API, Core, Parquet: Add filter hint support to InternalData.ReadBuilder#16583
Open
kamcheungting-db wants to merge 4 commits into
Open
API, Core, Parquet: Add filter hint support to InternalData.ReadBuilder#16583kamcheungting-db wants to merge 4 commits into
kamcheungting-db wants to merge 4 commits into
Conversation
Adds InternalData.read(format, file, filterHint) as the primary entry point for reads that benefit from I/O optimization. The filter hint is passed to the format reader as a best-effort optimization — Parquet uses it for row-group skipping via column statistics; Avro ignores it. Callers must always apply a residual filter for correctness. Changes: - InternalData.ReadBuilder: add withFilterHint(Expression) default no-op - InternalData.read(format, file, filterHint): new factory overload that applies the hint only when it is not alwaysTrue(); the no-arg overload delegates to this for a single implementation - Parquet.ReadBuilder: override withFilterHint() to wire into ReadConf row-group skipping; keep filter() intact for existing callers such as ParquetFormatModel; withFilterHint() ANDs with any existing filter to avoid silently losing predicates - PartitionStatisticsScan: add caseSensitive(boolean) default method - BasePartitionStatisticsScan: implement filter(), project(), and caseSensitive() with two-phase filtering — read(format, file, filter) for row-group skipping + CloseableIterable.filter(Evaluator) for exact per-row correctness; readSchema() unions projection and filter-referenced columns so the evaluator always has the data it needs Tests: - TestInternalData: testWithFilterHintIsNoOpForAvro verifies Avro returns all rows regardless of hint; testWithFilterHintEnablesRowGroupSkipping ForParquet writes two non-overlapping id ranges into separate row groups (forced via 1-byte row group size + per-record flush check) and asserts only high-range rows are returned without a residual filter, directly proving skipping happened; testWithFilterHintAndResidualFilter ReturnsMatchingRowsOnly verifies the two-phase pattern on both formats - PartitionStatisticsScanTestBase: testFilterNullThrows, testFilterAlwaysTrueReturnsAll, testFilterAlwaysFalseReturnsEmpty, testFilterByDataFileCount, testFilterColumnNotInProjectionIsStillApplied Co-authored-by: Isaac
…down # Conflicts: # api/src/main/java/org/apache/iceberg/PartitionStatisticsScan.java # core/src/main/java/org/apache/iceberg/BasePartitionStatisticsScan.java # core/src/test/java/org/apache/iceberg/PartitionStatisticsScanTestBase.java
…ection scan - TestInternalData.testWithFilterHintSkipsAllRowGroupsWhenNoneMatchForParquet: with one row group per record and no residual filter, a hint matching no row group (id > 100000) returns zero rows, proving the hint drives row-group elimination (all rows would return if the wiring regressed). - PartitionStatisticsScanTestBase.testCaseInsensitiveFilterWithProjectionOnColumnOutsideProjection: a case-insensitive filter on an uppercase stats column not in the projection exercises the residual Evaluator together with the projection/filter read-schema union. Co-authored-by: Isaac
Fixes spotlessJavaCheck violations flagged by build-checks CI: collapse the multi-line ternary in Parquet.ReadBuilder.withFilterHint and rewrap the InternalData.read javadoc. Co-authored-by: Isaac
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
InternalData.ReadBuilder had no way to pass a filter expression to the underlying format reader. For Parquet, this meant row-group skipping (already implemented via ParquetMetricsRowGroupFilter) was never reachable from internal metadata reads such as partition statistics scans.
This PR adds InternalData.read(format, file, filterHint) as the primary entry point for filtered reads.
The hint is a best-effort I/O optimization — Parquet uses it for row-group skipping; Avro ignores it. Callers are responsible for correctness via a residual filter.