[MINOR] Make sure all HoodieRecords are appropriately serializable by Kryo
#6977
Conversation
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
added 11 commits
October 17, 2022 19:57
… `KryoSerializable`
…ld not be overridden, b) provide for hooks to properly serialize record's payload; Implemented serialization hooks for `HoodieAvroIndexedRecord`; Implemented serialization hooks for `HoodieEmptyRecord`;
…ld not be overridden, b) provide for hooks to properly serialize record's payload; Implemented serialization hooks for `HoodieAvroIndexedRecord`; Implemented serialization hooks for `HoodieEmptyRecord`; Implemented serialization hooks for `HoodieAvroRecord`;
… that it could project row
alexeykudinkin
force-pushed
the
ak/rfc46-kryo-serde-fix
branch
from
30a32fe to
416dda2
Compare
danny0405
reviewed
Oct 18, 2022
|
|
||
| kryo.writeObjectOrNull(output, key, HoodieKey.class); | ||
| kryo.writeObjectOrNull(output, currentLocation, recLocSerializer); | ||
| kryo.writeObjectOrNull(output, newLocation, recLocSerializer); |
Contributor
There was a problem hiding this comment.
How about the performance gains for records SE/DE after this change ?
Contributor
Author
There was a problem hiding this comment.
It's less about performance gains in the HoodieRecord per se and more about fixing ser/de for HoodieSparkRecord
- Previous PR actually removed mandatory conversion to
UnsafeRoww/in theHoodieSparkRecord(for performance reasons) - Conversion to
UnsafeRowis still required though upon ser/de (for shuffling). This change is to handle that.
wzx140
reviewed
Oct 18, 2022
| * (by Kryo) | ||
| */ | ||
| public HoodieSparkRecord(InternalRow data) { | ||
| private final transient StructType schema; |
Contributor
There was a problem hiding this comment.
I think this is unnecessary. The schema is only used in construction.
Contributor
Author
There was a problem hiding this comment.
We actually now don't do projection in ctor, instead we only do it when serializing HoodieSparkRecord
| @Override | ||
| protected final void writeRecordPayload(InternalRow payload, Kryo kryo, Output output) { | ||
| // NOTE: [[payload]] could be null if record has already been deflated | ||
| UnsafeRow unsafeRow = convertToUnsafeRow(payload, schema); |
Contributor
There was a problem hiding this comment.
HoodieInternalRow must hold UnsafeRow. Does it need to unsafe projection? We use HoodieInternalRow because we want to modify metaField.
I don't really understand why we should make sure that SparkRecord only hold UnsafeRow. HoodieInternalRow hold UnsafeRow too. They basically make no difference.
Contributor
Author
There was a problem hiding this comment.
This is done to simplify ser/deserialization sequnce:
- We always project to
UnsafeRowbefore serialization, therefore - We need to decode just the
UnsafeRowwhen we deserialize - Ser/de of
UnsafeRowis as simple as just writing out the bytes
| * - partitionPath : the partition path of a record. | ||
| */ | ||
| public class HoodieKey implements Serializable { | ||
| public class HoodieKey implements Serializable, KryoSerializable { |
Contributor
There was a problem hiding this comment.
We have used kryo as the serializer of shuffle. I wonder that Why we explicitly specify KryoSerializable? Same to HoodieRecord.
Contributor
Author
There was a problem hiding this comment.
Correct. We want to enforce that
HoodieRecordare serializable by Kryo- That we control serialization sequence
Collaborator
Contributor
Author
|
CI is green:
|
wzx140
added a commit
to wzx140/hudi
that referenced
this pull request
Nov 30, 2022
[minor] add more test for rfc46 (apache#7003) ## Change Logs - Add HoodieSparkValidateDuplicateKeyRecordMerger behaving the same as ValidateDuplicateKeyPayload. We should use it with config "hoodie.sql.insert.mode=strict". - Fix nest field exist in HoodieCatalystExpressionUtils - Fix rewrite in HoodieInternalRowUtiles to support type promoted as avro - Fallback to avro when use "merge into" sql - Fix some schema handling issue - Support delta streamer - Convert parquet schema to spark schema and then avro schema(in org.apache.hudi.io.storage.HoodieSparkParquetReader#getSchema). Some types in avro are not compatible with parquet. For ex, decimal as int32/int64 in parquet will convert to int/long in avro. Because avro do not has decimal as int/long . We will lose the logic type info if we directly convert it to avro schema. - Support schema evolution in parquet block [Minor] fix multi deser avro payload (apache#7021) In HoodieAvroRecord, we will call isDelete, shouldIgnore before we write it to the file. Each method will deserialize HoodiePayload. So we add deserialization method in HoodieRecord and call this method once before calling isDelete or shouldIgnore. Co-authored-by: wangzixuan.wzxuan <[email protected]> Co-authored-by: Alexey Kudinkin <[email protected]> Co-authored-by: Alexey Kudinkin <[email protected]> [MINOR] Properly registering target classes w/ Kryo (apache#7026) * Added `HoodieKryoRegistrar` registering necessary Hudi's classes w/ Kryo to make their serialization more efficient (by serializing just the class id, in-liue the fully qualified class-name) * Redirected Kryo registration to `HoodieKryoRegistrar` * Registered additional classes likely to be serialized by Kryo * Updated tests * Fixed serialization of Avro's `Utf8` to serialize just the bytes * Added tests * Added custom `AvroUtf8Serializer`; Tidying up * Extracted `HoodieCommonKryoRegistrar` to leverage in `SerializationUtils` * `HoodieKryoRegistrar` > `HoodieSparkKryoRegistrar`; Rebased `HoodieSparkKryoRegistrar` onto `HoodieCommonKryoRegistrar` * `lint` * Fixing compilation for Spark 2.x * Disabling flaky test [MINOR] Make sure all `HoodieRecord`s are appropriately serializable by Kryo (apache#6977) * Make sure `HoodieRecord`, `HoodieKey`, `HoodieRecordLocation` are all `KryoSerializable` * Revisited `HoodieRecord` serialization hooks to make sure they a) could not be overridden, b) provide for hooks to properly serialize record's payload; Implemented serialization hooks for `HoodieAvroIndexedRecord`; Implemented serialization hooks for `HoodieEmptyRecord`; * Revisited `HoodieRecord` serialization hooks to make sure they a) could not be overridden, b) provide for hooks to properly serialize record's payload; Implemented serialization hooks for `HoodieAvroIndexedRecord`; Implemented serialization hooks for `HoodieEmptyRecord`; Implemented serialization hooks for `HoodieAvroRecord`; * Revisited `HoodieSparkRecord` to transiently hold on to the schema so that it could project row * Implemented serialization hooks for `HoodieSparkRecord` * Added `TestHoodieSparkRecord` * Added tests for Avro-based records * Added test for `HoodieEmptyRecord` * Fixed sealing/unsealing for `HoodieRecord` in `HoodieBackedTableMetadataWriter` * Properly handle deflated records * Fixing `Row`s encoding * Fixed `HoodieRecord` to be properly sealed/unsealed * Fixed serialization of the `HoodieRecordGlobalLocation` [MINOR] Additional fixes for apache#6745 (apache#6947) * Tidying up * Tidying up more * Cleaning up duplication * Tidying up * Revisited legacy operating mode configuration * Tidying up * Cleaned up `projectUnsafe` API * Fixing compilation * Cleaning up `HoodieSparkRecord` ctors; Revisited mandatory unsafe-projection * Fixing compilation * Cleaned up `ParquetReader` initialization * Revisited `HoodieSparkRecord` to accept either `UnsafeRow` or `HoodieInternalRow`, and avoid unnecessary copying after unsafe-projection * Cleaning up redundant exception spec * Make sure `updateMetadataFields` properly wraps `InternalRow` into `HoodieInternalRow` if necessary; Cleaned up `MetadataValues` * Fixed meta-fields extraction and `HoodieInternalRow` composition w/in `HoodieSparkRecord` * De-duplicate `HoodieSparkRecord` ctors; Make sure either only `UnsafeRow` or `HoodieInternalRow` are permitted inside `HoodieSparkRecord` * Removed unnecessary copying * Cleaned up projection for `HoodieSparkRecord` (dropping partition columns); Removed unnecessary copying * Fixing compilation * Fixing compilation (for Flink) * Cleaned up File Raders' interfaces: - Extracted `HoodieSeekingFileReader` interface (for key-ranged reads) - Pushed down concrete implementation methods into `HoodieAvroFileReaderBase` from the interfaces * Cleaned up File Readers impls (inline with then new interfaces) * Rebsaed `HoodieBackedTableMetadata` onto new `HoodieSeekingFileReader` * Tidying up * Missing licenses * Re-instate custom override for `HoodieAvroParquetReader`; Tidying up * Fixed missing cloning w/in `HoodieLazyInsertIterable` * Fixed missing cloning in deduplication flow * Allow `HoodieSparkRecord` to hold `ColumnarBatchRow` * Missing licenses * Fixing compilation * Missing changes * Fixed Spark 2.x validation whether the row was read as a batch Fix comment in RFC46 (apache#6745) * rename * add MetadataValues in updateMetadataValues * remove singleton in fileFactory * add truncateRecordKey * remove hoodieRecord#setData * rename HoodieAvroRecord * fix code style * fix HoodieSparkRecordSerializer * fix benchmark * fix SparkRecordUtils * instantiate HoodieWriteConfig on the fly * add test * fix HoodieSparkRecordSerializer. Replace Java's object serialization with kryo * add broadcast * fix comment * remove unnecessary broadcast * add unsafe check in spark record * fix getRecordColumnValues * remove spark.sql.parquet.writeLegacyFormat * fix unsafe projection * fix * pass external schema * update doc * rename back to HoodieAvroRecord * fix * remove comparable wrapper * fix comment * fix comment * fix comment * fix comment * simplify row copy * fix ParquetReaderIterator Co-authored-by: Shawy Geng <[email protected]> Co-authored-by: wangzixuan.wzxuan <[email protected]> [RFC-46][HUDI-4414] Update the RFC-46 doc to fix comments feedback (apache#6132) * Update the RFC-46 doc to fix comments feedback * fix Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4301] [HUDI-3384][HUDI-3385] Spark specific file reader/writer.(apache#5629) * [HUDI-4301] [HUDI-3384][HUDI-3385] Spark specific file reader/writer. * add schema finger print * add benchmark * a new way to config the merger * fix Co-authored-by: wangzixuan.wzxuan <[email protected]> Co-authored-by: gengxiaoyu <[email protected]> [HUDI-3350][HUDI-3351] Support HoodieMerge API and Spark engine-specific HoodieRecord (apache#5627) Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4344] fix usage of HoodieDataBlock#getRecordIterator (apache#6005) Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4292][RFC-46] Update doc to align with the Record Merge API changes (apache#5927) [MINOR] Fix type casting in TestHoodieHFileReaderWriter [HUDI-3378][HUDI-3379][HUDI-3381] Migrate usage of HoodieRecordPayload and raw Avro payload to HoodieRecord (apache#5522) Co-authored-by: Alexey Kudinkin <[email protected]> Co-authored-by: wangzixuan.wzxuan <[email protected]>
wzx140
added a commit
to wzx140/hudi
that referenced
this pull request
Dec 1, 2022
