalam Thirupathi, Brian Li, Jalpreet Singh Nanda (:imjalpreet), James Petty, James Sun, Neerad Somanchi, Shixuan Fan, Sreeni Viswanadha, Swapnil Tailor, Tal Galili, Timothy Meehan, Vivek, Xiang Fu, Ying, Zac Wen, Zhan Yuan, Zhenxiao Luo, Zhongting Hu, abhiseksaikia