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 allowNULL
as the result of input/combine functions, while also allowing only constant expressions and notNULL
as 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_strategy
session property or theexperimental.resource-aware-scheduling-strategy
configuration property toTTL
. See #16409.Add adjusted queue size information to the
v1/cluster
endpoint.
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
NodeTtlFetcher
andClusterTtlProvider
interfaces respectively. See #16409.
Hive Changes
Fix issue while reading Apache HUDI tables when
PrestoFileSystemCache
is refreshed. The issue occurs becauseHoodieROTablePathFilter
is cached with defaultConfiguration
object (#16611).Add session property
hive.cache_enabled
to 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