NOTE: This change is first part of the series to clean up Hudi's Spark DataSource related implementations, making sure there's minimal code duplication among them, implementations are consistent and performant
This PR is making sure that BaseFileOnlyViewRelation only reads projected columns as well as avoiding unnecessary serde from Row to InternalRow
Brief change log
- Introduced HoodieBaseRDD as a base for all custom RDD impls
- Extracted common fields/methods to HoodieBaseRelation
- Cleaned up and streamlined HoodieBaseFileViewOnlyRelation
- Fixed all of the Relations to avoid superfluous Row <> InternalRow conversions
Desc: Add a hive sync config(hoodie.datasource.hive_sync.sync_comment). This config defaults to false.
While syncing data source to hudi, add column comments to source avro schema, and the sync_comment is true, syncing column comments to the hive table.
Rework of #4761
This diff introduces following changes:
- Write stats are converted to metadata index records during the commit. Making them use the HoodieData type so that the record generation scales up with needs.
- Metadata index init support for bloom filter and column stats partitions.
- When building the BloomFilter from the index records, using the type param stored in the payload instead of hardcoded type.
- Delta writes can change column ranges and the column stats index need to be properly updated with new ranges to be consistent with the table dataset. This fix add column stats index update support for the delta writes.
Co-authored-by: Manoj Govindassamy <manoj.govindassamy@gmail.com>
- This change makes sure MT records are updated appropriately on HDFS: previously after Log File append operations MT records were updated w/ just the size of the deltas being appended to the original files, which have been found to be the cause of issues in case of Rollbacks that were instead updating MT with records bearing the full file-size.
- To make sure that we hedge against similar issues going f/w, this PR alleviates this discrepancy and streamlines the flow of MT table always ingesting records bearing full file-sizes.
* Bootstrapping initial support for Metadata Table in Spark Datasource
- Consolidated Avro/Row conversion utilities to center around Spark's AvroDeserializer ; removed duplication
- Bootstrapped HoodieBaseRelation
- Updated HoodieMergeOnReadRDD to be able to handle Metadata Table
- Modified MOR relations to be able to read different Base File formats (Parquet, HFile)
Fix dependency conflict
Fix repairs command
Implement putIfAbsent for DDB lock provider
Add upgrade step and validate while fetching configs
Validate checksum for latest table version only while fetching config
Move generateChecksum to BinaryUtil
Rebase and resolve conflict
Fix table version check
Currently, HadoopFsRelation will use the value of the real partition path as the value of the partition field. However, different from the normal table, Hudi will persist the partition value in the parquet file. And in some cases, it's different between the value of the real partition path and the value of the partition field.
So here we implement BaseFileOnlyViewRelation which lets Hudi manage its own relation.
- This adds a restore plan and serializes it to restore.requested meta file in timeline. This also means that we are introducing schedule and execution phases for restore which was not present before.
- This adds support in spark-datasource to just schedule table services inline so that users can leverage async execution w/o the need for lock service providers.
* [HUDI-1295] Metadata Index - Bloom filter and Column stats index to speed up index lookups
- Today, base files have bloom filter at their footers and index lookups
have to load the base file to perform any bloom lookups. Though we have
interval tree based file purging, we still end up in significant amount
of base file read for the bloom filter for the end index lookups for the
keys. This index lookup operation can be made more performant by having
all the bloom filters in a new metadata partition and doing pointed
lookups based on keys.
* [HUDI-1295] Metadata Index - Bloom filter and Column stats index to speed up index lookups
- Adding indexing support for clean, restore and rollback operations.
Each of these operations will now be converted to index records for
bloom filter and column stats additionally.
* [HUDI-1295] Metadata Index - Bloom filter and Column stats index to speed up index lookups
- Making hoodie key consistent for both column stats and bloom index by
including fileId instead of fileName, in both read and write paths.
- Performance optimization for looking up records in the metadata table.
- Avoiding multi column sorting needed for HoodieBloomMetaIndexBatchCheckFunction
* [HUDI-1295] Metadata Index - Bloom filter and Column stats index to speed up index lookups
- HoodieBloomMetaIndexBatchCheckFunction cleanup to remove unused classes
- Base file checking before reading the file footer for bloom or column stats
* [HUDI-1295] Metadata Index - Bloom filter and Column stats index to speed up index lookups
- Updating the bloom index and column stats index to have full file name
included in the key instead of just file id.
- Minor test fixes.
* [HUDI-1295] Metadata Index - Bloom filter and Column stats index to speed up index lookups
- Fixed flink commit method to handle metadata table all partition update records
- TestBloomIndex fixes
* [HUDI-1295] Metadata Index - Bloom filter and Column stats index to speed up index lookups
- SparkHoodieBloomIndexHelper code simplification for various config modes
- Signature change for getBloomFilters() and getColumnStats(). Callers can
just pass in interested partition and file names, the index key is then
constructed internally based on the passed in parameters.