[minor] add more test for rfc46 (apache#7003) ## Change Logs - Add HoodieSparkValidateDuplicateKeyRecordMerger behaving the same as ValidateDuplicateKeyPayload. We should use it with config "hoodie.sql.insert.mode=strict". - Fix nest field exist in HoodieCatalystExpressionUtils - Fix rewrite in HoodieInternalRowUtiles to support type promoted as avro - Fallback to avro when use "merge into" sql - Fix some schema handling issue - Support delta streamer - Convert parquet schema to spark schema and then avro schema(in org.apache.hudi.io.storage.HoodieSparkParquetReader#getSchema). Some types in avro are not compatible with parquet. For ex, decimal as int32/int64 in parquet will convert to int/long in avro. Because avro do not has decimal as int/long . We will lose the logic type info if we directly convert it to avro schema. - Support schema evolution in parquet block [Minor] fix multi deser avro payload (apache#7021) In HoodieAvroRecord, we will call isDelete, shouldIgnore before we write it to the file. Each method will deserialize HoodiePayload. So we add deserialization method in HoodieRecord and call this method once before calling isDelete or shouldIgnore. Co-authored-by: wangzixuan.wzxuan <[email protected]> Co-authored-by: Alexey Kudinkin <[email protected]> Co-authored-by: Alexey Kudinkin <[email protected]> [MINOR] Properly registering target classes w/ Kryo (apache#7026) * Added `HoodieKryoRegistrar` registering necessary Hudi's classes w/ Kryo to make their serialization more efficient (by serializing just the class id, in-liue the fully qualified class-name) * Redirected Kryo registration to `HoodieKryoRegistrar` * Registered additional classes likely to be serialized by Kryo * Updated tests * Fixed serialization of Avro's `Utf8` to serialize just the bytes * Added tests * Added custom `AvroUtf8Serializer`; Tidying up * Extracted `HoodieCommonKryoRegistrar` to leverage in `SerializationUtils` * `HoodieKryoRegistrar` > `HoodieSparkKryoRegistrar`; Rebased `HoodieSparkKryoRegistrar` onto `HoodieCommonKryoRegistrar` * `lint` * Fixing compilation for Spark 2.x * Disabling flaky test [MINOR] Make sure all `HoodieRecord`s are appropriately serializable by Kryo (apache#6977) * Make sure `HoodieRecord`, `HoodieKey`, `HoodieRecordLocation` are all `KryoSerializable` * Revisited `HoodieRecord` serialization hooks to make sure they a) could not be overridden, b) provide for hooks to properly serialize record's payload; Implemented serialization hooks for `HoodieAvroIndexedRecord`; Implemented serialization hooks for `HoodieEmptyRecord`; * Revisited `HoodieRecord` serialization hooks to make sure they a) could not be overridden, b) provide for hooks to properly serialize record's payload; Implemented serialization hooks for `HoodieAvroIndexedRecord`; Implemented serialization hooks for `HoodieEmptyRecord`; Implemented serialization hooks for `HoodieAvroRecord`; * Revisited `HoodieSparkRecord` to transiently hold on to the schema so that it could project row * Implemented serialization hooks for `HoodieSparkRecord` * Added `TestHoodieSparkRecord` * Added tests for Avro-based records * Added test for `HoodieEmptyRecord` * Fixed sealing/unsealing for `HoodieRecord` in `HoodieBackedTableMetadataWriter` * Properly handle deflated records * Fixing `Row`s encoding * Fixed `HoodieRecord` to be properly sealed/unsealed * Fixed serialization of the `HoodieRecordGlobalLocation` [MINOR] Additional fixes for apache#6745 (apache#6947) * Tidying up * Tidying up more * Cleaning up duplication * Tidying up * Revisited legacy operating mode configuration * Tidying up * Cleaned up `projectUnsafe` API * Fixing compilation * Cleaning up `HoodieSparkRecord` ctors; Revisited mandatory unsafe-projection * Fixing compilation * Cleaned up `ParquetReader` initialization * Revisited `HoodieSparkRecord` to accept either `UnsafeRow` or `HoodieInternalRow`, and avoid unnecessary copying after unsafe-projection * Cleaning up redundant exception spec * Make sure `updateMetadataFields` properly wraps `InternalRow` into `HoodieInternalRow` if necessary; Cleaned up `MetadataValues` * Fixed meta-fields extraction and `HoodieInternalRow` composition w/in `HoodieSparkRecord` * De-duplicate `HoodieSparkRecord` ctors; Make sure either only `UnsafeRow` or `HoodieInternalRow` are permitted inside `HoodieSparkRecord` * Removed unnecessary copying * Cleaned up projection for `HoodieSparkRecord` (dropping partition columns); Removed unnecessary copying * Fixing compilation * Fixing compilation (for Flink) * Cleaned up File Raders' interfaces: - Extracted `HoodieSeekingFileReader` interface (for key-ranged reads) - Pushed down concrete implementation methods into `HoodieAvroFileReaderBase` from the interfaces * Cleaned up File Readers impls (inline with then new interfaces) * Rebsaed `HoodieBackedTableMetadata` onto new `HoodieSeekingFileReader` * Tidying up * Missing licenses * Re-instate custom override for `HoodieAvroParquetReader`; Tidying up * Fixed missing cloning w/in `HoodieLazyInsertIterable` * Fixed missing cloning in deduplication flow * Allow `HoodieSparkRecord` to hold `ColumnarBatchRow` * Missing licenses * Fixing compilation * Missing changes * Fixed Spark 2.x validation whether the row was read as a batch Fix comment in RFC46 (apache#6745) * rename * add MetadataValues in updateMetadataValues * remove singleton in fileFactory * add truncateRecordKey * remove hoodieRecord#setData * rename HoodieAvroRecord * fix code style * fix HoodieSparkRecordSerializer * fix benchmark * fix SparkRecordUtils * instantiate HoodieWriteConfig on the fly * add test * fix HoodieSparkRecordSerializer. Replace Java's object serialization with kryo * add broadcast * fix comment * remove unnecessary broadcast * add unsafe check in spark record * fix getRecordColumnValues * remove spark.sql.parquet.writeLegacyFormat * fix unsafe projection * fix * pass external schema * update doc * rename back to HoodieAvroRecord * fix * remove comparable wrapper * fix comment * fix comment * fix comment * fix comment * simplify row copy * fix ParquetReaderIterator Co-authored-by: Shawy Geng <[email protected]> Co-authored-by: wangzixuan.wzxuan <[email protected]> [RFC-46][HUDI-4414] Update the RFC-46 doc to fix comments feedback (apache#6132) * Update the RFC-46 doc to fix comments feedback * fix Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4301] [HUDI-3384][HUDI-3385] Spark specific file reader/writer.(apache#5629) * [HUDI-4301] [HUDI-3384][HUDI-3385] Spark specific file reader/writer. * add schema finger print * add benchmark * a new way to config the merger * fix Co-authored-by: wangzixuan.wzxuan <[email protected]> Co-authored-by: gengxiaoyu <[email protected]> [HUDI-3350][HUDI-3351] Support HoodieMerge API and Spark engine-specific HoodieRecord (apache#5627) Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4344] fix usage of HoodieDataBlock#getRecordIterator (apache#6005) Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4292][RFC-46] Update doc to align with the Record Merge API changes (apache#5927) [MINOR] Fix type casting in TestHoodieHFileReaderWriter [HUDI-3378][HUDI-3379][HUDI-3381] Migrate usage of HoodieRecordPayload and raw Avro payload to HoodieRecord (apache#5522) Co-authored-by: Alexey Kudinkin <[email protected]> Co-authored-by: wangzixuan.wzxuan <[email protected]>
wzx140
added a commit
to wzx140/hudi
that referenced
this pull request
Dec 2, 2022
[minor] add more test for rfc46 (apache#7003) ## Change Logs - Add HoodieSparkValidateDuplicateKeyRecordMerger behaving the same as ValidateDuplicateKeyPayload. We should use it with config "hoodie.sql.insert.mode=strict". - Fix nest field exist in HoodieCatalystExpressionUtils - Fix rewrite in HoodieInternalRowUtiles to support type promoted as avro - Fallback to avro when use "merge into" sql - Fix some schema handling issue - Support delta streamer - Convert parquet schema to spark schema and then avro schema(in org.apache.hudi.io.storage.HoodieSparkParquetReader#getSchema). Some types in avro are not compatible with parquet. For ex, decimal as int32/int64 in parquet will convert to int/long in avro. Because avro do not has decimal as int/long . We will lose the logic type info if we directly convert it to avro schema. - Support schema evolution in parquet block [Minor] fix multi deser avro payload (apache#7021) In HoodieAvroRecord, we will call isDelete, shouldIgnore before we write it to the file. Each method will deserialize HoodiePayload. So we add deserialization method in HoodieRecord and call this method once before calling isDelete or shouldIgnore. Co-authored-by: wangzixuan.wzxuan <[email protected]> Co-authored-by: Alexey Kudinkin <[email protected]> Co-authored-by: Alexey Kudinkin <[email protected]> [MINOR] Properly registering target classes w/ Kryo (apache#7026) * Added `HoodieKryoRegistrar` registering necessary Hudi's classes w/ Kryo to make their serialization more efficient (by serializing just the class id, in-liue the fully qualified class-name) * Redirected Kryo registration to `HoodieKryoRegistrar` * Registered additional classes likely to be serialized by Kryo * Updated tests * Fixed serialization of Avro's `Utf8` to serialize just the bytes * Added tests * Added custom `AvroUtf8Serializer`; Tidying up * Extracted `HoodieCommonKryoRegistrar` to leverage in `SerializationUtils` * `HoodieKryoRegistrar` > `HoodieSparkKryoRegistrar`; Rebased `HoodieSparkKryoRegistrar` onto `HoodieCommonKryoRegistrar` * `lint` * Fixing compilation for Spark 2.x * Disabling flaky test [MINOR] Make sure all `HoodieRecord`s are appropriately serializable by Kryo (apache#6977) * Make sure `HoodieRecord`, `HoodieKey`, `HoodieRecordLocation` are all `KryoSerializable` * Revisited `HoodieRecord` serialization hooks to make sure they a) could not be overridden, b) provide for hooks to properly serialize record's payload; Implemented serialization hooks for `HoodieAvroIndexedRecord`; Implemented serialization hooks for `HoodieEmptyRecord`; * Revisited `HoodieRecord` serialization hooks to make sure they a) could not be overridden, b) provide for hooks to properly serialize record's payload; Implemented serialization hooks for `HoodieAvroIndexedRecord`; Implemented serialization hooks for `HoodieEmptyRecord`; Implemented serialization hooks for `HoodieAvroRecord`; * Revisited `HoodieSparkRecord` to transiently hold on to the schema so that it could project row * Implemented serialization hooks for `HoodieSparkRecord` * Added `TestHoodieSparkRecord` * Added tests for Avro-based records * Added test for `HoodieEmptyRecord` * Fixed sealing/unsealing for `HoodieRecord` in `HoodieBackedTableMetadataWriter` * Properly handle deflated records * Fixing `Row`s encoding * Fixed `HoodieRecord` to be properly sealed/unsealed * Fixed serialization of the `HoodieRecordGlobalLocation` [MINOR] Additional fixes for apache#6745 (apache#6947) * Tidying up * Tidying up more * Cleaning up duplication * Tidying up * Revisited legacy operating mode configuration * Tidying up * Cleaned up `projectUnsafe` API * Fixing compilation * Cleaning up `HoodieSparkRecord` ctors; Revisited mandatory unsafe-projection * Fixing compilation * Cleaned up `ParquetReader` initialization * Revisited `HoodieSparkRecord` to accept either `UnsafeRow` or `HoodieInternalRow`, and avoid unnecessary copying after unsafe-projection * Cleaning up redundant exception spec * Make sure `updateMetadataFields` properly wraps `InternalRow` into `HoodieInternalRow` if necessary; Cleaned up `MetadataValues` * Fixed meta-fields extraction and `HoodieInternalRow` composition w/in `HoodieSparkRecord` * De-duplicate `HoodieSparkRecord` ctors; Make sure either only `UnsafeRow` or `HoodieInternalRow` are permitted inside `HoodieSparkRecord` * Removed unnecessary copying * Cleaned up projection for `HoodieSparkRecord` (dropping partition columns); Removed unnecessary copying * Fixing compilation * Fixing compilation (for Flink) * Cleaned up File Raders' interfaces: - Extracted `HoodieSeekingFileReader` interface (for key-ranged reads) - Pushed down concrete implementation methods into `HoodieAvroFileReaderBase` from the interfaces * Cleaned up File Readers impls (inline with then new interfaces) * Rebsaed `HoodieBackedTableMetadata` onto new `HoodieSeekingFileReader` * Tidying up * Missing licenses * Re-instate custom override for `HoodieAvroParquetReader`; Tidying up * Fixed missing cloning w/in `HoodieLazyInsertIterable` * Fixed missing cloning in deduplication flow * Allow `HoodieSparkRecord` to hold `ColumnarBatchRow` * Missing licenses * Fixing compilation * Missing changes * Fixed Spark 2.x validation whether the row was read as a batch Fix comment in RFC46 (apache#6745) * rename * add MetadataValues in updateMetadataValues * remove singleton in fileFactory * add truncateRecordKey * remove hoodieRecord#setData * rename HoodieAvroRecord * fix code style * fix HoodieSparkRecordSerializer * fix benchmark * fix SparkRecordUtils * instantiate HoodieWriteConfig on the fly * add test * fix HoodieSparkRecordSerializer. Replace Java's object serialization with kryo * add broadcast * fix comment * remove unnecessary broadcast * add unsafe check in spark record * fix getRecordColumnValues * remove spark.sql.parquet.writeLegacyFormat * fix unsafe projection * fix * pass external schema * update doc * rename back to HoodieAvroRecord * fix * remove comparable wrapper * fix comment * fix comment * fix comment * fix comment * simplify row copy * fix ParquetReaderIterator Co-authored-by: Shawy Geng <[email protected]> Co-authored-by: wangzixuan.wzxuan <[email protected]> [RFC-46][HUDI-4414] Update the RFC-46 doc to fix comments feedback (apache#6132) * Update the RFC-46 doc to fix comments feedback * fix Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4301] [HUDI-3384][HUDI-3385] Spark specific file reader/writer.(apache#5629) * [HUDI-4301] [HUDI-3384][HUDI-3385] Spark specific file reader/writer. * add schema finger print * add benchmark * a new way to config the merger * fix Co-authored-by: wangzixuan.wzxuan <[email protected]> Co-authored-by: gengxiaoyu <[email protected]> [HUDI-3350][HUDI-3351] Support HoodieMerge API and Spark engine-specific HoodieRecord (apache#5627) Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4344] fix usage of HoodieDataBlock#getRecordIterator (apache#6005) Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4292][RFC-46] Update doc to align with the Record Merge API changes (apache#5927) [MINOR] Fix type casting in TestHoodieHFileReaderWriter [HUDI-3378][HUDI-3379][HUDI-3381] Migrate usage of HoodieRecordPayload and raw Avro payload to HoodieRecord (apache#5522) Co-authored-by: Alexey Kudinkin <[email protected]> Co-authored-by: wangzixuan.wzxuan <[email protected]>
wzx140
added a commit
to wzx140/hudi
that referenced
this pull request
Dec 3, 2022
[minor] add more test for rfc46 (apache#7003) ## Change Logs - Add HoodieSparkValidateDuplicateKeyRecordMerger behaving the same as ValidateDuplicateKeyPayload. We should use it with config "hoodie.sql.insert.mode=strict". - Fix nest field exist in HoodieCatalystExpressionUtils - Fix rewrite in HoodieInternalRowUtiles to support type promoted as avro - Fallback to avro when use "merge into" sql - Fix some schema handling issue - Support delta streamer - Convert parquet schema to spark schema and then avro schema(in org.apache.hudi.io.storage.HoodieSparkParquetReader#getSchema). Some types in avro are not compatible with parquet. For ex, decimal as int32/int64 in parquet will convert to int/long in avro. Because avro do not has decimal as int/long . We will lose the logic type info if we directly convert it to avro schema. - Support schema evolution in parquet block [Minor] fix multi deser avro payload (apache#7021) In HoodieAvroRecord, we will call isDelete, shouldIgnore before we write it to the file. Each method will deserialize HoodiePayload. So we add deserialization method in HoodieRecord and call this method once before calling isDelete or shouldIgnore. Co-authored-by: wangzixuan.wzxuan <[email protected]> Co-authored-by: Alexey Kudinkin <[email protected]> Co-authored-by: Alexey Kudinkin <[email protected]> [MINOR] Properly registering target classes w/ Kryo (apache#7026) * Added `HoodieKryoRegistrar` registering necessary Hudi's classes w/ Kryo to make their serialization more efficient (by serializing just the class id, in-liue the fully qualified class-name) * Redirected Kryo registration to `HoodieKryoRegistrar` * Registered additional classes likely to be serialized by Kryo * Updated tests * Fixed serialization of Avro's `Utf8` to serialize just the bytes * Added tests * Added custom `AvroUtf8Serializer`; Tidying up * Extracted `HoodieCommonKryoRegistrar` to leverage in `SerializationUtils` * `HoodieKryoRegistrar` > `HoodieSparkKryoRegistrar`; Rebased `HoodieSparkKryoRegistrar` onto `HoodieCommonKryoRegistrar` * `lint` * Fixing compilation for Spark 2.x * Disabling flaky test [MINOR] Make sure all `HoodieRecord`s are appropriately serializable by Kryo (apache#6977) * Make sure `HoodieRecord`, `HoodieKey`, `HoodieRecordLocation` are all `KryoSerializable` * Revisited `HoodieRecord` serialization hooks to make sure they a) could not be overridden, b) provide for hooks to properly serialize record's payload; Implemented serialization hooks for `HoodieAvroIndexedRecord`; Implemented serialization hooks for `HoodieEmptyRecord`; * Revisited `HoodieRecord` serialization hooks to make sure they a) could not be overridden, b) provide for hooks to properly serialize record's payload; Implemented serialization hooks for `HoodieAvroIndexedRecord`; Implemented serialization hooks for `HoodieEmptyRecord`; Implemented serialization hooks for `HoodieAvroRecord`; * Revisited `HoodieSparkRecord` to transiently hold on to the schema so that it could project row * Implemented serialization hooks for `HoodieSparkRecord` * Added `TestHoodieSparkRecord` * Added tests for Avro-based records * Added test for `HoodieEmptyRecord` * Fixed sealing/unsealing for `HoodieRecord` in `HoodieBackedTableMetadataWriter` * Properly handle deflated records * Fixing `Row`s encoding * Fixed `HoodieRecord` to be properly sealed/unsealed * Fixed serialization of the `HoodieRecordGlobalLocation` [MINOR] Additional fixes for apache#6745 (apache#6947) * Tidying up * Tidying up more * Cleaning up duplication * Tidying up * Revisited legacy operating mode configuration * Tidying up * Cleaned up `projectUnsafe` API * Fixing compilation * Cleaning up `HoodieSparkRecord` ctors; Revisited mandatory unsafe-projection * Fixing compilation * Cleaned up `ParquetReader` initialization * Revisited `HoodieSparkRecord` to accept either `UnsafeRow` or `HoodieInternalRow`, and avoid unnecessary copying after unsafe-projection * Cleaning up redundant exception spec * Make sure `updateMetadataFields` properly wraps `InternalRow` into `HoodieInternalRow` if necessary; Cleaned up `MetadataValues` * Fixed meta-fields extraction and `HoodieInternalRow` composition w/in `HoodieSparkRecord` * De-duplicate `HoodieSparkRecord` ctors; Make sure either only `UnsafeRow` or `HoodieInternalRow` are permitted inside `HoodieSparkRecord` * Removed unnecessary copying * Cleaned up projection for `HoodieSparkRecord` (dropping partition columns); Removed unnecessary copying * Fixing compilation * Fixing compilation (for Flink) * Cleaned up File Raders' interfaces: - Extracted `HoodieSeekingFileReader` interface (for key-ranged reads) - Pushed down concrete implementation methods into `HoodieAvroFileReaderBase` from the interfaces * Cleaned up File Readers impls (inline with then new interfaces) * Rebsaed `HoodieBackedTableMetadata` onto new `HoodieSeekingFileReader` * Tidying up * Missing licenses * Re-instate custom override for `HoodieAvroParquetReader`; Tidying up * Fixed missing cloning w/in `HoodieLazyInsertIterable` * Fixed missing cloning in deduplication flow * Allow `HoodieSparkRecord` to hold `ColumnarBatchRow` * Missing licenses * Fixing compilation * Missing changes * Fixed Spark 2.x validation whether the row was read as a batch Fix comment in RFC46 (apache#6745) * rename * add MetadataValues in updateMetadataValues * remove singleton in fileFactory * add truncateRecordKey * remove hoodieRecord#setData * rename HoodieAvroRecord * fix code style * fix HoodieSparkRecordSerializer * fix benchmark * fix SparkRecordUtils * instantiate HoodieWriteConfig on the fly * add test * fix HoodieSparkRecordSerializer. Replace Java's object serialization with kryo * add broadcast * fix comment * remove unnecessary broadcast * add unsafe check in spark record * fix getRecordColumnValues * remove spark.sql.parquet.writeLegacyFormat * fix unsafe projection * fix * pass external schema * update doc * rename back to HoodieAvroRecord * fix * remove comparable wrapper * fix comment * fix comment * fix comment * fix comment * simplify row copy * fix ParquetReaderIterator Co-authored-by: Shawy Geng <[email protected]> Co-authored-by: wangzixuan.wzxuan <[email protected]> [RFC-46][HUDI-4414] Update the RFC-46 doc to fix comments feedback (apache#6132) * Update the RFC-46 doc to fix comments feedback * fix Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4301] [HUDI-3384][HUDI-3385] Spark specific file reader/writer.(apache#5629) * [HUDI-4301] [HUDI-3384][HUDI-3385] Spark specific file reader/writer. * add schema finger print * add benchmark * a new way to config the merger * fix Co-authored-by: wangzixuan.wzxuan <[email protected]> Co-authored-by: gengxiaoyu <[email protected]> [HUDI-3350][HUDI-3351] Support HoodieMerge API and Spark engine-specific HoodieRecord (apache#5627) Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4344] fix usage of HoodieDataBlock#getRecordIterator (apache#6005) Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4292][RFC-46] Update doc to align with the Record Merge API changes (apache#5927) [MINOR] Fix type casting in TestHoodieHFileReaderWriter [HUDI-3378][HUDI-3379][HUDI-3381] Migrate usage of HoodieRecordPayload and raw Avro payload to HoodieRecord (apache#5522) Co-authored-by: Alexey Kudinkin <[email protected]> Co-authored-by: wangzixuan.wzxuan <[email protected]>
wzx140
added a commit
to wzx140/hudi
that referenced
this pull request
Dec 9, 2022
