hongzhi-gao commented on code in PR #19244: URL: https://github.com/apache/iotdb/pull/19244#discussion_r4561999448
########## iotdb-core/consensus/src/main/java/org/apache/iotdb/consensus/config/IoTConsensusConfig.java: ########## @@ -22,6 +22,8 @@ import org.apache.iotdb.commons.client.property.ClientPoolProperty.DefaultProperty; import org.apache.iotdb.commons.memory.AtomicLongMemoryBlock; import org.apache.iotdb.commons.memory.IMemoryBlock; +import org.apache.iotdb.consensus.i18n.IoTConsensusMessages; +import org.apache.iotdb.consensus.iot.thrift.TLogCompressionType; Review Comment: Addressed in 434292b65c5. Internal configuration now uses TsFile CompressionType, with NONE mapped to UNCOMPRESSED. The wire enum remains separate because its numeric IDs differ from TsFile's; an explicit mapping preserves the RPC contract. Unsupported TsFile codecs are rejected by the consensus configuration builder, with regression coverage. ########## iotdb-core/consensus/src/main/java/org/apache/iotdb/consensus/iot/logdispatcher/LogCompression.java: ########## @@ -0,0 +1,332 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one + * or more contributor license agreements. See the NOTICE file + * distributed with this work for additional information + * regarding copyright ownership. The ASF licenses this file + * to you under the Apache License, Version 2.0 (the + * "License"); you may not use this file except in compliance + * with the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, + * software distributed under the License is distributed on an + * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY + * KIND, either express or implied. See the License for the + * specific language governing permissions and limitations + * under the License. + */ + +package org.apache.iotdb.consensus.iot.logdispatcher; + +import org.apache.iotdb.consensus.i18n.IoTConsensusMessages; +import org.apache.iotdb.consensus.iot.thrift.TLogBufferCompression; +import org.apache.iotdb.consensus.iot.thrift.TLogCompressionType; +import org.apache.iotdb.consensus.iot.thrift.TLogEntry; +import org.apache.iotdb.consensus.iot.thrift.TSyncLogEntriesReq; + +import com.github.luben.zstd.Zstd; +import com.github.luben.zstd.ZstdException; +import org.apache.tsfile.compress.ICompressor; +import org.apache.tsfile.compress.IUnCompressor; +import org.apache.tsfile.file.metadata.enums.CompressionType; + +import java.io.IOException; +import java.nio.ByteBuffer; +import java.util.ArrayList; +import java.util.Arrays; +import java.util.List; +import java.util.concurrent.atomic.AtomicLong; +import java.util.zip.CRC32C; + +/** Wire-only encoding: the original batch and its buffer positions remain available for retries. */ +public final class LogCompression { + // Conservative allowance for the additional Thrift descriptor and list field on the wire. + private static final int DESCRIPTOR_BYTES = 32; + private static final ICompressor LZ4 = ICompressor.getCompressor(CompressionType.LZ4); + private static final ICompressor SNAPPY = ICompressor.getCompressor(CompressionType.SNAPPY); + private static final IUnCompressor LZ4_DECODER = + IUnCompressor.getUnCompressor(CompressionType.LZ4); + private static final IUnCompressor SNAPPY_DECODER = + IUnCompressor.getUnCompressor(CompressionType.SNAPPY); + private static final IUnCompressor ZSTD_DECODER = + IUnCompressor.getUnCompressor(CompressionType.ZSTD); + + private LogCompression() {} + + /** Accounts for arrays throughout the async attempt, including synchronous send failure. */ + public static final class EncodedRequest implements AutoCloseable { + private TSyncLogEntriesReq request; + private final AtomicLong reserved; + private final long rawBytes; + private long encodedBytes; + private boolean memoryLimited; + + private EncodedRequest( + TSyncLogEntriesReq request, long reserved, long rawBytes, boolean memoryLimited) { + this.request = request; + this.reserved = new AtomicLong(reserved); + this.rawBytes = rawBytes; + this.encodedBytes = rawBytes; + this.memoryLimited = memoryLimited; + } + + public long getRawBytes() { + return rawBytes; + } + + public long getEncodedBytes() { + return encodedBytes; + } + + public boolean isMemoryLimited() { + return memoryLimited; + } + + public TSyncLogEntriesReq getRequest() { + return request; + } + + private void retain(long bytes) { + long released = reserved.getAndSet(bytes) - bytes; + if (released > 0) IoTConsensusMemoryManager.getInstance().freeLogCompression(released); Review Comment: Addressed in 31f35e67b48: expanded the control-flow formatting, split declarations, and used clearer variable