- KeyLookupHandle and KeyLookupResults code refactoring
- Metadata schema changes - removed the reserved field
* [HUDI-1295] Metadata Index - Bloom filter and Column stats index to speed up index lookups
- Removing HoodieColumnStatsMetadata and using HoodieColumnRangeMetadata instead.
Fixed the users of the the removed class.
* [HUDI-1295] Metadata Index - Bloom filter and Column stats index to speed up index lookups
- Extending meta index test to cover deletes, compactions, clean
and restore table operations. Also, fixed the getBloomFilters()
and getColumnStats() to account for deleted entries.
* [HUDI-1295] Metadata Index - Bloom filter and Column stats index to speed up index lookups
- Addressing review comments - java doc for new classes, keys sorting for
lookup, index methods renaming.
* [HUDI-1295] Metadata Index - Bloom filter and Column stats index to speed up index lookups
- Consolidated the bloom filter checking for keys in to one
HoodieMetadataBloomIndexCheckFunction instead of a spearate batch
and lazy mode. Removed all the configs around it.
- Made the metadata table partition file group count configurable.
- Fixed the HoodieKeyLookupHandle to have auto closable file reader
when checking bloom filter and range keys.
- Config property renames. Test fixes.
* [HUDI-1295] Metadata Index - Bloom filter and Column stats index to speed up index lookups
- Enabling column stats indexing for all columns by default
- Handling column stat generation errors and test update
* [HUDI-1295] Metadata Index - Bloom filter and Column stats index to speed up index lookups
- Metadata table partition file group count taken from the slices when
the table is bootstrapped.
- Prep records for the commit refactored to the base class
- HoodieFileReader interface changes for filtering keys
- Multi column and data types support for colums stats index
* [HUDI-1295] Metadata Index - Bloom filter and Column stats index to speed up index lookups
- rebase to latest master and merge fixes for the build and test failures
* [HUDI-1295] Metadata Index - Bloom filter and Column stats index to speed up index lookups
- Extending the metadata column stats type payload schema to include
more statistics about the column ranges to help query integration.
* [HUDI-1295] Metadata Index - Bloom filter and Column stats index to speed up index lookups
- Addressing review comments
This change is addressing issues in regards to Metadata Table observing ingesting duplicated records leading to it persisting incorrect file-sizes for the files referred to in those records.
There are multiple issues that were leading to that:
- [HUDI-3322] Incorrect Rollback Plan generation: Rollback Plan generated for MOR tables was overly expansively listing all log-files with the latest base-instant as the ones that have been affected by the rollback, leading to invalid MT records being ingested referring to those.
- [HUDI-3343] Metadata Table including Uncommitted Log Files during Bootstrap: Since MT is bootstrapped at the end of the commit operation execution (after FS activity, but before committing to the timeline), it was actually incorrectly ingesting some files that were part of the intermediate state of the operation being committed.
This change will unblock Stack of PRs based off #4556
* [HUDI-2763] Metadata table records - support for key deduplication and virtual keys
- The backing log format for the metadata table is HFile, a KeyValue type.
Since the key field in the metadata record payload is a duplicate of the
Key in the Cell, the redundant key field in the record can be emptied
to save on the cost.
- HoodieHFileWriter and HoodieHFileDataBlock will now serialize records
with the key field emptied by default. HFile writer tries to find if
the record has metadata payload schema field 'key' and if so it does
the key trimming from the record payload.
- HoodieHFileReader when reading the serialized records back from disk,
it materializes the missing keyFields if any. HFile reader tries to
find if the record has metadata payload schema fiels 'key' and if so
it does the key materialization in the record payload.
- Tests have been added to verify the default virtual keys and key
deduplication support for the metadata table records.
Co-authored-by: Vinoth Chandar <vinoth@apache.org>
Refactoring layout optimization (clustering) flow to
- Enable support for linear (lexicographic) ordering as one of the ordering strategies (along w/ Z-order, Hilbert)
- Reconcile Layout Optimization and Clustering configuration to be more congruent
* Introduce hudi-spark3-common and hudi-spark2-common modules to place classes that would be reused in different spark versions, also introduce hudi-spark3.1.x to support spark 3.1.x.
* Introduce hudi format under hudi-spark2, hudi-spark3, hudi-spark3.1.x modules and change the hudi format in original hudi-spark module to hudi_v1 format.
* Manually tested on Spark 3.1.2 and Spark 3.2.0 SQL.
* Added a README.md file under hudi-spark-datasource module.
