Hudi will be taking on promise for it bundles to stay compatible with Spark minor versions (for ex 2.4, 3.1, 3.2): meaning that single build of Hudi (for ex "hudi-spark3.2-bundle") will be compatible with ALL patch versions in that minor branch (in that case 3.2.1, 3.2.0, etc)
To achieve that we'll have to remove (and ban) "spark-avro" as a dependency, which on a few occasions was the root-cause of incompatibility b/w consecutive Spark patch versions (most recently 3.2.1 and 3.2.0, due to this PR).
Instead of bundling "spark-avro" as dependency, we will be copying over some of the classes Hudi depends on and maintain them along the Hudi code-base to make sure we're able to provide for the aforementioned guarantee. To workaround arising compatibility issues we will be applying local patches to guarantee compatibility of Hudi bundles w/in the Spark minor version branches.
Following Hudi modules to Spark minor branches is currently maintained:
"hudi-spark3" -> 3.2.x
"hudi-spark3.1.x" -> 3.1.x
"hudi-spark2" -> 2.4.x
Following classes hierarchies (borrowed from "spark-avro") are maintained w/in these Spark-specific modules to guarantee compatibility with respective minor version branches:
AvroSerializer
AvroDeserializer
AvroUtils
Each of these classes has been correspondingly copied from Spark 3.2.1 (for 3.2.x branch), 3.1.2 (for 3.1.x branch), 2.4.4 (for 2.4.x branch) into their respective modules.
SchemaConverters class in turn is shared across all those modules given its relative stability (there're only cosmetical changes from 2.4.4 to 3.2.1).
All of the aforementioned classes have their corresponding scope of visibility limited to corresponding packages (org.apache.spark.sql.avro, org.apache.spark.sql) to make sure broader code-base does not become dependent on them and instead relies on facades abstracting them.
Additionally, given that Hudi plans on supporting all the patch versions of Spark w/in aforementioned minor versions branches of Spark, additional build steps were added to validate that Hudi could be properly compiled against those versions. Testing, however, is performed against the most recent patch versions of Spark with the help of Azure CI.
Brief change log:
- Removing spark-avro bundling from Hudi by default
- Scaffolded Spark 3.2.x hierarchy
- Bootstrapped Spark 3.1.x Avro serializer/deserializer hierarchy
- Bootstrapped Spark 2.4.x Avro serializer/deserializer hierarchy
- Moved ExpressionCodeGen,ExpressionPayload into hudi-spark module
- Fixed AvroDeserializer to stay compatible w/ both Spark 3.2.1 and 3.2.0
- Modified bot.yml to build full matrix of support Spark versions
- Removed "spark-avro" dependency from all modules
- Fixed relocation of spark-avro classes in bundles to assist in running integ-tests.
* [HUDI-845] Added locking capability to allow multiple writers
1. Added LockProvider API for pluggable lock methodologies
2. Added Resolution Strategy API to allow for pluggable conflict resolution
3. Added TableService client API to schedule table services
4. Added Transaction Manager for wrapping actions within transactions
* Fix flaky MOR unit test
* Update Spark APIs to make it be compatible with both spark2 & spark3
* Refactor bulk insert v2 part to make Hudi be able to compile with Spark3
* Add spark3 profile to handle fasterxml & spark version
* Create hudi-spark-common module & refactor hudi-spark related modules
Co-authored-by: Wenning Ding <wenningd@amazon.com>
1. Added the --clean-input and --clean-output parameters to clean the input and output directories before starting the job
2. Added the --delete-old-input parameter to deleted older batches for data already ingested. This helps keep number of redundant files low.
3. Added the --input-parallelism parameter to restrict the parallelism when generating input data. This helps keeping the number of generated input files low.
4. Added an option start_offset to Dag Nodes. Without ability to specify start offsets, data is generated into existing partitions. With start offset, DAG can control on which partition, the data is to be written.
5. Fixed generation of records for correct number of partitions
- In the existing implementation, the partition is chosen as a random long. This does not guarantee exact number of requested partitions to be created.
6. Changed variable blacklistedFields to be a Set as that is faster than List for membership checks.
7. Fixed integer division for Math.ceil. If two integers are divided, the result is not double unless one of the integer is casted to double.
