Considerations when Using GPORCA
To run queries optimally with GPORCA, consider the query criteria closely.
Ensure the following criteria are met:
- The table does not contain multi-column partition keys.
- The multi-level partitioned table is a uniform multi-level partitioned table. See About Uniform Multi-level Partitioned Tables.
The server configuration parameter
optimizer_enable_master_only_queriesis set to
onwhen running against master only tables such as the system table pg_attribute. For information about the parameter, see the Greenplum Database Reference Guide.
Note: Enabling this parameter decreases performance of short running catalog queries. To avoid this issue, set this parameter only for a session or a query.
Statistics have been collected on the root partition of a partitioned table.
If the partitioned table contains more than 20,000 partitions, consider a redesign of the table schema.
These server configuration parameters affect GPORCA query processing.
optimizer_cte_inlining_boundcontrols the amount of inlining performed for common table expression (CTE) queries (queries that contain a
optimizer_force_comprehensive_join_implementationaffects GPORCA’s consideration of nested loop join and hash join alternatives. When the value is
false(the default), GPORCA does not consider nested loop join alternatives when a hash join is available.
optimizer_force_multistage_aggforces GPORCA to choose a multi-stage aggregate plan for a scalar distinct qualified aggregate. When the value is
off(the default), GPORCA chooses between a one-stage and two-stage aggregate plan based on cost.
optimizer_force_three_stage_scalar_dqaforces GPORCA to choose a plan with multistage aggregates when such a plan alternative is generated.
optimizer_join_ordersets the query optimization level for join ordering by specifying which types of join ordering alternatives to evaluate.
optimizer_join_order_thresholdspecifies the maximum number of join children for which GPORCA uses the dynamic programming-based join ordering algorithm.
optimizer_nestloop_factorcontrols nested loop join cost factor to apply to during query optimization.
optimizer_parallel_unioncontrols the amount of parallelization that occurs for queries that contain a
UNION ALLclause. When the value is
on, GPORCA can generate a query plan of the child operations of a
UNION ALLoperation run in parallel on segment instances.
optimizer_sort_factorcontrols the cost factor that GPORCA applies to sorting operations during query optimization. The cost factor can be adjusted for queries when data skew is present.
gp_enable_relsize_collectioncontrols how GPORCA (and the Postgres Planner) handle a table without statistics. By default, GPORCA uses a default value to estimate the number of rows if statistics are not available. When this value is
on, GPORCA uses the estimated size of a table if there are no statistics for the table.
This parameter is ignored for a root partition of a partitioned table. If the root partition does not have statistics, GPORCA always uses the default value. You can use
ANALZYE ROOTPARTITIONto collect statistics on the root partition. See ANALYZE.
These server configuration parameters control the display and logging of information.
optimizer_print_missing_statscontrols the display of column information about columns with missing statistics for a query (default is
optimizer_print_optimization_statscontrols the logging of GPORCA query optimization metrics for a query (default is
For information about the parameters, see the Greenplum Database Reference Guide.
GPORCA generates minidumps to describe the optimization context for a given query. The minidump files are used by VMware support to analyze Greenplum Database issues. The information in the file is not in a format that can be easily used for debugging or troubleshooting. The minidump file is located under the master data directory and uses the following naming format:
For information about the minidump file, see the server configuration parameter
optimizer_minidump in the Greenplum Database Reference Guide.
EXPLAIN ANALYZE command uses GPORCA, the
EXPLAIN plan shows only the number of partitions that are being eliminated. The scanned partitions are not shown. To show the name of the scanned partitions in the segment logs set the server configuration parameter
on. This example
SET command enables the parameter.
SET gp_log_dynamic_partition_pruning = on;
Parent topic:About GPORCA