GH-3530: Bypass Hadoop codec abstraction to optimize compression performance#3570
Open
iemejia wants to merge 14 commits into
Open
GH-3530: Bypass Hadoop codec abstraction to optimize compression performance#3570iemejia wants to merge 14 commits into
iemejia wants to merge 14 commits into
Conversation
iemejia
force-pushed
the
parquet-perf-v2-par6-compression
branch
2 times, most recently
from
fe836a6 to
0571b9e
Compare
…n performance Bypass Hadoop CompressionCodec for Snappy (xerial JNI), LZ4_RAW (airlift), ZSTD (zstd-jni), GZIP (JDK), LZO (airlift), and BROTLI (brotli4j) in both CodecFactory and DirectCodecFactory, eliminating per-page codec pool lookups, stream wrapper allocation, and unnecessary buffer copies. SNAPPY: direct byte-array JNI calls via Snappy.compress/uncompress, avoiding the Hadoop stream abstraction and intermediate direct ByteBuffer copies. ZSTD: replace streaming ZstdOutputStreamNoFinalizer/ZstdInputStreamNoFinalizer with reusable ZstdCompressCtx/ZstdDecompressCtx single-call APIs. Supports compression level and multi-threaded workers configuration. LZ4_RAW: direct airlift Lz4Compressor/Lz4Decompressor with reusable direct ByteBuffers, bypassing Hadoop's NonBlockedCompressor overhead. GZIP: bypass Hadoop's GzipCodec and codec-pool/stream-wrapper overhead with direct JDK GZIPOutputStream/GZIPInputStream. Compression level is read from the existing "zlib.compress.level" Hadoop configuration key. LZO: use aircompressor's LzoHadoopStreams directly, bypassing the GPL-licensed com.hadoop.compression.lzo.LzoCodec. Wire-compatible with Hadoop's LzoCodec. BROTLI: migrate from jbrotli (unmaintained) to brotli4j via reflection, using single-call Encoder.compress/Decoder.decompress byte-array APIs. End-to-end interop tests (TestCompressionInterop) validate that files written with the old Hadoop CompressionCodec path are readable by the new direct path and vice versa, for all 6 codecs including multi-row-group scenarios.
Remove the legacy Hadoop CompressionCodec-based compression/decompression path from CodecFactory and DirectCodecFactory. All codecs now use only the optimized direct implementations (snappy-java, zstd-jni, airlift LZ4/LZO, JDK GZIP, brotli4j) introduced in the previous commit. Removed from CodecFactory: - HeapBytesCompressor/HeapBytesDecompressor (Hadoop stream wrappers) - getCodec() method (Hadoop codec loading via reflection) - CODEC_BY_NAME static cache and cacheKey() helper - Hadoop codec fallback in createCompressor()/createDecompressor() Removed from DirectCodecFactory: - IndirectDecompressor (Hadoop Decompressor wrapper) - FullDirectDecompressor (Hadoop DirectDecompressionCodec via reflection) - DirectCodecPool (pooled Hadoop compressor/decompressor instances) - Static reflection fields for DirectDecompressionCodec discovery Unsupported codecs and BROTLI without brotli4j now throw UnsupportedOperationException instead of falling through to Hadoop. Also removes the commons-pool:commons-pool dependency which was only used by DirectCodecPool.
Delete the Hadoop CompressionCodec implementations that are no longer used now that CodecFactory exclusively uses direct library calls: Deleted codec classes: - SnappyCodec, SnappyCompressor, SnappyDecompressor - ZstandardCodec, ZstdCompressorStream, ZstdDecompressorStream - Lz4RawCodec, Lz4RawCompressor, Lz4RawDecompressor - NonBlockedCompressor, NonBlockedDecompressor - NonBlockedCompressorStream, NonBlockedDecompressorStream Deleted tests that exercised the Hadoop codec classes directly: - TestCompressionCodec, TestSnappyCodec, TestZstandardCodec Move ZSTD configuration constants (PARQUET_COMPRESS_ZSTD_LEVEL, PARQUET_COMPRESS_ZSTD_WORKERS) from the deleted ZstandardCodec into CodecFactory where they are now consumed. Update TestDirectCodecFactory and TestDataPageChecksums to use CodecFactory constants and direct Snappy JNI respectively.
Create a new parquet-compression module that contains all compression codec implementations (Snappy, ZSTD, LZ4, GZIP, LZO, Brotli) without any Hadoop dependencies. This allows non-Hadoop users to get compression support without pulling in the Hadoop classpath. New module: parquet-compression - DefaultCompressionCodecFactory: full compression/decompression logic - Dependencies: parquet-common, snappy-java, zstd-jni, aircompressor, brotli4j (optional/runtime) - Zero Hadoop dependencies parquet-hadoop changes: - CodecFactory now extends DefaultCompressionCodecFactory, providing only Hadoop Configuration-based constructors for backward compat - DirectCodecFactory unchanged (still extends CodecFactory) - Compression library dependencies moved to parquet-compression - Added parquet-compression as a dependency
Add TestDefaultCompressionCodecFactory with 12 tests covering: - Round-trip compress/decompress for all 6 codecs (SNAPPY, GZIP, ZSTD, LZ4_RAW, LZO, BROTLI) - Multiple data sizes (0 bytes to 256KB) - ZSTD compression level configuration - ZSTD multi-threaded workers configuration - GZIP compression level configuration - Compressor reuse safety (consecutive compressions) - UNCOMPRESSED pass-through These tests run without Hadoop, validating the compression module operates independently.