names. Scoped Spotless checks pass. ########## iotdb-core/consensus/src/main/java/org/apache/iotdb/consensus/iot/logdispatcher/LogCompression.java: ########## @@ -0,0 +1,332 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one + * or more contributor license agreements. See the NOTICE file + * distributed with this work for additional information + * regarding copyright ownership. The ASF licenses this file + * to you under the Apache License, Version 2.0 (the + * "License"); you may not use this file except in compliance + * with the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, + * software distributed under the License is distributed on an + * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY + * KIND, either express or implied. See the License for the + * specific language governing permissions and limitations + * under the License. + */ + +package org.apache.iotdb.consensus.iot.logdispatcher; + +import org.apache.iotdb.consensus.i18n.IoTConsensusMessages; +import org.apache.iotdb.consensus.iot.thrift.TLogBufferCompression; +import org.apache.iotdb.consensus.iot.thrift.TLogCompressionType; +import org.apache.iotdb.consensus.iot.thrift.TLogEntry; +import org.apache.iotdb.consensus.iot.thrift.TSyncLogEntriesReq; + +import com.github.luben.zstd.Zstd; +import com.github.luben.zstd.ZstdException; +import org.apache.tsfile.compress.ICompressor; +import org.apache.tsfile.compress.IUnCompressor; +import org.apache.tsfile.file.metadata.enums.CompressionType; + +import java.io.IOException; +import java.nio.ByteBuffer; +import java.util.ArrayList; +import java.util.Arrays; +import java.util.List; +import java.util.concurrent.atomic.AtomicLong; +import java.util.zip.CRC32C; + +/** Wire-only encoding: the original batch and its buffer positions remain available for retries. */ +public final class LogCompression { + // Conservative allowance for the additional Thrift descriptor and list field on the wire. + private static final int DESCRIPTOR_BYTES = 32; + private static final ICompressor LZ4 = ICompressor.getCompressor(CompressionType.LZ4); + private static final ICompressor SNAPPY = ICompressor.getCompressor(CompressionType.SNAPPY); + private static final IUnCompressor LZ4_DECODER = + IUnCompressor.getUnCompressor(CompressionType.LZ4); + private static final IUnCompressor SNAPPY_DECODER = + IUnCompressor.getUnCompressor(CompressionType.SNAPPY); + private static final IUnCompressor ZSTD_DECODER = + IUnCompressor.getUnCompressor(CompressionType.ZSTD); + + private LogCompression() {} + + /** Accounts for arrays throughout the async attempt, including synchronous send failure. */ + public static final class EncodedRequest implements AutoCloseable { + private TSyncLogEntriesReq request; + private final AtomicLong reserved; + private final long rawBytes; + private long encodedBytes; + private boolean memoryLimited; + + private EncodedRequest( + TSyncLogEntriesReq request, long reserved, long rawBytes, boolean memoryLimited) { + this.request = request; + this.reserved = new AtomicLong(reserved); + this.rawBytes = rawBytes; + this.encodedBytes = rawBytes; + this.memoryLimited = memoryLimited; + } + + public long getRawBytes() { + return rawBytes; + } + + public long getEncodedBytes() { + return encodedBytes; + } + + public boolean isMemoryLimited() { + return memoryLimited; + } + + public TSyncLogEntriesReq getRequest() { + return request; + } + + private void retain(long bytes) { + long released = reserved.getAndSet(bytes) - bytes; + if (released > 0) IoTConsensusMemoryManager.getInstance().freeLogCompression(released); + } + + @Override + public void close() { + long bytes = reserved.getAndSet(0); + if (bytes > 0) IoTConsensusMemoryManager.getInstance().freeLogCompression(bytes); + } + } + + public static EncodedRequest compress( + TSyncLogEntriesReq original, TLogCompressionType codec, int minSize, int maxDecodedBytes) + throws IOException { + long buffers = 0, rawSize = 0; + boolean hasEligibleBuffer = false; + for (TLogEntry entry : original.getLogEntries()) { + for (ByteBuffer buffer : entry.getData()) { + int size = buffer.remaining(); + rawSize += size; + if (size >= minSize && size > 0) { + hasEligibleBuffer = true; + } + buffers++; + } + } + // Preserve the original path for oversized legal entries and when scratch memory is scarce. + if (codec == TLogCompressionType.NONE || !hasEligibleBuffer || rawSize > maxDecodedBytes) + return new EncodedRequest(original, 0, rawSize, false); + // Charge descriptor/list/entry overhead first, then one block's working arrays at a time. + // Retain only actual compact outputs between blocks, not