[SPARK-56231][SQL] Bucket pruning and bucket join optimization for V2 file read path#55229
Draft
LuciferYang wants to merge 9 commits intoapache:masterfrom
Draft
[SPARK-56231][SQL] Bucket pruning and bucket join optimization for V2 file read path#55229LuciferYang wants to merge 9 commits intoapache:masterfrom
LuciferYang wants to merge 9 commits intoapache:masterfrom
Conversation
…Frame API writes and delete FallBackFileSourceV2 Key changes: - FileWrite: added partitionSchema, customPartitionLocations, dynamicPartitionOverwrite, isTruncate; path creation and truncate logic; dynamic partition overwrite via FileCommitProtocol - FileTable: createFileWriteBuilder with SupportsDynamicOverwrite and SupportsTruncate; capabilities now include TRUNCATE and OVERWRITE_DYNAMIC; fileIndex skips file existence checks when userSpecifiedSchema is provided (write path) - All file format writes (Parquet, ORC, CSV, JSON, Text, Avro) use createFileWriteBuilder with partition/truncate/overwrite support - DataFrameWriter.lookupV2Provider: enabled FileDataSourceV2 for non-partitioned Append and Overwrite via df.write.save(path) - DataFrameWriter.insertInto: V1 fallback for file sources (TODO: SPARK-56175) - DataFrameWriter.saveAsTable: V1 fallback for file sources (TODO: SPARK-56230, needs StagingTableCatalog) - DataSourceV2Utils.getTableProvider: V1 fallback for file sources (TODO: SPARK-56175) - Removed FallBackFileSourceV2 rule - V2SessionCatalog.createTable: V1 FileFormat data type validation
…catalog table loading, and gate removal Key changes: - FileTable extends SupportsPartitionManagement with createPartition, dropPartition, listPartitionIdentifiers, partitionSchema - Partition operations sync to catalog metastore (best-effort) - V2SessionCatalog.loadTable returns FileTable instead of V1Table, sets catalogTable and useCatalogFileIndex on FileTable - V2SessionCatalog.getDataSourceOptions includes storage.properties for proper option propagation (header, ORC bloom filter, etc.) - V2SessionCatalog.createTable validates data types via FileTable - FileTable.columns() restores NOT NULL constraints from catalogTable - FileTable.partitioning() falls back to userSpecifiedPartitioning or catalog partition columns - FileTable.fileIndex uses CatalogFileIndex when catalog has registered partitions (custom partition locations) - FileTable.schema checks column name duplication for non-catalog tables only - DataSourceV2Utils.getTableProvider: removed FileDataSourceV2 gate - DataFrameWriter.insertInto: enabled V2 for file sources - DataFrameWriter.saveAsTable: V1 fallback (TODO: SPARK-56230) - ResolveSessionCatalog: V1 fallback for FileTable-backed commands (AnalyzeTable, AnalyzeColumn, TruncateTable, TruncatePartition, ShowPartitions, RecoverPartitions, AddPartitions, RenamePartitions, DropPartitions, SetTableLocation, CREATE TABLE validation, REPLACE TABLE blocking) - FindDataSourceTable: streaming V1 fallback for FileTable (TODO: SPARK-56233) - DataSource.planForWritingFileFormat: graceful V2 handling
Enable bucketed writes for V2 file tables via catalog BucketSpec. Key changes: - FileWrite: add bucketSpec field, use V1WritesUtils.getWriterBucketSpec() instead of hardcoded None - FileTable: createFileWriteBuilder passes catalogTable.bucketSpec to the write pipeline - FileDataSourceV2: getTable uses collect to skip BucketTransform (handled via catalogTable.bucketSpec instead) - FileWriterFactory: use DynamicPartitionDataConcurrentWriter for bucketed writes since V2's RequiresDistributionAndOrdering cannot express hash-based ordering - All 6 format Write/Table classes updated with BucketSpec parameter Note: bucket pruning and bucket join (read-path optimization) are not included in this patch (tracked under SPARK-56231).
Add RepairTableExec to sync filesystem partition directories with catalog metastore for V2 file tables. Key changes: - New RepairTableExec: scans filesystem partitions via FileTable.listPartitionIdentifiers(), compares with catalog, registers missing partitions and drops orphaned entries - DataSourceV2Strategy: route RepairTable and RecoverPartitions for FileTable to new V2 exec node
Implement SupportsOverwriteV2 for V2 file tables to support static partition overwrite (INSERT OVERWRITE TABLE t PARTITION(p=1) SELECT ...). Key changes: - FileTable: replace SupportsTruncate with SupportsOverwriteV2 on WriteBuilder, implement overwrite(predicates) - FileWrite: extend toBatch() to delete only the matching partition directory, ordered by partitionSchema - FileTable.CAPABILITIES: add OVERWRITE_BY_FILTER - All 6 format Write/Table classes: plumb overwritePredicates parameter This is a prerequisite for SPARK-56304 (ifPartitionNotExists).
Contributor
Author
|
This is the 9th PR for SPARK-56170. The commit |
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.
PR Description
What changes were proposed in this pull request?
Enables bucket pruning and bucket join optimization for V2 file tables (
BatchScanExecwithFileScan), matching the V1FileSourceScanExecbehavior.BucketSpecfromFileTablethroughFileScanBuildertoFileScanand all 6 concrete scan classes (Parquet, ORC, CSV, JSON, Text, Avro)FileScan.partitions— files are grouped by bucket ID extracted from filenames, with optional bucket pruning and coalescingHashPartitioningfromDataSourceV2ScanExecBase.outputPartitioningfor bucketed scans, enabling shuffle-free joinsDisableUnnecessaryBucketedScanto handleBatchScanExec— disables bucketed scan when no downstream operator benefits from itCoalesceBucketsInJointo handleBatchScanExec— coalesces bucket counts for joins between tables with different bucket numbersFileSourceStrategy.genBucketSet(widened toprivate[sql]) for bucket pruning filter analysisWhy are the changes needed?
V2 file tables (default after the gate removal in SPARK-56170) do not support bucket pruning or bucket join optimizations. Workloads that rely on bucketed tables for performance would regress when using V2 file tables.
Does this PR introduce any user-facing change?
No. This is a performance optimization that makes V2 file tables match V1 behavior for bucketed reads.
How was this patch tested?
New
V2BucketedReadSuitewith 6 tests covering bucket pruning (equality + IN filters), bucketed join shuffle avoidance, disable unnecessary bucketed scan, bucket coalescing, and config-based bucketing disable. ExistingBucketedReadSuite(31 tests),DisableUnnecessaryBucketedScanSuite, andCoalesceBucketsInJoinSuiteall pass (50 tests total).Was this patch authored or co-authored using generative AI tooling?
Generated-by: Claude Code