Technologies/Apache DataFusion/datafusion.parquet.pushdown_rows_filtered
Apache DataFusionApache DataFusionMetric

datafusion.parquet.pushdown_rows_filtered

Rows filtered by pushdown
Dimensions:None
Available on:Native (1)
Interface Metrics (1)
Native
Number of rows filtered by pushdown predicates at the file format level
Dimensions:None

Technical Annotations (22)

Configuration Parameters (6)
datafusion.execution.parquet.pushdown_filtersrecommended: false
default to avoid regressions until performance issues resolved
datafusion.execution.target_partitionsrecommended: 1
set to 1 during benchmarking to reduce variability
datafusion.execution.parquet.binary_as_stringrecommended: true
required for ClickBench data processing
datafusion.optimizer.enable_aggregate_dynamic_filter_pushdownrecommended: false
disable to avoid aggregate filter overhead in Parquet reader
datafusion.execution.parquet.reorder_filtersrecommended: false
Default false; enable to optimize filter ordering when pushdown enabled
with_parquet_pruningrecommended: True
Enables row group and page-level pruning for Parquet files
Error Signatures (1)
lit(true)log pattern
CLI Commands (1)
set datafusion.optimizer.enable_aggregate_dynamic_filter_pushdown = false;diagnostic
Technical References (14)
ArrowPredicate APIcomponentlate materializationconceptRowFiltercomponentdatafusion/physical-plan/src/aggregates/no_grouping.rsfile pathdatafusion/pruning/src/file_pruner.rsfile patharrow-rscomponentpage index pruningconceptfilter_pushdown.rsfile pathphysical-optimizer/filter_pushdown.rsfile pathFileOpener::opencomponentpartition boundsconceptdynamic filterconceptrow group filteringconceptpredicate pushdownconcept
Related Insights (7)
Parquet filter pushdown causes query slowdowns for specific query patternswarning
Aggregate dynamic filters cause overhead when pushed to Parquet readerwarning
Low-selectivity filters cause unnecessary row-level evaluation overheadinfo
Parquet predicate evaluation errors lost when pushdown filters enabledinfo
Disabled Parquet filter pushdown reduces query performancewarning
Dynamic filters not generated or built on wrong table due to lost statisticswarning
Race condition in HashJoin dynamic filter construction prevents immediate probe-side filter visibilitywarning