rahil-c commented on code in PR #17632: URL: https://github.com/apache/hudi/pull/17632#discussion_r2640791764
########## hudi-client/hudi-spark-client/src/main/java/org/apache/hudi/io/storage/HoodieSparkLanceReader.java: ########## @@ -0,0 +1,326 @@ +/* + * 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.hudi.io.storage; + +import com.lancedb.lance.file.LanceFileReader; +import org.apache.arrow.memory.BufferAllocator; +import org.apache.arrow.memory.RootAllocator; +import org.apache.arrow.vector.VectorSchemaRoot; +import org.apache.arrow.vector.ipc.ArrowReader; +import org.apache.hudi.HoodieSchemaConversionUtils; +import org.apache.hudi.common.bloom.BloomFilter; +import org.apache.hudi.common.model.HoodieRecord; +import org.apache.hudi.common.model.HoodieSparkRecord; +import org.apache.hudi.common.schema.HoodieSchema; +import org.apache.hudi.common.schema.HoodieSchemaUtils; +import org.apache.hudi.common.util.collection.ClosableIterator; +import org.apache.hudi.common.util.collection.CloseableMappingIterator; +import org.apache.hudi.common.util.collection.Pair; +import org.apache.hudi.exception.HoodieException; +import org.apache.hudi.exception.HoodieIOException; +import org.apache.hudi.storage.HoodieStorage; +import org.apache.hudi.storage.StoragePath; +import org.apache.spark.sql.catalyst.InternalRow; +import org.apache.spark.sql.catalyst.expressions.UnsafeProjection; +import org.apache.spark.sql.catalyst.expressions.UnsafeRow; +import org.apache.spark.sql.types.StructField; +import org.apache.spark.sql.types.StructType; +import org.apache.spark.sql.util.LanceArrowUtils; +import org.apache.spark.sql.vectorized.ColumnVector; +import org.apache.spark.sql.vectorized.ColumnarBatch; +import org.apache.spark.sql.vectorized.LanceArrowColumnVector; + +import java.io.IOException; +import java.util.ArrayList; +import java.util.HashSet; +import java.util.Iterator; +import java.util.List; +import java.util.Set; + +import static org.apache.hudi.common.util.TypeUtils.unsafeCast; + +/** + * {@link HoodieSparkFileReader} implementation for Lance file format. + */ +public class HoodieSparkLanceReader implements HoodieSparkFileReader { + // number of rows to read + private static final int DEFAULT_BATCH_SIZE = 512; + private final StoragePath path; + private final HoodieStorage storage; + + public HoodieSparkLanceReader(HoodieStorage storage, StoragePath path) { + this.path = path; + this.storage = storage; + } + + @Override + public String[] readMinMaxRecordKeys() { + throw new UnsupportedOperationException("Min/max record key tracking is not yet supported for Lance file format"); + } + + @Override + public BloomFilter readBloomFilter() { + throw new UnsupportedOperationException("Bloom filter is not yet supported for Lance file format"); + } + + @Override + public Set<Pair<String, Long>> filterRowKeys(Set<String> candidateRowKeys) { + Set<Pair<String, Long>> result = new HashSet<>(); + long position = 0; + + try (ClosableIterator<String> keyIterator = getRecordKeyIterator()) { + while (keyIterator.hasNext()) { + String recordKey = keyIterator.next(); + // If filter is empty/null, then all keys will be added. + // if filter has specific keys, then ensure only those are added + if (candidateRowKeys == null || candidateRowKeys.isEmpty() + || candidateRowKeys.contains(recordKey)) { + result.add(Pair.of(recordKey, position)); + } + position++; + } + } catch (IOException e) { + throw new HoodieIOException("Failed to filter row keys from Lance file: " + path, e); + } + + return result; + } + + @Override + public ClosableIterator<HoodieRecord<InternalRow>> getRecordIterator(HoodieSchema readerSchema, HoodieSchema requestedSchema) throws IOException { + return getRecordIterator(requestedSchema); + } + + @Override + public ClosableIterator<HoodieRecord<InternalRow>> getRecordIterator(HoodieSchema schema) throws IOException { + ClosableIterator<UnsafeRow> iterator = getUnsafeRowIterator(); + return new CloseableMappingIterator<>(iterator, data -> unsafeCast(new HoodieSparkRecord(data))); + } + + @Override + public ClosableIterator<String> getRecordKeyIterator() throws IOException { + //TODO to revisit adding support for when metadata fields are not persisted. + + // Get schema with only the record key field for efficient column pruning + HoodieSchema recordKeySchema = HoodieSchemaUtils.getRecordKeySchema(); + ClosableIterator<UnsafeRow> iterator = getUnsafeRowIterator(recordKeySchema); + + // Map each UnsafeRow to extract the record key string directly from index 0 + // The record key is at index 0 because we're using lance column projection which has only the record key field + return new