[minor] add more test for rfc46 (apache#7003) ## Change Logs - Add HoodieSparkValidateDuplicateKeyRecordMerger behaving the same as ValidateDuplicateKeyPayload. We should use it with config "hoodie.sql.insert.mode=strict". - Fix nest field exist in HoodieCatalystExpressionUtils - Fix rewrite in HoodieInternalRowUtiles to support type promoted as avro - Fallback to avro when use "merge into" sql - Fix some schema handling issue - Support delta streamer - Convert parquet schema to spark schema and then avro schema(in org.apache.hudi.io.storage.HoodieSparkParquetReader#getSchema). Some types in avro are not compatible with parquet. For ex, decimal as int32/int64 in parquet will convert to int/long in avro. Because avro do not has decimal as int/long . We will lose the logic type info if we directly convert it to avro schema. - Support schema evolution in parquet block [Minor] fix multi deser avro payload (apache#7021) In HoodieAvroRecord, we will call isDelete, shouldIgnore before we write it to the file. Each method will deserialize HoodiePayload. So we add deserialization method in HoodieRecord and call this method once before calling isDelete or shouldIgnore. Co-authored-by: wangzixuan.wzxuan <[email protected]> Co-authored-by: Alexey Kudinkin <[email protected]> Co-authored-by: Alexey Kudinkin <[email protected]> [MINOR] Properly registering target classes w/ Kryo (apache#7026) * Added `HoodieKryoRegistrar` registering necessary Hudi's classes w/ Kryo to make their serialization more efficient (by serializing just the class id, in-liue the fully qualified class-name) * Redirected Kryo registration to `HoodieKryoRegistrar` * Registered additional classes likely to be serialized by Kryo * Updated tests * Fixed serialization of Avro's `Utf8` to serialize just the bytes * Added tests * Added custom `AvroUtf8Serializer`; Tidying up * Extracted `HoodieCommonKryoRegistrar` to leverage in `SerializationUtils` * `HoodieKryoRegistrar` > `HoodieSparkKryoRegistrar`; Rebased `HoodieSparkKryoRegistrar` onto `HoodieCommonKryoRegistrar` * `lint` * Fixing compilation for Spark 2.x * Disabling flaky test [MINOR] Make sure all `HoodieRecord`s are appropriately serializable by Kryo (apache#6977) * Make sure `HoodieRecord`, `HoodieKey`, `HoodieRecordLocation` are all `KryoSerializable` * Revisited `HoodieRecord` serialization hooks to make sure they a) could not be overridden, b) provide for hooks to properly serialize record's payload; Implemented serialization hooks for `HoodieAvroIndexedRecord`; Implemented serialization hooks for `HoodieEmptyRecord`; * Revisited `HoodieRecord` serialization hooks to make sure they a) could not be overridden, b) provide for hooks to properly serialize record's payload; Implemented serialization hooks for `HoodieAvroIndexedRecord`; Implemented serialization hooks for `HoodieEmptyRecord`; Implemented serialization hooks for `HoodieAvroRecord`; * Revisited `HoodieSparkRecord` to transiently hold on to the schema so that it could project row * Implemented serialization hooks for `HoodieSparkRecord` * Added `TestHoodieSparkRecord` * Added tests for Avro-based records * Added test for `HoodieEmptyRecord` * Fixed sealing/unsealing for `HoodieRecord` in `HoodieBackedTableMetadataWriter` * Properly handle deflated records * Fixing `Row`s encoding * Fixed `HoodieRecord` to be properly sealed/unsealed * Fixed serialization of the `HoodieRecordGlobalLocation` [MINOR] Additional fixes for apache#6745 (apache#6947) * Tidying up * Tidying up more * Cleaning up duplication * Tidying up * Revisited legacy operating mode configuration * Tidying up * Cleaned up `projectUnsafe` API * Fixing compilation * Cleaning up `HoodieSparkRecord` ctors; Revisited mandatory unsafe-projection * Fixing compilation * Cleaned up `ParquetReader` initialization * Revisited `HoodieSparkRecord` to accept either `UnsafeRow` or `HoodieInternalRow`, and avoid unnecessary copying after unsafe-projection * Cleaning up redundant exception spec * Make sure `updateMetadataFields` properly wraps `InternalRow` into `HoodieInternalRow` if necessary; Cleaned up `MetadataValues` * Fixed meta-fields extraction and `HoodieInternalRow` composition w/in `HoodieSparkRecord` * De-duplicate `HoodieSparkRecord` ctors; Make sure either only `UnsafeRow` or `HoodieInternalRow` are permitted inside `HoodieSparkRecord` * Removed unnecessary copying * Cleaned up projection for `HoodieSparkRecord` (dropping partition columns); Removed unnecessary copying * Fixing compilation * Fixing compilation (for Flink) * Cleaned up File Raders' interfaces: - Extracted `HoodieSeekingFileReader` interface (for key-ranged reads) - Pushed down concrete implementation methods into `HoodieAvroFileReaderBase` from the interfaces * Cleaned up File Readers impls (inline with then new interfaces) * Rebsaed `HoodieBackedTableMetadata` onto new `HoodieSeekingFileReader` * Tidying up * Missing licenses * Re-instate custom override for `HoodieAvroParquetReader`; Tidying up * Fixed missing cloning w/in `HoodieLazyInsertIterable` * Fixed missing cloning in deduplication flow * Allow `HoodieSparkRecord` to hold `ColumnarBatchRow` * Missing licenses * Fixing compilation * Missing changes * Fixed Spark 2.x validation whether the row was read as a batch Fix comment in RFC46 (apache#6745) * rename * add MetadataValues in updateMetadataValues * remove singleton in fileFactory * add truncateRecordKey * remove hoodieRecord#setData * rename HoodieAvroRecord * fix code style * fix HoodieSparkRecordSerializer * fix benchmark * fix SparkRecordUtils * instantiate HoodieWriteConfig on the fly * add test * fix HoodieSparkRecordSerializer. Replace Java's object serialization with kryo * add broadcast * fix comment * remove unnecessary broadcast * add unsafe check in spark record * fix getRecordColumnValues * remove spark.sql.parquet.writeLegacyFormat * fix unsafe projection * fix * pass external schema * update doc * rename back to HoodieAvroRecord * fix * remove comparable wrapper * fix comment * fix comment * fix comment * fix comment * simplify row copy * fix ParquetReaderIterator Co-authored-by: Shawy Geng <[email protected]> Co-authored-by: wangzixuan.wzxuan <[email protected]> [RFC-46][HUDI-4414] Update the RFC-46 doc to fix comments feedback (apache#6132) * Update the RFC-46 doc to fix comments feedback * fix Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4301] [HUDI-3384][HUDI-3385] Spark specific file reader/writer.(apache#5629) * [HUDI-4301] [HUDI-3384][HUDI-3385] Spark specific file reader/writer. * add schema finger print * add benchmark * a new way to config the merger * fix Co-authored-by: wangzixuan.wzxuan <[email protected]> Co-authored-by: gengxiaoyu <[email protected]> [HUDI-3350][HUDI-3351] Support HoodieMerge API and Spark engine-specific HoodieRecord (apache#5627) Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4344] fix usage of HoodieDataBlock#getRecordIterator (apache#6005) Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4292][RFC-46] Update doc to align with the Record Merge API changes (apache#5927) [MINOR] Fix type casting in TestHoodieHFileReaderWriter [HUDI-3378][HUDI-3379][HUDI-3381] Migrate usage of HoodieRecordPayload and raw Avro payload to HoodieRecord (apache#5522) Co-authored-by: Alexey Kudinkin <[email protected]> Co-authored-by: wangzixuan.wzxuan <[email protected]>
wzx140
added a commit
to wzx140/hudi
that referenced
this pull request
Dec 13, 2022
[minor] add more test for rfc46 (apache#7003) ## Change Logs - Add HoodieSparkValidateDuplicateKeyRecordMerger behaving the same as ValidateDuplicateKeyPayload. We should use it with config "hoodie.sql.insert.mode=strict". - Fix nest field exist in HoodieCatalystExpressionUtils - Fix rewrite in HoodieInternalRowUtiles to support type promoted as avro - Fallback to avro when use "merge into" sql - Fix some schema handling issue - Support delta streamer - Convert parquet schema to spark schema and then avro schema(in org.apache.hudi.io.storage.HoodieSparkParquetReader#getSchema). Some types in avro are not compatible with parquet. For ex, decimal as int32/int64 in parquet will convert to int/long in avro. Because avro do not has decimal as int/long . We will lose the logic type info if we directly convert it to avro schema. - Support schema evolution in parquet block [Minor] fix multi deser avro payload (apache#7021) In HoodieAvroRecord, we will call isDelete, shouldIgnore before we write it to the file. Each method will deserialize HoodiePayload. So we add deserialization method in HoodieRecord and call this method once before calling isDelete or shouldIgnore. Co-authored-by: wangzixuan.wzxuan <[email protected]> Co-authored-by: Alexey Kudinkin <[email protected]> Co-authored-by: Alexey Kudinkin <[email protected]> [MINOR] Properly registering target classes w/ Kryo (apache#7026) * Added `HoodieKryoRegistrar` registering necessary Hudi's classes w/ Kryo to make their serialization more efficient (by serializing just the class id, in-liue the fully qualified class-name) * Redirected Kryo registration to `HoodieKryoRegistrar` * Registered additional classes likely to be serialized by Kryo * Updated tests * Fixed serialization of Avro's `Utf8` to serialize just the bytes * Added tests * Added custom `AvroUtf8Serializer`; Tidying up * Extracted `HoodieCommonKryoRegistrar` to leverage in `SerializationUtils` * `HoodieKryoRegistrar` > `HoodieSparkKryoRegistrar`; Rebased `HoodieSparkKryoRegistrar` onto `HoodieCommonKryoRegistrar` * `lint` * Fixing compilation for Spark 2.x * Disabling flaky test [MINOR] Make sure all `HoodieRecord`s are appropriately serializable by Kryo (apache#6977) * Make sure `HoodieRecord`, `HoodieKey`, `HoodieRecordLocation` are all `KryoSerializable` * Revisited `HoodieRecord` serialization hooks to make sure they a) could not be overridden, b) provide for hooks to properly serialize record's payload; Implemented serialization hooks for `HoodieAvroIndexedRecord`; Implemented serialization hooks for `HoodieEmptyRecord`; * Revisited `HoodieRecord` serialization hooks to make sure they a) could not be overridden, b) provide for hooks to properly serialize record's payload; Implemented serialization hooks for `HoodieAvroIndexedRecord`; Implemented serialization hooks for `HoodieEmptyRecord`; Implemented serialization hooks for `HoodieAvroRecord`; * Revisited `HoodieSparkRecord` to transiently hold on to the schema so that it could project row * Implemented serialization hooks for `HoodieSparkRecord` * Added `TestHoodieSparkRecord` * Added tests for Avro-based records * Added test for `HoodieEmptyRecord` * Fixed sealing/unsealing for `HoodieRecord` in `HoodieBackedTableMetadataWriter` * Properly handle deflated records * Fixing `Row`s encoding * Fixed `HoodieRecord` to be properly sealed/unsealed * Fixed serialization of the `HoodieRecordGlobalLocation` [MINOR] Additional fixes for apache#6745 (apache#6947) * Tidying up * Tidying up more * Cleaning up duplication * Tidying up * Revisited legacy operating mode configuration * Tidying up * Cleaned up `projectUnsafe` API * Fixing compilation * Cleaning up `HoodieSparkRecord` ctors; Revisited mandatory unsafe-projection * Fixing compilation * Cleaned up `ParquetReader` initialization * Revisited `HoodieSparkRecord` to accept either `UnsafeRow` or `HoodieInternalRow`, and avoid unnecessary copying after unsafe-projection * Cleaning up redundant exception spec * Make sure `updateMetadataFields` properly wraps `InternalRow` into `HoodieInternalRow` if necessary; Cleaned up `MetadataValues` * Fixed meta-fields extraction and `HoodieInternalRow` composition w/in `HoodieSparkRecord` * De-duplicate `HoodieSparkRecord` ctors; Make sure either only `UnsafeRow` or `HoodieInternalRow` are permitted inside `HoodieSparkRecord` * Removed unnecessary copying * Cleaned up projection for `HoodieSparkRecord` (dropping partition columns); Removed unnecessary copying * Fixing compilation * Fixing compilation (for Flink) * Cleaned up File Raders' interfaces: - Extracted `HoodieSeekingFileReader` interface (for key-ranged reads) - Pushed down concrete implementation methods into `HoodieAvroFileReaderBase` from the interfaces * Cleaned up File Readers impls (inline with then new interfaces) * Rebsaed `HoodieBackedTableMetadata` onto new `HoodieSeekingFileReader` * Tidying up * Missing licenses * Re-instate custom override for `HoodieAvroParquetReader`; Tidying up * Fixed missing cloning w/in `HoodieLazyInsertIterable` * Fixed missing cloning in deduplication flow * Allow `HoodieSparkRecord` to hold `ColumnarBatchRow` * Missing licenses * Fixing compilation * Missing changes * Fixed Spark 2.x validation whether the row was read as a batch Fix comment in RFC46 (apache#6745) * rename * add MetadataValues in updateMetadataValues * remove singleton in fileFactory * add truncateRecordKey * remove hoodieRecord#setData * rename HoodieAvroRecord * fix code style * fix HoodieSparkRecordSerializer * fix benchmark * fix SparkRecordUtils * instantiate HoodieWriteConfig on the fly * add test * fix HoodieSparkRecordSerializer. Replace Java's object serialization with kryo * add broadcast * fix comment * remove unnecessary broadcast * add unsafe check in spark record * fix getRecordColumnValues * remove spark.sql.parquet.writeLegacyFormat * fix unsafe projection * fix * pass external schema * update doc * rename back to HoodieAvroRecord * fix * remove comparable wrapper * fix comment * fix comment * fix comment * fix comment * simplify row copy * fix ParquetReaderIterator Co-authored-by: Shawy Geng <[email protected]> Co-authored-by: wangzixuan.wzxuan <[email protected]> [RFC-46][HUDI-4414] Update the RFC-46 doc to fix comments feedback (apache#6132) * Update the RFC-46 doc to fix comments feedback * fix Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4301] [HUDI-3384][HUDI-3385] Spark specific file reader/writer.