a whole-batch worst-case estimate. + long budget = buffers * 256 + original.getLogEntriesSize() * 128L; + if (!IoTConsensusMemoryManager.getInstance().reserveLogCompression(budget)) { + return new EncodedRequest(original, 0, rawSize, true); + } + EncodedRequest result = new EncodedRequest(original, budget, rawSize, false); + try { + List<TLogEntry> entries = new ArrayList<>(original.getLogEntriesSize()); + long retained = 0; + boolean changed = false; + for (TLogEntry entry : original.getLogEntries()) { + List<ByteBuffer> data = new ArrayList<>(entry.getDataSize()); + List<TLogBufferCompression> info = new ArrayList<>(entry.getDataSize()); + long raw = 0, wire = 0, arrays = 0; + boolean compressed = false; + for (ByteBuffer buffer : entry.getData()) { + int size = buffer.remaining(); + raw += size; + ByteBuffer encoded = buffer; + TLogCompressionType actualCodec = TLogCompressionType.NONE; + if (size >= minSize && size > 0) { + // Includes an optional input copy, codec-bound output and compact output, with + // allowance for array headers and codec bounds on very small buffers. + long workspace = 4L * size + 256; + long before = result.reserved.get(); + if (IoTConsensusMemoryManager.getInstance().reserveLogCompression(workspace)) { + result.reserved.addAndGet(workspace); + byte[] input = buffer.hasArray() ? buffer.array() : bytes(buffer); + int offset = buffer.hasArray() ? buffer.arrayOffset() + buffer.position() : 0; + int bound = + codec == TLogCompressionType.ZSTD + ? (int) Zstd.compressBound(size) + : compressor(codec).getMaxBytesForCompression(size); Review Comment: The special path originally kept Zstd at level 1 while the available TsFile API used its own default level. Removed that special handling in 434292b65c5: ZSTD now uses TsFile ICompressor/IUnCompressor and its default level, just like the other codecs. The pinned TsFile default is the maximum Zstd level. I am rerunning native cluster benchmarks; the existing Zstd-1 numbers are explicitly marked as historical and do not represent the new default. ########## iotdb-core/consensus/src/main/java/org/apache/iotdb/consensus/iot/logdispatcher/LogCompression.java: ########## @@ -0,0 +1,332 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one + * or more contributor license agreements. See the NOTICE file + * distributed with this work for additional information + * regarding copyright ownership. The ASF licenses this file + * to you under the Apache License, Version 2.0 (the + * "License"); you may not use this file except in compliance + * with the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, + * software distributed under the License is distributed on an + * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY + * KIND, either express or implied. See the License for the + * specific language governing permissions and limitations + * under the License. + */ + +package org.apache.iotdb.consensus.iot.logdispatcher; + +import org.apache.iotdb.consensus.i18n.IoTConsensusMessages; +import org.apache.iotdb.consensus.iot.thrift.TLogBufferCompression; +import org.apache.iotdb.consensus.iot.thrift.TLogCompressionType; +import org.apache.iotdb.consensus.iot.thrift.TLogEntry; +import org.apache.iotdb.consensus.iot.thrift.TSyncLogEntriesReq; + +import com.github.luben.zstd.Zstd; +import com.github.luben.zstd.ZstdException; +import org.apache.tsfile.compress.ICompressor; +import org.apache.tsfile.compress.IUnCompressor; +import org.apache.tsfile.file.metadata.enums.CompressionType; + +import java.io.IOException; +import java.nio.ByteBuffer; +import java.util.ArrayList; +import java.util.Arrays; +import java.util.List; +import java.util.concurrent.atomic.AtomicLong; +import java.util.zip.CRC32C; + +/** Wire-only encoding: the original batch and its buffer positions remain available for retries. */ +public final class LogCompression { + // Conservative allowance for the additional Thrift descriptor and list field on the wire. + private static final int DESCRIPTOR_BYTES = 32; + private static final ICompressor LZ4 = ICompressor.getCompressor(CompressionType.LZ4); + private static final ICompressor SNAPPY = ICompressor.getCompressor(CompressionType.SNAPPY); + private static final IUnCompressor LZ4_DECODER = + IUnCompressor.getUnCompressor(CompressionType.LZ4); + private static final IUnCompressor SNAPPY_DECODER = + IUnCompressor.getUnCompressor(CompressionType.SNAPPY); + private static final IUnCompressor ZSTD_DECODER = + IUnCompressor.getUnCompressor(CompressionType.ZSTD); + + private LogCompression() {} + + /** Accounts for arrays throughout the async attempt, including synchronous