CloseableMappingIterator<>(iterator, data -> data.getUTF8String(0).toString()); + } + + /** + * Get an iterator over UnsafeRows from the Lance file. + * + * @return ClosableIterator over UnsafeRows + * @throws IOException if reading fails + */ + public ClosableIterator<UnsafeRow> getUnsafeRowIterator() { + BufferAllocator allocator = new RootAllocator(Long.MAX_VALUE); + + try { + LanceFileReader lanceReader = LanceFileReader.open(path.toString(), allocator); + + // Get schema from Lance file and convert to Spark StructType + org.apache.arrow.vector.types.pojo.Schema arrowSchema = lanceReader.schema(); + StructType sparkSchema = LanceArrowUtils.fromArrowSchema(arrowSchema); + + ArrowReader arrowReader = lanceReader.readAll(null, null, DEFAULT_BATCH_SIZE); + + return new LanceRecordIterator(allocator, lanceReader, arrowReader, sparkSchema); + } catch (Exception e) { + allocator.close(); + throw new HoodieException("Failed to create Lance reader for: " + path, e); + } + } + + /** + * Get an iterator over UnsafeRows from the Lance file with column projection. + * This allows reading only specific columns for better performance. + * + * @param requestedSchema Avro schema specifying which columns to read + * @return ClosableIterator over UnsafeRows + * @throws IOException if reading fails + */ + public ClosableIterator<UnsafeRow> getUnsafeRowIterator(HoodieSchema requestedSchema) { + // Convert HoodieSchema to Spark StructType + StructType requestedSparkSchema = HoodieSchemaConversionUtils.convertHoodieSchemaToStructType(requestedSchema); + + BufferAllocator allocator = new RootAllocator(Long.MAX_VALUE); + + try { + LanceFileReader lanceReader = LanceFileReader.open(path.toString(), allocator); + + // Build list of column names to project (read only requested columns) + List<String> columnNames = new ArrayList<>(); + for (StructField field : requestedSparkSchema.fields()) { + columnNames.add(field.name()); + } + + // Read only the requested columns from Lance file for efficiency + ArrowReader arrowReader = lanceReader.readAll(columnNames, null, DEFAULT_BATCH_SIZE); + + return new LanceRecordIterator(allocator, lanceReader, arrowReader, requestedSparkSchema); + } catch (Exception e) { + allocator.close(); + throw new HoodieException("Failed to create Lance reader for: " + path, e); + } + } + + @Override + public HoodieSchema getSchema() { + // Read Arrow schema from Lance file and convert to Avro + try (BufferAllocator allocator = new RootAllocator(Long.MAX_VALUE); Review Comment: @the-other-tim-brown @voonhous Just want to share my mental model here so we are all aligned. ```BufferAllocator allocator = new RootAllocator(Long.MAX_VALUE)``` Correct that the following is not a lance class and is native to arrow. However this class is heavily used in Lance readers and writers as a core dependecy. My intention of sharing the screenshot or code pointers from lance is just to point how their community is leveraging this specific arrow class in their implementation (my assumption was that they have a would good sense of how best to configure `BufferAllocator`) and that we can follow a similar pattern since we are essentially leveraging lance's readers/writers for integrating with their format. The Lance's spark `Runtime` class is a wrapper around the underlying arrow `BufferAllocator` class: https://github.com/lance-format/lance-spark/blob/main/lance-spark-base_2.12/src/main/java/org/lance/spark/LanceRuntime.java#L57 and defines a global `RootAllocator` with a default size of `LONG.MAX_VALUE`, if it does not get any value set from a sys env var. From there when spark attempts to read or write lance files, it will only access that same `RootAllocator` via usages of `LanceRuntime.allocator())`, basically following a singleton pattern to ensure that only memory is access from that same allocator https://github.com/lance-format/lance-spark/blob/main/lance-spark-base_2.12/src/main/java/org/lance/spark/internal/LanceFragmentScanner.java#L54. If these lance examples do not provide enough clarity, from the arrow documentation I think this that pattern that lance community followed seems to match the development guidelines section [Development Guidelines](https://arrow.apache.org/docs/java/memory.html#id10 ) ``` Applications should generally: Use the BufferAllocator interface in APIs instead of RootAllocator. Create one RootAllocator at the start of the program and explicitly pass it when needed. ``` Let me know though is there is something concrete you wanted changed in the code, or a pattern followed. Otherwise I think what @voonhous mentioned makes sense to me. -- 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]