…asses Split the monolithic 832-line DefaultCompressionCodecFactory into focused single-responsibility classes with minimum visibility: - DefaultCompressionCodecFactory (public): factory logic, caching, constants, BytesCompressor/BytesDecompressor abstract classes - SnappyCodec (package-private): Snappy JNI compress/decompress - ZstdCodec (package-private): zstd-jni context-based compress/decompress - Lz4RawCodec (package-private): airlift LZ4 with direct buffers - GzipCodec (package-private): JDK GZIPOutputStream/GZIPInputStream - LzoCodec (package-private): airlift LzoHadoopStreams - BrotliCodec (package-private): delegates to Brotli4j helper - Brotli4j (public): reflection-based brotli4j access, public because DirectCodecFactory in parquet-hadoop also needs it Each codec class is final with a private constructor (utility pattern). Inner Compressor/Decompressor classes are package-private static final.
…on helper Promote brotli4j from runtime/optional to a compile dependency and remove the Brotli4j reflection helper class entirely. BrotliCodec and DirectCodecFactory now use brotli4j's Encoder/Decoder APIs directly. Changes: - parquet-compression pom: brotli4j is now a regular compile dependency - Deleted Brotli4j.java (reflection-based helper) - Rewrote BrotliCodec to use Encoder.compress/Decoder.decompress directly - Rewrote DirectCodecFactory Brotli classes to use brotli4j directly - Removed Brotli4j.AVAILABLE guards (brotli4j is always on classpath) - Removed unused slf4j-api dependency from parquet-compression - Removed redundant brotli4j dependency from parquet-hadoop pom
snappy-java, aircompressor, and zstd-jni are now provided transitively via the parquet-compression dependency.
Move DirectCodecFactory from parquet-hadoop to parquet-compression, eliminating the last direct compression library imports from parquet-hadoop. DirectCodecFactory now takes ParquetConfiguration instead of Hadoop Configuration. CodecFactory in parquet-hadoop becomes a minimal deprecated shim that provides only Hadoop Configuration-based constructors. parquet-hadoop's dependency on parquet-compression is now purely for the deprecated CodecFactory shim's inheritance. All compression logic lives entirely in parquet-compression with zero Hadoop dependencies.
iemejia
force-pushed
the
parquet-perf-v2-par6-compression
branch
from
0571b9e to
5de3272
Compare
Apply palantir-java-format rules to DirectCodecFactory and TestDefaultCompressionCodecFactory: join short lines that fit within the column limit, and break method chains and long argument lists according to the project style. Assisted-by: GitHub Copilot:claude-opus-4.6
iemejia
force-pushed
the
parquet-perf-v2-par6-compression
branch
2 times, most recently
from
c493dd6 to
402f4c5
Compare
The japicmp plugin fails because: 1. commons-pool classes (BasePoolableObjectFactory, GenericObjectPool) are no longer transitively available after removing the Hadoop codec pool path 2. Public codec classes were moved from parquet-hadoop to parquet-compression 3. CodecFactory was refactored to extend DefaultCompressionCodecFactory 4. Deprecated constructors in ParquetRecordWriter and ColumnChunkPageWriteStore changed their BytesCompressor parameter type Add ignoreMissingClassesByRegularExpression for org.apache.commons.pool and exclude all moved/refactored classes from the compatibility check.
iemejia
force-pushed
the
parquet-perf-v2-par6-compression
branch
from
402f4c5 to
537ce63
Compare
TestDataPageChecksums directly uses org.xerial.snappy.Snappy for computing expected compressed checksums. After extracting compression code to parquet-compression, snappy-java is only transitively available but the dependency plugin requires it to be explicitly declared.
CodecFactory.createDirectCodecFactory() now returns DefaultCompressionCodecFactory, not CodecFactory. Update the benchmark to use the parent type directly.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Part of #3530 — Apache Parquet Java Performance Improvements
Summary
Bypass the Hadoop
CompressionCodecabstraction for all six supported codecs, eliminating per-page codec-pool lookups, stream-wrapper allocation, and unnecessary buffer copies in bothCodecFactoryandDirectCodecFactory.SnappyCodecstream wrappersSnappy.compress/uncompressdirect callsLZ4Compressor/LZ4DecompressordirectZstdOutputStreamNoFinalizer/ZstdInputStreamNoFinalizerZstdCompressCtx/ZstdDecompressCtxsingle-call APIsGzipCodecwith codec-pool overheadGZIPOutputStream/GZIPInputStreamdirectcom.hadoop.compression.lzo.LzoCodecLzoHadoopStreams(Apache 2.0, wire-compatible)brotli-codec(jbrotli, 2016, x86-only)brotli4j1.23.0 (10 platforms incl. aarch64, reflection-loaded)Notable side effects:
JMH benchmarks:
CompressionBenchmark,CpuReadBenchmark,CpuWriteBenchmark,FileReadBenchmark,FileWriteBenchmark,ConcurrentReadWriteBenchmark.Benchmark results
Environment: JDK 25.0.3 (Temurin), OpenJDK 64-Bit Server VM, JMH 1.37, Linux x86_64.
End-to-end file write (100K rows, SingleShotTime, ms/op lower is better):
End-to-end file read (ms/op lower is better):
Raw codec throughput (
DirectCodecFactory): Snappy/ZSTD/LZ4/GZIP unchanged (already had native access). Brotli decompression improved 2.3-2.7x (brotli4j >> jbrotli).V2 shows consistently larger speedups than V1 because V2 encoding produces more, smaller pages, meaning more codec invocations per file where the per-invocation Hadoop overhead accumulates.