(apache#5629) * [HUDI-4301] [HUDI-3384][HUDI-3385] Spark specific file reader/writer. * add schema finger print * add benchmark * a new way to config the merger * fix Co-authored-by: wangzixuan.wzxuan <[email protected]> Co-authored-by: gengxiaoyu <[email protected]> [HUDI-3350][HUDI-3351] Support HoodieMerge API and Spark engine-specific HoodieRecord (apache#5627) Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4344] fix usage of HoodieDataBlock#getRecordIterator (apache#6005) Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4292][RFC-46] Update doc to align with the Record Merge API changes (apache#5927) [MINOR] Fix type casting in TestHoodieHFileReaderWriter [HUDI-3378][HUDI-3379][HUDI-3381] Migrate usage of HoodieRecordPayload and raw Avro payload to HoodieRecord (apache#5522) Co-authored-by: Alexey Kudinkin <[email protected]> Co-authored-by: wangzixuan.wzxuan <[email protected]>
alexeykudinkin
pushed a commit
to wzx140/hudi
that referenced
this pull request
Dec 13, 2022
[minor] add more test for rfc46 (apache#7003) - Add HoodieSparkValidateDuplicateKeyRecordMerger behaving the same as ValidateDuplicateKeyPayload. We should use it with config "hoodie.sql.insert.mode=strict". - Fix nest field exist in HoodieCatalystExpressionUtils - Fix rewrite in HoodieInternalRowUtiles to support type promoted as avro - Fallback to avro when use "merge into" sql - Fix some schema handling issue - Support delta streamer - Convert parquet schema to spark schema and then avro schema(in org.apache.hudi.io.storage.HoodieSparkParquetReader#getSchema). Some types in avro are not compatible with parquet. For ex, decimal as int32/int64 in parquet will convert to int/long in avro. Because avro do not has decimal as int/long . We will lose the logic type info if we directly convert it to avro schema. - Support schema evolution in parquet block [Minor] fix multi deser avro payload (apache#7021) In HoodieAvroRecord, we will call isDelete, shouldIgnore before we write it to the file. Each method will deserialize HoodiePayload. So we add deserialization method in HoodieRecord and call this method once before calling isDelete or shouldIgnore. Co-authored-by: wangzixuan.wzxuan <[email protected]> Co-authored-by: Alexey Kudinkin <[email protected]> Co-authored-by: Alexey Kudinkin <[email protected]> [MINOR] Properly registering target classes w/ Kryo (apache#7026) * Added `HoodieKryoRegistrar` registering necessary Hudi's classes w/ Kryo to make their serialization more efficient (by serializing just the class id, in-liue the fully qualified class-name) * Redirected Kryo registration to `HoodieKryoRegistrar` * Registered additional classes likely to be serialized by Kryo * Updated tests * Fixed serialization of Avro's `Utf8` to serialize just the bytes * Added tests * Added custom `AvroUtf8Serializer`; Tidying up * Extracted `HoodieCommonKryoRegistrar` to leverage in `SerializationUtils` * `HoodieKryoRegistrar` > `HoodieSparkKryoRegistrar`; Rebased `HoodieSparkKryoRegistrar` onto `HoodieCommonKryoRegistrar` * `lint` * Fixing compilation for Spark 2.x * Disabling flaky test [MINOR] Make sure all `HoodieRecord`s are appropriately serializable by Kryo (apache#6977) * Make sure `HoodieRecord`, `HoodieKey`, `HoodieRecordLocation` are all `KryoSerializable` * Revisited `HoodieRecord` serialization hooks to make sure they a) could not be overridden, b) provide for hooks to properly serialize record's payload; Implemented serialization hooks for `HoodieAvroIndexedRecord`; Implemented serialization hooks for `HoodieEmptyRecord`; * Revisited `HoodieRecord` serialization hooks to make sure they a) could not be overridden, b) provide for hooks to properly serialize record's payload; Implemented serialization hooks for `HoodieAvroIndexedRecord`; Implemented serialization hooks for `HoodieEmptyRecord`; Implemented serialization hooks for `HoodieAvroRecord`; * Revisited `HoodieSparkRecord` to transiently hold on to the schema so that it could project row * Implemented serialization hooks for `HoodieSparkRecord` * Added `TestHoodieSparkRecord` * Added tests for Avro-based records * Added test for `HoodieEmptyRecord` * Fixed sealing/unsealing for `HoodieRecord` in `HoodieBackedTableMetadataWriter` * Properly handle deflated records * Fixing `Row`s encoding * Fixed `HoodieRecord` to be properly sealed/unsealed * Fixed serialization of the `HoodieRecordGlobalLocation` [MINOR] Additional fixes for apache#6745 (apache#6947) * Tidying up * Tidying up more * Cleaning up duplication * Tidying up * Revisited legacy operating mode configuration * Tidying up * Cleaned up `projectUnsafe` API * Fixing compilation * Cleaning up `HoodieSparkRecord` ctors; Revisited mandatory unsafe-projection * Fixing compilation * Cleaned up `ParquetReader` initialization * Revisited `HoodieSparkRecord` to accept either `UnsafeRow` or `HoodieInternalRow`, and avoid unnecessary copying after unsafe-projection * Cleaning up redundant exception spec * Make sure `updateMetadataFields` properly wraps `InternalRow` into `HoodieInternalRow` if necessary; Cleaned up `MetadataValues` * Fixed meta-fields extraction and `HoodieInternalRow` composition w/in `HoodieSparkRecord` * De-duplicate `HoodieSparkRecord` ctors; Make sure either only `UnsafeRow` or `HoodieInternalRow` are permitted inside `HoodieSparkRecord` * Removed unnecessary copying * Cleaned up projection for `HoodieSparkRecord` (dropping partition columns); Removed unnecessary copying * Fixing compilation * Fixing compilation (for Flink) * Cleaned up File Raders' interfaces: - Extracted `HoodieSeekingFileReader` interface (for key-ranged reads) - Pushed down concrete implementation methods into `HoodieAvroFileReaderBase` from the interfaces * Cleaned up File Readers impls (inline with then new interfaces) * Rebsaed `HoodieBackedTableMetadata` onto new `HoodieSeekingFileReader` * Tidying up * Missing licenses * Re-instate custom override for `HoodieAvroParquetReader`; Tidying up * Fixed missing cloning w/in `HoodieLazyInsertIterable` * Fixed missing cloning in deduplication flow * Allow `HoodieSparkRecord` to hold `ColumnarBatchRow` * Missing licenses * Fixing compilation * Missing changes * Fixed Spark 2.x validation whether the row was read as a batch Fix comment in RFC46 (apache#6745) * rename * add MetadataValues in updateMetadataValues * remove singleton in fileFactory * add truncateRecordKey * remove hoodieRecord#setData * rename HoodieAvroRecord * fix code style * fix HoodieSparkRecordSerializer * fix benchmark * fix SparkRecordUtils * instantiate HoodieWriteConfig on the fly * add test * fix HoodieSparkRecordSerializer. Replace Java's object serialization with kryo * add broadcast * fix comment * remove unnecessary broadcast * add unsafe check in spark record * fix getRecordColumnValues * remove spark.sql.parquet.writeLegacyFormat * fix unsafe projection * fix * pass external schema * update doc * rename back to HoodieAvroRecord * fix * remove comparable wrapper * fix comment * fix comment * fix comment * fix comment * simplify row copy * fix ParquetReaderIterator Co-authored-by: Shawy Geng <[email protected]> Co-authored-by: wangzixuan.wzxuan <[email protected]> [RFC-46][HUDI-4414] Update the RFC-46 doc to fix comments feedback (apache#6132) * Update the RFC-46 doc to fix comments feedback * fix Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4301] [HUDI-3384][HUDI-3385] Spark specific file reader/writer.(apache#5629) * [HUDI-4301] [HUDI-3384][HUDI-3385] Spark specific file reader/writer. * add schema finger print * add benchmark * a new way to config the merger * fix Co-authored-by: wangzixuan.wzxuan <[email protected]> Co-authored-by: gengxiaoyu <[email protected]> [HUDI-3350][HUDI-3351] Support HoodieMerge API and Spark engine-specific HoodieRecord (apache#5627) Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4344] fix usage of HoodieDataBlock#getRecordIterator (apache#6005) Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4292][RFC-46] Update doc to align with the Record Merge API changes (apache#5927) [MINOR] Fix type casting in TestHoodieHFileReaderWriter [HUDI-3378][HUDI-3379][HUDI-3381] Migrate usage of HoodieRecordPayload and raw Avro payload to HoodieRecord (apache#5522) Co-authored-by: Alexey Kudinkin <[email protected]> Co-authored-by: wangzixuan.wzxuan <[email protected]>
wzx140
added a commit
to wzx140/hudi
that referenced
this pull request
Dec 13, 2022
[minor] add more test for rfc46 (apache#7003) ## Change Logs - Add HoodieSparkValidateDuplicateKeyRecordMerger behaving the same as ValidateDuplicateKeyPayload. We should use it with config "hoodie.sql.insert.mode=strict". - Fix nest field exist in HoodieCatalystExpressionUtils - Fix rewrite in HoodieInternalRowUtiles to support type promoted as avro - Fallback to avro when use "merge into" sql - Fix some schema handling issue - Support delta streamer - Convert parquet schema to spark schema and then avro schema(in org.apache.hudi.io.storage.HoodieSparkParquetReader#getSchema). Some types in avro are not compatible with parquet. For ex, decimal as int32/int64 in parquet will convert to int/long in avro. Because avro do not has decimal as int/long . We will lose the logic type info if we directly convert it to avro schema. - Support schema evolution in parquet block [Minor] fix multi deser avro payload (apache#7021) In HoodieAvroRecord, we will call isDelete, shouldIgnore before we write it to the file. Each method will deserialize HoodiePayload. So we add deserialization method in HoodieRecord and call this method once before calling isDelete or shouldIgnore. Co-authored-by: wangzixuan.wzxuan <[email protected]> Co-authored-by: Alexey Kudinkin <[email protected]> Co-authored-by: Alexey Kudinkin <[email protected]> [MINOR] Properly registering target classes w/ Kryo (apache#7026) * Added `HoodieKryoRegistrar` registering necessary Hudi's classes w/ Kryo to make their serialization more efficient (by serializing just the class id, in-liue the fully qualified class-name) * Redirected Kryo registration to `HoodieKryoRegistrar` * Registered additional classes likely to be serialized by Kryo * Updated tests * Fixed serialization of Avro's `Utf8` to serialize just the bytes * Added tests * Added custom `AvroUtf8Serializer`; Tidying up * Extracted `HoodieCommonKryoRegistrar` to leverage in `SerializationUtils` * `HoodieKryoRegistrar` > `HoodieSparkKryoRegistrar`; Rebased `HoodieSparkKryoRegistrar` onto `HoodieCommonKryoRegistrar` * `lint` * Fixing compilation for Spark 2.x * Disabling flaky test [MINOR] Make sure all `HoodieRecord`s are appropriately serializable by Kryo (apache#6977) * Make sure `HoodieRecord`, `HoodieKey`, `HoodieRecordLocation` are