voonhous commented on code in PR #18938: URL: https://github.com/apache/hudi/pull/18938#discussion_r3434907476
########## hudi-hadoop-common/src/main/java/org/apache/hudi/io/storage/hadoop/VariantReconstruction.java: ########## @@ -0,0 +1,198 @@ +/* + * 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.hadoop; + +import org.apache.hudi.avro.VariantSchemaUtils; +import org.apache.hudi.avro.VariantShreddingProvider; +import org.apache.hudi.common.config.HoodieStorageConfig; +import org.apache.hudi.common.schema.HoodieSchema; +import org.apache.hudi.common.schema.HoodieSchemaField; +import org.apache.hudi.common.schema.HoodieSchemaType; +import org.apache.hudi.common.util.Option; +import org.apache.hudi.common.util.ReflectionUtils; +import org.apache.hudi.storage.HoodieStorage; + +import org.apache.avro.Schema; +import org.apache.avro.generic.GenericData; +import org.apache.avro.generic.GenericRecord; +import org.apache.avro.generic.IndexedRecord; +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; + +import java.util.ArrayList; +import java.util.List; + +/** + * Reconstructs unshredded variants when reading an already-shredded base file on the Avro + * ({@code HoodieRecordType.AVRO}) read path. + * + * <p>parquet-avro does not understand variant shredding, so a shredded variant column comes back as + * a raw {@code {metadata, value, typed_value}} record. This reads such columns at their shredded + * (typed_value-bearing) schema, then rebuilds the standard unshredded {@code {metadata, value}} + * variant via {@link VariantShreddingProvider#rebuildVariantRecord} before records reach the + * merger/writer. The Spark/InternalRow read path reconstructs natively and does not use this. + * + * <p>See https://github.com/apache/hudi/issues/18931. + */ +final class VariantReconstruction { + + private static final Logger LOG = LoggerFactory.getLogger(VariantReconstruction.class); + + // Classpath fallback when the provider class config is not set; mirrors HoodieAvroFileWriterFactory. + private static final String[] PROVIDER_CANDIDATES = { + "org.apache.hudi.variant.Spark4VariantShreddingProvider" + }; + + private final HoodieSchema intermediateSchema; + private final Schema outputAvroSchema; + private final VariantShreddingProvider provider; + // Indexed by field position in the (requested == output) record. For target fields, the file's + // shredded sub-schema and the unshredded target sub-schema for rebuild; null for non-targets. + private final boolean[] isTarget; + private final Schema[] shreddedSubSchemas; + private final Schema[] unshreddedSubSchemas; + + private VariantReconstruction(HoodieSchema intermediateSchema, Schema outputAvroSchema, + VariantShreddingProvider provider, boolean[] isTarget, + Schema[] shreddedSubSchemas, Schema[] unshreddedSubSchemas) { + this.intermediateSchema = intermediateSchema; + this.outputAvroSchema = outputAvroSchema; + this.provider = provider; + this.isTarget = isTarget; + this.shreddedSubSchemas = shreddedSubSchemas; + this.unshreddedSubSchemas = unshreddedSubSchemas; + } + + /** + * Schema to read the parquet file with: the requested schema, but with shredded variant columns + * swapped to their file (typed_value-bearing) form so parquet-avro materializes {@code typed_value}. + */ + HoodieSchema intermediateSchema() { + return intermediateSchema; + } + + /** + * Builds a reconstruction for the given file and requested schemas, or returns {@code null} when + * none is needed (no shredded variant columns in the file, reading shredded variants disabled, or + * no provider available - in which case the read proceeds unchanged). + */ + static VariantReconstruction create(HoodieSchema fileSchema, HoodieSchema requestedSchema, HoodieStorage storage) { + if (requestedSchema.getType() != HoodieSchemaType.RECORD || fileSchema.getType() != HoodieSchemaType.RECORD) { + return null; + } Review Comment: Good catch -- you're right, that path silently mangles data. With `create()` returning `null`, the caller falls back to the unshredded requested schema, so parquet-avro reads `value`/`metadata` and drops `typed_value`; any row whose payload lived in `typed_value` comes back as a variant with a null `value`. No loud failure, just corruption. Switched it to fail fast, mirroring the write path's missing-provider behavior: ```java VariantShreddingProvider provider = loadProvider(storage); if (provider == null) { // Reading would drop typed_value and silently corrupt variants whose payload lives there, so fail fast. throw new HoodieException("Base file has shredded variant column(s) and reading shredded variants is " + "enabled, but no VariantShreddingProvider is available to reconstruct them. Set " + HoodieStorageConfig.PARQUET_VARIANT_SHREDDING_PROVIDER_CLASS.key() + " or add a provider implementation (e.g. the Spark variant module) to the classpath."); } ``` This only triggers when a shredded variant column is actually requested (otherwise `create()` already returns `null` and the read is unaffected), so non-variant reads on the same file are fine. Removed the now-unused logger too. -- 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]