- This change breaks `hudi-client` into `hudi-client-common` and `hudi-spark-client` modules
- Simple usages of Spark using jsc.parallelize() has been redone using EngineContext#map, EngineContext#flatMap etc
- Code changes in the PR, break classes into `BaseXYZ` parent classes with no spark dependencies living in `hudi-client-common`
- Classes on `hudi-spark-client` are named `SparkXYZ` extending the parent classes with all the Spark dependencies
- To simplify/cleanup, HoodieIndex#fetchRecordLocation has been removed and its usages in tests replaced with alternatives
Co-authored-by: Vinoth Chandar <vinoth@apache.org>
For Delete API, "hoodie.delete.shuffle.parallelism" isn't used as opposed to "hoodie.upsert.shuffle.parallelism" is used for upsert, this creates the performance difference between delete by upsert API with "EmptyHoodieRecordPayload" and delete API for certain cases.
This patch makes the following fixes in this regard.
- Let deduplicateKeys method use "hoodie.delete.shuffle.parallelism"
- Repartition inputRDD as "hoodie.delete.shuffle.parallelism" in case "hoodie.combine.before.delete=false"
* [HUDI-960] Implementation of the HFile base and log file format.
1. Includes HFileWriter and HFileReader
2. Includes HFileInputFormat for both snapshot and realtime input format for Hive
3. Unit test for new code
4. IT for using HFile format and querying using Hive (Presto and SparkSQL are not supported)
Advantage:
HFile file format saves data as binary key-value pairs. This implementation chooses the following values:
1. Key = Hoodie Record Key (as bytes)
2. Value = Avro encoded GenericRecord (as bytes)
HFile allows efficient lookup of a record by key or range of keys. Hence, this base file format is well suited to applications like RFC-15, RFC-08 which will benefit from the ability to lookup records by key or search in a range of keys without having to read the entire data/log format.
Limitations:
HFile storage format has certain limitations when used as a general purpose data storage format.
1. Does not have a implemented reader for Presto and SparkSQL
2. Is not a columnar file format and hence may lead to lower compression levels and greater IO on query side due to lack of column pruning
Other changes:
- Remove databricks/avro from pom
- Fix HoodieClientTestUtils from not using scala imports/reflection based conversion etc
- Breaking up limitFileSize(), per parquet and hfile base files
- Added three new configs for HoodieHFileConfig - prefetchBlocksOnOpen, cacheDataInL1, dropBehindCacheCompaction
- Throw UnsupportedException in HFileReader.getRecordKeys()
- Updated HoodieCopyOnWriteTable to create the correct merge handle (HoodieSortedMergeHandle for HFile and HoodieMergeHandle otherwise)
* Fixing checkstyle
Co-authored-by: Vinoth Chandar <vinoth@apache.org>
- [HUDI-418] Bootstrap Index Implementation using HFile with unit-test
- [HUDI-421] FileSystem View Changes to support Bootstrap with unit-tests
- [HUDI-424] Implement Query Side Integration for querying tables containing bootstrap file slices
- [HUDI-423] Implement upsert functionality for handling updates to these bootstrap file slices
- [HUDI-421] Bootstrap Write Client with tests
- [HUDI-425] Added HoodieDeltaStreamer support
- [HUDI-899] Add a knob to change partition-path style while performing metadata bootstrap
- [HUDI-900] Metadata Bootstrap Key Generator needs to handle complex keys correctly
- [HUDI-424] Simplify Record reader implementation
- [HUDI-423] Implement upsert functionality for handling updates to these bootstrap file slices
- [HUDI-420] Hoodie Demo working with hive and sparkSQL. Also, Hoodie CLI working with bootstrap tables
Co-authored-by: Mehrotra <uditme@amazon.com>
Co-authored-by: Vinoth Chandar <vinoth@apache.org>
Co-authored-by: Balaji Varadarajan <varadarb@uber.com>
Notable changes:
1. HoodieFileWriter and HoodieFileReader abstractions for writer/reader side of a base file format
2. HoodieDataBlock abstraction for creation specific data blocks for base file formats. (e.g. Parquet has HoodieAvroDataBlock)
3. All hardocded references to Parquet / Parquet based classes have been abstracted to call methods which accept a base file format
4. HiveSyncTool accepts the base file format as a CLI parameter
5. HoodieDeltaStreamer accepts the base file format as a CLI parameter
6. HoodieSparkSqlWriter accepts the base file format as a parameter