all `KryoSerializable` * Revisited `HoodieRecord` serialization hooks to make sure they a) could not be overridden, b) provide for hooks to properly serialize record's payload; Implemented serialization hooks for `HoodieAvroIndexedRecord`; Implemented serialization hooks for `HoodieEmptyRecord`; * Revisited `HoodieRecord` serialization hooks to make sure they a) could not be overridden, b) provide for hooks to properly serialize record's payload; Implemented serialization hooks for `HoodieAvroIndexedRecord`; Implemented serialization hooks for `HoodieEmptyRecord`; Implemented serialization hooks for `HoodieAvroRecord`; * Revisited `HoodieSparkRecord` to transiently hold on to the schema so that it could project row * Implemented serialization hooks for `HoodieSparkRecord` * Added `TestHoodieSparkRecord` * Added tests for Avro-based records * Added test for `HoodieEmptyRecord` * Fixed sealing/unsealing for `HoodieRecord` in `HoodieBackedTableMetadataWriter` * Properly handle deflated records * Fixing `Row`s encoding * Fixed `HoodieRecord` to be properly sealed/unsealed * Fixed serialization of the `HoodieRecordGlobalLocation` [MINOR] Additional fixes for apache#6745 (apache#6947) * Tidying up * Tidying up more * Cleaning up duplication * Tidying up * Revisited legacy operating mode configuration * Tidying up * Cleaned up `projectUnsafe` API * Fixing compilation * Cleaning up `HoodieSparkRecord` ctors; Revisited mandatory unsafe-projection * Fixing compilation * Cleaned up `ParquetReader` initialization * Revisited `HoodieSparkRecord` to accept either `UnsafeRow` or `HoodieInternalRow`, and avoid unnecessary copying after unsafe-projection * Cleaning up redundant exception spec * Make sure `updateMetadataFields` properly wraps `InternalRow` into `HoodieInternalRow` if necessary; Cleaned up `MetadataValues` * Fixed meta-fields extraction and `HoodieInternalRow` composition w/in `HoodieSparkRecord` * De-duplicate `HoodieSparkRecord` ctors; Make sure either only `UnsafeRow` or `HoodieInternalRow` are permitted inside `HoodieSparkRecord` * Removed unnecessary copying * Cleaned up projection for `HoodieSparkRecord` (dropping partition columns); Removed unnecessary copying * Fixing compilation * Fixing compilation (for Flink) * Cleaned up File Raders' interfaces: - Extracted `HoodieSeekingFileReader` interface (for key-ranged reads) - Pushed down concrete implementation methods into `HoodieAvroFileReaderBase` from the interfaces * Cleaned up File Readers impls (inline with then new interfaces) * Rebsaed `HoodieBackedTableMetadata` onto new `HoodieSeekingFileReader` * Tidying up * Missing licenses * Re-instate custom override for `HoodieAvroParquetReader`; Tidying up * Fixed missing cloning w/in `HoodieLazyInsertIterable` * Fixed missing cloning in deduplication flow * Allow `HoodieSparkRecord` to hold `ColumnarBatchRow` * Missing licenses * Fixing compilation * Missing changes * Fixed Spark 2.x validation whether the row was read as a batch Fix comment in RFC46 (apache#6745) * rename * add MetadataValues in updateMetadataValues * remove singleton in fileFactory * add truncateRecordKey * remove hoodieRecord#setData * rename HoodieAvroRecord * fix code style * fix HoodieSparkRecordSerializer * fix benchmark * fix SparkRecordUtils * instantiate HoodieWriteConfig on the fly * add test * fix HoodieSparkRecordSerializer. Replace Java's object serialization with kryo * add broadcast * fix comment * remove unnecessary broadcast * add unsafe check in spark record * fix getRecordColumnValues * remove spark.sql.parquet.writeLegacyFormat * fix unsafe projection * fix * pass external schema * update doc * rename back to HoodieAvroRecord * fix * remove comparable wrapper * fix comment * fix comment * fix comment * fix comment * simplify row copy * fix ParquetReaderIterator Co-authored-by: Shawy Geng <[email protected]> Co-authored-by: wangzixuan.wzxuan <[email protected]> [RFC-46][HUDI-4414] Update the RFC-46 doc to fix comments feedback (apache#6132) * Update the RFC-46 doc to fix comments feedback * fix Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4301] [HUDI-3384][HUDI-3385] Spark specific file reader/writer.(apache#5629) * [HUDI-4301] [HUDI-3384][HUDI-3385] Spark specific file reader/writer. * add schema finger print * add benchmark * a new way to config the merger * fix Co-authored-by: wangzixuan.wzxuan <[email protected]> Co-authored-by: gengxiaoyu <[email protected]> [HUDI-3350][HUDI-3351] Support HoodieMerge API and Spark engine-specific HoodieRecord (apache#5627) Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4344] fix usage of HoodieDataBlock#getRecordIterator (apache#6005) Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4292][RFC-46] Update doc to align with the Record Merge API changes (apache#5927) [MINOR] Fix type casting in TestHoodieHFileReaderWriter [HUDI-3378][HUDI-3379][HUDI-3381] Migrate usage of HoodieRecordPayload and raw Avro payload to HoodieRecord (apache#5522) Co-authored-by: Alexey Kudinkin <[email protected]> Co-authored-by: wangzixuan.wzxuan <[email protected]>
alexeykudinkin
pushed a commit
to wzx140/hudi
that referenced
this pull request
Dec 13, 2022
[minor] add more test for rfc46 (apache#7003) - Add HoodieSparkValidateDuplicateKeyRecordMerger behaving the same as ValidateDuplicateKeyPayload. We should use it with config "hoodie.sql.insert.mode=strict". - Fix nest field exist in HoodieCatalystExpressionUtils - Fix rewrite in HoodieInternalRowUtiles to support type promoted as avro - Fallback to avro when use "merge into" sql - Fix some schema handling issue - Support delta streamer - Convert parquet schema to spark schema and then avro schema(in org.apache.hudi.io.storage.HoodieSparkParquetReader#getSchema). Some types in avro are not compatible with parquet. For ex, decimal as int32/int64 in parquet will convert to int/long in avro. Because avro do not has decimal as int/long . We will lose the logic type info if we directly convert it to avro schema. - Support schema evolution in parquet block [Minor] fix multi deser avro payload (apache#7021) In HoodieAvroRecord, we will call isDelete, shouldIgnore before we write it to the file. Each method will deserialize HoodiePayload. So we add deserialization method in HoodieRecord and call this method once before calling isDelete or shouldIgnore. Co-authored-by: wangzixuan.wzxuan <[email protected]> Co-authored-by: Alexey Kudinkin <[email protected]> Co-authored-by: Alexey Kudinkin <[email protected]> [MINOR] Properly registering target classes w/ Kryo (apache#7026) * Added `HoodieKryoRegistrar` registering necessary Hudi's classes w/ Kryo to make their serialization more efficient (by serializing just the class id, in-liue the fully qualified class-name) * Redirected Kryo registration to `HoodieKryoRegistrar` * Registered additional classes likely to be serialized by Kryo * Updated tests * Fixed serialization of Avro's `Utf8` to serialize just the bytes * Added tests * Added custom `AvroUtf8Serializer`; Tidying up * Extracted `HoodieCommonKryoRegistrar` to leverage in `SerializationUtils` * `HoodieKryoRegistrar` > `HoodieSparkKryoRegistrar`; Rebased `HoodieSparkKryoRegistrar` onto `HoodieCommonKryoRegistrar` * `lint` * Fixing compilation for Spark 2.x * Disabling flaky test [MINOR] Make sure all `HoodieRecord`s are appropriately serializable by Kryo (apache#6977) * Make sure `HoodieRecord`, `HoodieKey`, `HoodieRecordLocation` are all `KryoSerializable` * Revisited `HoodieRecord` serialization hooks to make sure they a) could not be overridden, b) provide for hooks to properly serialize record's payload; Implemented serialization hooks for `HoodieAvroIndexedRecord`; Implemented serialization hooks for `HoodieEmptyRecord`; * Revisited `HoodieRecord` serialization hooks to make sure they a) could not be overridden, b) provide for hooks to properly serialize record's payload; Implemented serialization hooks for `HoodieAvroIndexedRecord`; Implemented serialization hooks for `HoodieEmptyRecord`; Implemented serialization hooks for `HoodieAvroRecord`; * Revisited `HoodieSparkRecord` to transiently hold on to the schema so that it could project row * Implemented serialization hooks for `HoodieSparkRecord` * Added `TestHoodieSparkRecord` * Added tests for Avro-based records * Added test for `HoodieEmptyRecord` * Fixed sealing/unsealing for `HoodieRecord` in `HoodieBackedTableMetadataWriter` * Properly handle deflated records * Fixing `Row`s encoding * Fixed `HoodieRecord` to be properly sealed/unsealed * Fixed serialization of the `HoodieRecordGlobalLocation` [MINOR] Additional fixes for apache#6745 (apache#6947) * Tidying up * Tidying up more * Cleaning up duplication * Tidying up * Revisited legacy operating mode configuration * Tidying up * Cleaned up `projectUnsafe` API * Fixing compilation * Cleaning up `HoodieSparkRecord` ctors; Revisited mandatory unsafe-projection * Fixing compilation * Cleaned up `ParquetReader` initialization * Revisited `HoodieSparkRecord` to accept either `UnsafeRow` or `HoodieInternalRow`, and avoid unnecessary copying after unsafe-projection * Cleaning up redundant exception spec * Make sure `updateMetadataFields` properly wraps `InternalRow` into `HoodieInternalRow` if necessary; Cleaned up `MetadataValues` * Fixed meta-fields extraction and `HoodieInternalRow` composition w/in `HoodieSparkRecord` * De-duplicate `HoodieSparkRecord` ctors; Make sure either only `UnsafeRow` or `HoodieInternalRow` are permitted inside `HoodieSparkRecord` * Removed unnecessary copying * Cleaned up projection for `HoodieSparkRecord` (dropping partition columns); Removed unnecessary copying * Fixing compilation * Fixing compilation (for Flink) * Cleaned up File Raders' interfaces: - Extracted `HoodieSeekingFileReader` interface (for key-ranged reads) - Pushed down concrete implementation methods into `HoodieAvroFileReaderBase` from the interfaces * Cleaned up File Readers impls (inline with then new interfaces) * Rebsaed `HoodieBackedTableMetadata` onto new `HoodieSeekingFileReader` * Tidying up * Missing licenses * Re-instate custom override for `HoodieAvroParquetReader`; Tidying up * Fixed missing cloning w/in `HoodieLazyInsertIterable` * Fixed missing cloning in deduplication flow * Allow `HoodieSparkRecord` to hold `ColumnarBatchRow` * Missing licenses * Fixing compilation * Missing changes * Fixed Spark 2.x validation whether the row was read as a batch Fix comment in RFC46 (apache#6745) * rename * add MetadataValues in updateMetadataValues * remove singleton in fileFactory * add truncateRecordKey * remove hoodieRecord#setData * rename HoodieAvroRecord * fix code style * fix HoodieSparkRecordSerializer * fix benchmark * fix SparkRecordUtils * instantiate HoodieWriteConfig on the fly * add test * fix HoodieSparkRecordSerializer. Replace Java's object serialization with kryo * add broadcast * fix comment * remove unnecessary broadcast * add unsafe check in spark record * fix getRecordColumnValues * remove spark.sql.parquet.writeLegacyFormat * fix unsafe projection * fix * pass external schema * update doc * rename back to HoodieAvroRecord * fix * remove comparable wrapper * fix comment * fix comment * fix comment * fix comment * simplify row copy * fix ParquetReaderIterator Co-authored-by: Shawy Geng <[email protected]> Co-authored-by: wangzixuan.wzxuan <[email protected]> [RFC-46][HUDI-4414] Update the RFC-46 doc to fix comments feedback (apache#6132) * Update the RFC-46 doc to fix comments feedback * fix Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4301] [HUDI-3384][HUDI-3385] Spark specific file reader/writer.(apache#5629) * [HUDI-4301] [HUDI-3384][HUDI-3385] Spark specific file reader/writer. * add schema finger print * add benchmark * a new way to config the merger * fix Co-authored-by: wangzixuan.wzxuan <[email protected]> Co-authored-by: gengxiaoyu <[email protected]> [HUDI-3350][HUDI-3351] Support HoodieMerge API and Spark engine-specific HoodieRecord (apache#5627) Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4344] fix usage of HoodieDataBlock#getRecordIterator (apache#6005) Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4292][RFC-46] Update doc to align with the Record Merge API changes (apache#5927) [MINOR] Fix type casting in TestHoodieHFileReaderWriter [HUDI-3378][HUDI-3379][HUDI-3381] Migrate usage of HoodieRecordPayload and raw Avro payload to HoodieRecord (apache#5522) Co-authored-by: Alexey Kudinkin <[email protected]> Co-authored-by: wangzixuan.wzxuan <[email protected]>