send failure. */ + public static final class EncodedRequest implements AutoCloseable { + private TSyncLogEntriesReq request; + private final AtomicLong reserved; + private final long rawBytes; + private long encodedBytes; + private boolean memoryLimited; + + private EncodedRequest( + TSyncLogEntriesReq request, long reserved, long rawBytes, boolean memoryLimited) { + this.request = request; + this.reserved = new AtomicLong(reserved); + this.rawBytes = rawBytes; + this.encodedBytes = rawBytes; + this.memoryLimited = memoryLimited; + } + + public long getRawBytes() { + return rawBytes; + } + + public long getEncodedBytes() { + return encodedBytes; + } + + public boolean isMemoryLimited() { + return memoryLimited; + } + + public TSyncLogEntriesReq getRequest() { + return request; + } + + private void retain(long bytes) { + long released = reserved.getAndSet(bytes) - bytes; + if (released > 0) IoTConsensusMemoryManager.getInstance().freeLogCompression(released); + } + + @Override + public void close() { + long bytes = reserved.getAndSet(0); + if (bytes > 0) IoTConsensusMemoryManager.getInstance().freeLogCompression(bytes); + } + } + + public static EncodedRequest compress( + TSyncLogEntriesReq original, TLogCompressionType codec, int minSize, int maxDecodedBytes) + throws IOException { + long buffers = 0, rawSize = 0; + boolean hasEligibleBuffer = false; + for (TLogEntry entry : original.getLogEntries()) { + for (ByteBuffer buffer : entry.getData()) { + int size = buffer.remaining(); + rawSize += size; + if (size >= minSize && size > 0) { + hasEligibleBuffer = true; + } + buffers++; + } + } + // Preserve the original path for oversized legal entries and when scratch memory is scarce. + if (codec == TLogCompressionType.NONE || !hasEligibleBuffer || rawSize > maxDecodedBytes) + return new EncodedRequest(original, 0, rawSize, false); + // Charge descriptor/list/entry overhead first, then one block's working arrays at a time. + // Retain only actual compact outputs between blocks, not a whole-batch worst-case estimate. + long budget = buffers * 256 + original.getLogEntriesSize() * 128L; + if (!IoTConsensusMemoryManager.getInstance().reserveLogCompression(budget)) { + return new EncodedRequest(original, 0, rawSize, true); + } + EncodedRequest result = new EncodedRequest(original, budget, rawSize, false); + try { + List<TLogEntry> entries = new ArrayList<>(original.getLogEntriesSize()); + long retained = 0; + boolean changed = false; + for (TLogEntry entry : original.getLogEntries()) { + List<ByteBuffer> data = new ArrayList<>(entry.getDataSize()); + List<TLogBufferCompression> info = new ArrayList<>(entry.getDataSize()); + long raw = 0, wire = 0, arrays = 0; + boolean compressed = false; + for (ByteBuffer buffer : entry.getData()) { + int size = buffer.remaining(); + raw += size; + ByteBuffer encoded = buffer; + TLogCompressionType actualCodec = TLogCompressionType.NONE; + if (size >= minSize && size > 0) { + // Includes an optional input copy, codec-bound output and compact output, with + // allowance for array headers and codec bounds on very small buffers. + long workspace = 4L * size + 256; + long before = result.reserved.get(); + if (IoTConsensusMemoryManager.getInstance().reserveLogCompression(workspace)) { + result.reserved.addAndGet(workspace); + byte[] input = buffer.hasArray() ? buffer.array() : bytes(buffer); + int offset = buffer.hasArray() ? buffer.arrayOffset() + buffer.position() : 0; + int bound = + codec == TLogCompressionType.ZSTD + ? (int) Zstd.compressBound(size) + : compressor(codec).getMaxBytesForCompression(size); + byte[] output = new byte[bound]; + int length = + codec == TLogCompressionType.ZSTD + ? (int) + Zstd.compressByteArray(output, 0, output.length, input, offset, size, 1) + : compressor(codec).compress(input, offset, size, output); + if (length < 0) throw invalid(Failure.CODEC, size, length); + if ((long) length + DESCRIPTOR_BYTES < size) { Review Comment: Renamed the variables to compressedLength/uncompressedLength in 31f35e67b48. The existing internal 4 KiB cutoff is a pre-compression eligibility check, separate from the post-compression benefit rule. Currently a buffer must save space including its descriptor, and the entry must also save space including all descriptors; a fixed 5% minimum-savings rule has not been added. I am rerunning performance tests on the current implementation and will inspect near-break-even payloads before choosing an additional benefit cutoff. Leaving this part open rather than claiming the size cutoff addresses decoding cost. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
