Starburst optimizer properties#
SEP includes additional configuration and session properties to configure the included extensions of the optimizer.
Configuration properties on this page are set as follows:
Kubernetes deployments - Set under the
config.properties:
nodes in coordinator and worker sections of the SEP Helm chart configuration.Other deployments - Set in the config.properties file.
Carefully monitor the impacts of changes of these expert-level configurations. The following properties can be configured:
optimizer.join-partitioned-build-min-row-count
#
Type: integer
Default value:
1000000
Session property:
join_partitioned_build_min_row_count
Set the minimum number of rows required on the build side of a join to partition
the table. Because partitioning of join hash table incurs overhead when
processing the probe side data of the join data during query runtime, setting
this property allows SEP to skip partitioning for tables too small to benefit
from parallel hash table construction. Build side hash tables for joins can be
either a single unpartitioned table or partitioned into the number of buckets
defined by task_concurrency
.
optimizer.pre-aggregate-case-aggregations.enabled
#
Type: boolean
Default value:
true
Session property:
pre_aggregate_case_aggregations
Enable pre-aggregation of CASE
statements that contain multiple, similar
CASE
statements. Disabling this causes performance degradation in most
cases. Only disable the feature if you have a very specific, technical need to
do so, and only until that need is resolved.
optimizer.accelerated-repartitioning.enabled
#
Type: boolean
Default value:
true
Session property:
accelerated_repartitioning_enabled
Use the accelerated operator for repartitioning data across worker nodes to increase parallel query processing performance that require repartitioning.