alexeykudinkin
pushed a commit
to onehouseinc/hudi
that referenced
this pull request
Dec 14, 2022
[minor] add more test for rfc46 (apache#7003) - Add HoodieSparkValidateDuplicateKeyRecordMerger behaving the same as ValidateDuplicateKeyPayload. We should use it with config "hoodie.sql.insert.mode=strict". - Fix nest field exist in HoodieCatalystExpressionUtils - Fix rewrite in HoodieInternalRowUtiles to support type promoted as avro - Fallback to avro when use "merge into" sql - Fix some schema handling issue - Support delta streamer - Convert parquet schema to spark schema and then avro schema(in org.apache.hudi.io.storage.HoodieSparkParquetReader#getSchema). Some types in avro are not compatible with parquet. For ex, decimal as int32/int64 in parquet will convert to int/long in avro. Because avro do not has decimal as int/long . We will lose the logic type info if we directly convert it to avro schema. - Support schema evolution in parquet block [Minor] fix multi deser avro payload (apache#7021) In HoodieAvroRecord, we will call isDelete, shouldIgnore before we write it to the file. Each method will deserialize HoodiePayload. So we add deserialization method in HoodieRecord and call this method once before calling isDelete or shouldIgnore. Co-authored-by: wangzixuan.wzxuan <[email protected]> Co-authored-by: Alexey Kudinkin <[email protected]> Co-authored-by: Alexey Kudinkin <[email protected]> [MINOR] Properly registering target classes w/ Kryo (apache#7026) * Added `HoodieKryoRegistrar` registering necessary Hudi's classes w/ Kryo to make their serialization more efficient (by serializing just the class id, in-liue the fully qualified class-name) * Redirected Kryo registration to `HoodieKryoRegistrar` * Registered additional classes likely to be serialized by Kryo * Updated tests * Fixed serialization of Avro's `Utf8` to serialize just the bytes * Added tests * Added custom `AvroUtf8Serializer`; Tidying up * Extracted `HoodieCommonKryoRegistrar` to leverage in `SerializationUtils` * `HoodieKryoRegistrar` > `HoodieSparkKryoRegistrar`; Rebased `HoodieSparkKryoRegistrar` onto `HoodieCommonKryoRegistrar` * `lint` * Fixing compilation for Spark 2.x * Disabling flaky test [MINOR] Make sure all `HoodieRecord`s are appropriately serializable by Kryo (apache#6977) * Make sure `HoodieRecord`, `HoodieKey`, `HoodieRecordLocation` are all `KryoSerializable` * Revisited `HoodieRecord` serialization hooks to make sure they a) could not be overridden, b) provide for hooks to properly serialize record's payload; Implemented serialization hooks for `HoodieAvroIndexedRecord`; Implemented serialization hooks for `HoodieEmptyRecord`; * Revisited `HoodieRecord` serialization hooks to make sure they a) could not be overridden, b) provide for hooks to properly serialize record's payload; Implemented serialization hooks for `HoodieAvroIndexedRecord`; Implemented serialization hooks for `HoodieEmptyRecord`; Implemented serialization hooks for `HoodieAvroRecord`; * Revisited `HoodieSparkRecord` to transiently hold on to the schema so that it could project row * Implemented serialization hooks for `HoodieSparkRecord` * Added `TestHoodieSparkRecord` * Added tests for Avro-based records * Added test for `HoodieEmptyRecord` * Fixed sealing/unsealing for `HoodieRecord` in `HoodieBackedTableMetadataWriter` * Properly handle deflated records * Fixing `Row`s encoding * Fixed `HoodieRecord` to be properly sealed/unsealed * Fixed serialization of the `HoodieRecordGlobalLocation` [MINOR] Additional fixes for apache#6745 (apache#6947) * Tidying up * Tidying up more * Cleaning up duplication * Tidying up * Revisited legacy operating mode configuration * Tidying up * Cleaned up `projectUnsafe` API * Fixing compilation * Cleaning up `HoodieSparkRecord` ctors; Revisited mandatory unsafe-projection * Fixing compilation * Cleaned up `ParquetReader` initialization * Revisited `HoodieSparkRecord` to accept either `UnsafeRow` or `HoodieInternalRow`, and avoid unnecessary copying after unsafe-projection * Cleaning up redundant exception spec * Make sure `updateMetadataFields` properly wraps `InternalRow` into `HoodieInternalRow` if necessary; Cleaned up `MetadataValues` * Fixed meta-fields extraction and `HoodieInternalRow` composition w/in `HoodieSparkRecord` * De-duplicate `HoodieSparkRecord` ctors; Make sure either only `UnsafeRow` or `HoodieInternalRow` are permitted inside `HoodieSparkRecord` * Removed unnecessary copying * Cleaned up projection for `HoodieSparkRecord` (dropping partition columns); Removed unnecessary copying * Fixing compilation * Fixing compilation (for Flink) * Cleaned up File Raders' interfaces: - Extracted `HoodieSeekingFileReader` interface (for key-ranged reads) - Pushed down concrete implementation methods into `HoodieAvroFileReaderBase` from the interfaces * Cleaned up File Readers impls (inline with then new interfaces) * Rebsaed `HoodieBackedTableMetadata` onto new `HoodieSeekingFileReader` * Tidying up * Missing licenses * Re-instate custom override for `HoodieAvroParquetReader`; Tidying up * Fixed missing cloning w/in `HoodieLazyInsertIterable` * Fixed missing cloning in deduplication flow * Allow `HoodieSparkRecord` to hold `ColumnarBatchRow` * Missing licenses * Fixing compilation * Missing changes * Fixed Spark 2.x validation whether the row was read as a batch Fix comment in RFC46 (apache#6745) * rename * add MetadataValues in updateMetadataValues * remove singleton in fileFactory * add truncateRecordKey * remove hoodieRecord#setData * rename HoodieAvroRecord * fix code style * fix HoodieSparkRecordSerializer * fix benchmark * fix SparkRecordUtils * instantiate HoodieWriteConfig on the fly * add test * fix HoodieSparkRecordSerializer. Replace Java's object serialization with kryo * add broadcast * fix comment * remove unnecessary broadcast * add unsafe check in spark record * fix getRecordColumnValues * remove spark.sql.parquet.writeLegacyFormat * fix unsafe projection * fix * pass external schema * update doc * rename back to HoodieAvroRecord * fix * remove comparable wrapper * fix comment * fix comment * fix comment * fix comment * simplify row copy * fix ParquetReaderIterator Co-authored-by: Shawy Geng <[email protected]> Co-authored-by: wangzixuan.wzxuan <[email protected]> [RFC-46][HUDI-4414] Update the RFC-46 doc to fix comments feedback (apache#6132) * Update the RFC-46 doc to fix comments feedback * fix Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4301] [HUDI-3384][HUDI-3385] Spark specific file reader/writer.(apache#5629) * [HUDI-4301] [HUDI-3384][HUDI-3385] Spark specific file reader/writer. * add schema finger print * add benchmark * a new way to config the merger * fix Co-authored-by: wangzixuan.wzxuan <[email protected]> Co-authored-by: gengxiaoyu <[email protected]> [HUDI-3350][HUDI-3351] Support HoodieMerge API and Spark engine-specific HoodieRecord (apache#5627) Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4344] fix usage of HoodieDataBlock#getRecordIterator (apache#6005) Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4292][RFC-46] Update doc to align with the Record Merge API changes (apache#5927) [MINOR] Fix type casting in TestHoodieHFileReaderWriter [HUDI-3378][HUDI-3379][HUDI-3381] Migrate usage of HoodieRecordPayload and raw Avro payload to HoodieRecord (apache#5522) Co-authored-by: Alexey Kudinkin <[email protected]> Co-authored-by: wangzixuan.wzxuan <[email protected]>
wzx140
added a commit
to wzx140/hudi
that referenced
this pull request
Dec 14, 2022
[minor] add more test for rfc46 (apache#7003) ## Change Logs - Add HoodieSparkValidateDuplicateKeyRecordMerger behaving the same as ValidateDuplicateKeyPayload. We should use it with config "hoodie.sql.insert.mode=strict". - Fix nest field exist in HoodieCatalystExpressionUtils - Fix rewrite in HoodieInternalRowUtiles to support type promoted as avro - Fallback to avro when use "merge into" sql - Fix some schema handling issue - Support delta streamer - Convert parquet schema to spark schema and then avro schema(in org.apache.hudi.io.storage.HoodieSparkParquetReader#getSchema). Some types in avro are not compatible with parquet. For ex, decimal as int32/int64 in parquet will convert to int/long in avro. Because avro do not has decimal as int/long . We will lose the logic type info if we directly convert it to avro schema. - Support schema evolution in parquet block [Minor] fix multi deser avro payload (apache#7021) In HoodieAvroRecord, we will call isDelete, shouldIgnore before we write it to the file. Each method will deserialize HoodiePayload. So we add deserialization method in HoodieRecord and call this method once before calling isDelete or shouldIgnore. Co-authored-by: wangzixuan.wzxuan <[email protected]> Co-authored-by: Alexey Kudinkin <[email protected]> Co-authored-by: Alexey Kudinkin <[email protected]> [MINOR] Properly registering target classes w/ Kryo (apache#7026) * Added `HoodieKryoRegistrar` registering necessary Hudi's classes w/ Kryo to make their serialization more efficient (by serializing just the class id, in-liue the fully qualified class-name) * Redirected Kryo registration to `HoodieKryoRegistrar` * Registered additional classes likely to be serialized by Kryo * Updated tests * Fixed serialization of Avro's `Utf8` to serialize just the bytes * Added tests * Added custom `AvroUtf8Serializer`; Tidying up * Extracted `HoodieCommonKryoRegistrar` to leverage in `SerializationUtils` * `HoodieKryoRegistrar` > `HoodieSparkKryoRegistrar`; Rebased `HoodieSparkKryoRegistrar` onto `HoodieCommonKryoRegistrar` * `lint` * Fixing compilation for Spark 2.x * Disabling flaky test [MINOR] Make sure all `HoodieRecord`s are appropriately serializable by Kryo (apache#6977) * Make sure `HoodieRecord`, `HoodieKey`, `HoodieRecordLocation` are all `KryoSerializable` * Revisited `HoodieRecord` serialization hooks to make sure they a) could not be overridden, b) provide for hooks to properly serialize record's payload; Implemented serialization hooks for `HoodieAvroIndexedRecord`; Implemented serialization hooks for `HoodieEmptyRecord`; * Revisited `HoodieRecord` serialization hooks to make sure they a) could not be overridden, b) provide for hooks to properly serialize record's payload; Implemented serialization hooks for `HoodieAvroIndexedRecord`; Implemented serialization hooks for `HoodieEmptyRecord`; Implemented serialization hooks for `HoodieAvroRecord`; * Revisited `HoodieSparkRecord` to transiently hold on to the schema so that it could project row * Implemented serialization hooks for `HoodieSparkRecord` * Added `TestHoodieSparkRecord` * Added tests for Avro-based records * Added test for `HoodieEmptyRecord` * Fixed sealing/unsealing for `HoodieRecord` in `HoodieBackedTableMetadataWriter` * Properly handle deflated records * Fixing `Row`s encoding * Fixed `HoodieRecord` to be properly sealed/unsealed * Fixed serialization of the `HoodieRecordGlobalLocation` [MINOR] Additional fixes for apache#6745 (apache#6947) * Tidying up * Tidying up more * Cleaning up duplication * Tidying up * Revisited legacy operating mode configuration * Tidying up * Cleaned up `projectUnsafe` API * Fixing compilation * Cleaning up `HoodieSparkRecord` ctors; Revisited mandatory unsafe-projection * Fixing compilation * Cleaned up `ParquetReader` initialization * Revisited `HoodieSparkRecord` to accept either `UnsafeRow` or `HoodieInternalRow`, and avoid unnecessary copying after unsafe-projection * Cleaning up redundant exception spec * Make sure `updateMetadataFields` properly wraps `InternalRow` into `HoodieInternalRow` if necessary; Cleaned up `MetadataValues` * Fixed meta-fields extraction and `HoodieInternalRow` composition w/in `HoodieSparkRecord` * De-duplicate `HoodieSparkRecord` ctors; Make sure either only `UnsafeRow` or `HoodieInternalRow` are permitted inside `HoodieSparkRecord` * Removed unnecessary copying * Cleaned up projection for `HoodieSparkRecord` (dropping partition columns); Removed unnecessary copying * Fixing compilation * Fixing compilation (for Flink) * Cleaned up File Raders' interfaces: - Extracted `HoodieSeekingFileReader` interface (for key-ranged reads) - Pushed down concrete implementation methods into `HoodieAvroFileReaderBase` from the interfaces * Cleaned up File Readers impls (inline with then new interfaces) * Rebsaed `HoodieBackedTableMetadata` onto new `HoodieSeekingFileReader` * Tidying up * Missing licenses * Re-instate custom override for `HoodieAvroParquetReader`; Tidying up * Fixed missing cloning w/in `HoodieLazyInsertIterable` * Fixed missing cloning in deduplication flow * Allow `HoodieSparkRecord` to hold `ColumnarBatchRow` * Missing licenses * Fixing compilation * Missing changes * Fixed Spark 2.x validation whether the row was read as a batch Fix comment in RFC46 (apache#6745) * rename * add MetadataValues in updateMetadataValues * remove singleton in fileFactory * add truncateRecordKey * remove hoodieRecord#setData * rename HoodieAvroRecord * fix code style * fix HoodieSparkRecordSerializer * fix benchmark * fix SparkRecordUtils * instantiate HoodieWriteConfig on the fly * add test * fix HoodieSparkRecordSerializer. Replace Java's object serialization with kryo * add broadcast * fix comment * remove unnecessary broadcast * add unsafe check in spark record * fix getRecordColumnValues * remove spark.sql.parquet.writeLegacyFormat * fix unsafe projection * fix * pass external schema * update doc * rename back to HoodieAvroRecord * fix * remove comparable wrapper * fix comment * fix comment * fix comment * fix comment * simplify row copy * fix ParquetReaderIterator Co-authored-by: Shawy Geng <[email protected]> Co-authored-by: wangzixuan.wzxuan <[email protected]> [RFC-46][HUDI-4414] Update the RFC-46 doc to fix comments feedback (apache#6132) * Update the RFC-46 doc to fix comments feedback * fix Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4301] [HUDI-3384][HUDI-3385] Spark specific file reader/writer.(apache#5629) * [HUDI-4301] [HUDI-3384][HUDI-3385] Spark specific file reader/writer. * add schema finger print * add benchmark * a new way to config the merger * fix Co-authored-by: wangzixuan.wzxuan <[email protected]> Co-authored-by: gengxiaoyu <[email protected]> [HUDI-3350][HUDI-3351] Support HoodieMerge API and Spark engine-specific HoodieRecord (apache#5627) Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4344] fix usage of HoodieDataBlock#getRecordIterator (apache#6005) Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4292][RFC-46] Update doc to align with the Record Merge API changes (apache#5927) [MINOR] Fix type casting in TestHoodieHFileReaderWriter [HUDI-3378][HUDI-3379][HUDI-3381] Migrate usage of HoodieRecordPayload and raw Avro payload to HoodieRecord (apache#5522) Co-authored-by: Alexey Kudinkin <[email protected]> Co-authored-by: wangzixuan.wzxuan <[email protected]>
xushiyan
pushed a commit
that referenced
this pull request
Dec 14, 2022
## Change Logs - Add HoodieSparkValidateDuplicateKeyRecordMerger behaving the same as ValidateDuplicateKeyPayload. We should use it with config "hoodie.sql.insert.mode=strict". - Fix nest field exist in HoodieCatalystExpressionUtils - Fix rewrite in HoodieInternalRowUtiles to support type promoted as avro - Fallback to avro when use "merge into" sql - Fix some schema handling issue - Support delta streamer - Convert parquet schema to spark schema and then avro schema(in org.apache.hudi.io.storage.HoodieSparkParquetReader#getSchema). Some types in avro are not compatible with parquet. For ex, decimal as int32/int64 in parquet will convert to int/long in avro. Because avro do not has decimal as int/long . We will lose the logic type info if we directly convert it to avro schema. - Support schema evolution in parquet block [Minor] fix multi deser avro payload (#7021) In HoodieAvroRecord, we will call isDelete, shouldIgnore before we write it to the file. Each method will deserialize HoodiePayload. So we add deserialization method in HoodieRecord and call this method once before calling isDelete or shouldIgnore. Co-authored-by: wangzixuan.wzxuan <[email protected]> Co-authored-by: Alexey Kudinkin <[email protected]> Co-authored-by: Alexey Kudinkin <[email protected]> [MINOR] Properly registering target classes w/ Kryo (#7026) * Added `HoodieKryoRegistrar` registering necessary Hudi's classes w/ Kryo to make their serialization more efficient (by serializing just the class id, in-liue the fully qualified class-name) [MINOR] Make sure all `HoodieRecord`s are appropriately serializable by Kryo (#6977) * Make sure `HoodieRecord`, `HoodieKey`, `HoodieRecordLocation` are all `KryoSerializable` * Revisited `HoodieRecord` serialization hooks to make sure they a) could not be overridden, b) provide for hooks to properly serialize record's payload; Implemented serialization hooks for `HoodieAvroIndexedRecord`; Implemented serialization hooks for `HoodieEmptyRecord`; Implemented serialization hooks for `HoodieAvroRecord`; * Revisited `HoodieSparkRecord` to transiently hold on to the schema so that it could project row [MINOR] Additional fixes for #6745 (#6947) Co-authored-by: Shawy Geng <[email protected]> Co-authored-by: wangzixuan.wzxuan <[email protected]> [RFC-46][HUDI-4414] Update the RFC-46 doc to fix comments feedback (#6132) Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4301] [HUDI-3384][HUDI-3385] Spark specific file reader/writer.(#5629) Co-authored-by: wangzixuan.wzxuan <[email protected]> Co-authored-by: gengxiaoyu <[email protected]> [HUDI-3350][HUDI-3351] Support HoodieMerge API and Spark engine-specific HoodieRecord (#5627) Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4344] fix usage of HoodieDataBlock#getRecordIterator (#6005) Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4292][RFC-46] Update doc to align with the Record Merge API changes (#5927) [MINOR] Fix type casting in TestHoodieHFileReaderWriter [HUDI-3378][HUDI-3379][HUDI-3381] Migrate usage of HoodieRecordPayload and raw Avro payload to HoodieRecord (#5522) Co-authored-by: Alexey Kudinkin <[email protected]> Co-authored-by: wangzixuan.wzxuan <[email protected]> Co-authored-by: gengxiaoyu <[email protected]>
fengjian428
pushed a commit
to fengjian428/hudi
that referenced
this pull request
Apr 5, 2023
## Change Logs - Add HoodieSparkValidateDuplicateKeyRecordMerger behaving the same as ValidateDuplicateKeyPayload. We should use it with config "hoodie.sql.insert.mode=strict". - Fix nest field exist in HoodieCatalystExpressionUtils - Fix rewrite in HoodieInternalRowUtiles to support type promoted as avro - Fallback to avro when use "merge into" sql - Fix some schema handling issue - Support delta streamer - Convert parquet schema to spark schema and then avro schema(in org.apache.hudi.io.storage.HoodieSparkParquetReader#getSchema). Some types in avro are not compatible with parquet. For ex, decimal as int32/int64 in parquet will convert to int/long in avro. Because avro do not has decimal as int/long . We will lose the logic type info if we directly convert it to avro schema. - Support schema evolution in parquet block [Minor] fix multi deser avro payload (apache#7021) In HoodieAvroRecord, we will call isDelete, shouldIgnore before we write it to the file. Each method will deserialize HoodiePayload. So we add deserialization method in HoodieRecord and call this method once before calling isDelete or shouldIgnore. Co-authored-by: wangzixuan.wzxuan <[email protected]> Co-authored-by: Alexey Kudinkin <[email protected]> Co-authored-by: Alexey Kudinkin <[email protected]> [MINOR] Properly registering target classes w/ Kryo (apache#7026) * Added `HoodieKryoRegistrar` registering necessary Hudi's classes w/ Kryo to make their serialization more efficient (by serializing just the class id, in-liue the fully qualified class-name) [MINOR] Make sure all `HoodieRecord`s are appropriately serializable by Kryo (apache#6977) * Make sure `HoodieRecord`, `HoodieKey`, `HoodieRecordLocation` are all `KryoSerializable` * Revisited `HoodieRecord` serialization hooks to make sure they a) could not be overridden, b) provide for hooks to properly serialize record's payload; Implemented serialization hooks for `HoodieAvroIndexedRecord`; Implemented serialization hooks for `HoodieEmptyRecord`; Implemented serialization hooks for `HoodieAvroRecord`; * Revisited `HoodieSparkRecord` to transiently hold on to the schema so that it could project row [MINOR] Additional fixes for apache#6745 (apache#6947) Co-authored-by: Shawy Geng <[email protected]> Co-authored-by: wangzixuan.wzxuan <[email protected]> [RFC-46][HUDI-4414] Update the RFC-46 doc to fix comments feedback (apache#6132) Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4301] [HUDI-3384][HUDI-3385] Spark specific file reader/writer.(apache#5629) Co-authored-by: wangzixuan.wzxuan <[email protected]> Co-authored-by: gengxiaoyu <[email protected]> [HUDI-3350][HUDI-3351] Support HoodieMerge API and Spark engine-specific HoodieRecord (apache#5627) Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4344] fix usage of HoodieDataBlock#getRecordIterator (apache#6005) Co-authored-by: wangzixuan.wzxuan <[email protected]> [HUDI-4292][RFC-46] Update doc to align with the Record Merge API changes (apache#5927) [MINOR] Fix type casting in TestHoodieHFileReaderWriter [HUDI-3378][HUDI-3379][HUDI-3381] Migrate usage of HoodieRecordPayload and raw Avro payload to HoodieRecord (apache#5522) Co-authored-by: Alexey Kudinkin <[email protected]> Co-authored-by: wangzixuan.wzxuan <[email protected]> Co-authored-by: gengxiaoyu <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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.
Change Logs
This PR adjusts the way new and existing
HoodieRecords implementations are being serialized by Kryo t/h explicit implementation ofKryoSerializableinterfaces forHodieRecord(and inheritors)HoodieKeyHoodieRecordLocationAdditionally,
HoodieSparkRecordserialization is streamlined to always project theRowintoUnsafeRowbefore serialization to avoid penalty of serializing whole object tree.Impact
No impact
Risk level (write none, low medium or high below)
Risk: medium
Documentation Update
N/A
Contributor's checklist