rahulsmahadev commented on code in PR #17622: URL: https://github.com/apache/iceberg/pull/17622#discussion_r3808793421
########## spark/v4.1/spark/src/main/java/org/apache/iceberg/spark/actions/RepairTableSparkAction.java: ########## @@ -0,0 +1,747 @@ +/* + * 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.iceberg.spark.actions; + +import static org.apache.iceberg.MetadataTableType.ENTRIES; + +import java.io.Serializable; +import java.util.Iterator; +import java.util.List; +import java.util.Set; +import java.util.UUID; +import java.util.stream.Collectors; +import org.apache.hadoop.fs.Path; +import org.apache.iceberg.ContentFile; +import org.apache.iceberg.DataFile; +import org.apache.iceberg.DeleteFile; +import org.apache.iceberg.FileFormat; +import org.apache.iceberg.HasTableOperations; +import org.apache.iceberg.ManifestContent; +import org.apache.iceberg.ManifestFile; +import org.apache.iceberg.ManifestFiles; +import org.apache.iceberg.ManifestWriter; +import org.apache.iceberg.Metrics; +import org.apache.iceberg.MetricsConfig; +import org.apache.iceberg.PartitionSpec; +import org.apache.iceberg.Partitioning; +import org.apache.iceberg.RollingManifestWriter; +import org.apache.iceberg.Snapshot; +import org.apache.iceberg.Table; +import org.apache.iceberg.TableOperations; +import org.apache.iceberg.TableProperties; +import org.apache.iceberg.actions.ImmutableRepairTable; +import org.apache.iceberg.actions.RepairTable; +import org.apache.iceberg.exceptions.CleanableFailure; +import org.apache.iceberg.exceptions.CommitStateUnknownException; +import org.apache.iceberg.io.FileIO; +import org.apache.iceberg.io.OutputFile; +import org.apache.iceberg.io.SupportsBulkOperations; +import org.apache.iceberg.mapping.NameMapping; +import org.apache.iceberg.relocated.com.google.common.collect.ImmutableList; +import org.apache.iceberg.relocated.com.google.common.collect.Iterables; +import org.apache.iceberg.relocated.com.google.common.collect.Lists; +import org.apache.iceberg.spark.JobGroupInfo; +import org.apache.iceberg.spark.SparkContentFile; +import org.apache.iceberg.spark.SparkDataFile; +import org.apache.iceberg.spark.SparkDeleteFile; +import org.apache.iceberg.spark.source.SerializableTableWithSize; +import org.apache.iceberg.types.Types; +import org.apache.iceberg.util.PropertyUtil; +import org.apache.iceberg.util.ThreadPools; +import org.apache.spark.api.java.function.MapPartitionsFunction; +import org.apache.spark.broadcast.Broadcast; +import org.apache.spark.sql.Dataset; +import org.apache.spark.sql.Encoder; +import org.apache.spark.sql.Encoders; +import org.apache.spark.sql.Row; +import org.apache.spark.sql.SparkSession; +import org.apache.spark.sql.types.StructType; +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; + +/** + * An action that repairs incorrect statistics in the manifests of a table. + * + * <p>The statistics of every live manifest entry are compared against the file the entry refers to. + * Only manifests that contain at least one incorrect entry are rewritten, so the cost of the commit + * is proportional to the number of incorrect entries rather than to the size of the table. + */ +public class RepairTableSparkAction extends BaseSnapshotUpdateSparkAction<RepairTableSparkAction> + implements RepairTable { + + public static final String USE_CACHING = "use-caching"; + public static final boolean USE_CACHING_DEFAULT = false; + + /** + * Whether to compare and repair column level statistics, which requires reading the footer of + * every file. When disabled, only record counts and file sizes are repaired. + */ + public static final String REPAIR_COLUMN_METRICS = "repair-column-metrics"; + + public static final boolean REPAIR_COLUMN_METRICS_DEFAULT = true; + + private static final Logger LOG = LoggerFactory.getLogger(RepairTableSparkAction.class); + + private static final RepairTable.Result EMPTY_RESULT = + ImmutableRepairTable.Result.builder() + .repairedManifests(ImmutableList.of()) + .repairedEntryCount(0L) + .build(); + + private static final String NEW_MANIFEST_PREFIX = "repaired-m-"; + + private final Table table; + private final int formatVersion; + private final long targetManifestSizeBytes; + private final boolean shouldStageManifests; + private final String outputLocation; + + private boolean dryRun = false; + + RepairTableSparkAction(SparkSession spark, Table table) { + super(spark); + this.table = table; + this.targetManifestSizeBytes = + PropertyUtil.propertyAsLong( + table.properties(), + TableProperties.MANIFEST_TARGET_SIZE_BYTES, + TableProperties.MANIFEST_TARGET_SIZE_BYTES_DEFAULT); + + TableOperations ops = ((HasTableOperations) table).operations(); + Path metadataFilePath = new Path(ops.metadataFileLocation("file")); + this.outputLocation = metadataFilePath.getParent().toString(); + this.formatVersion = ops.current().formatVersion(); + + boolean snapshotIdInheritanceEnabled = + PropertyUtil.propertyAsBoolean( + table.properties(), + TableProperties.SNAPSHOT_ID_INHERITANCE_ENABLED, + TableProperties.SNAPSHOT_ID_INHERITANCE_ENABLED_DEFAULT); + this.shouldStageManifests = formatVersion == 1 && !snapshotIdInheritanceEnabled; + } + + @Override + protected RepairTableSparkAction self() { + return this; + } + + @Override + public RepairTableSparkAction repairFileMetrics() { + // repairing entry stats is currently the only repair this action performs + return this; + } + + @Override + public RepairTableSparkAction dryRun() { + this.dryRun = true; + return this; + } + + @Override + public RepairTable.Result execute() { + String desc = String.format("Repairing manifests in %s (dryRun=%s)", table.name(), dryRun); + JobGroupInfo info = newJobGroupInfo("REPAIR-TABLE", desc); + return withJobGroupInfo(info, this::doExecute); + } + + private RepairTable.Result doExecute() { + Snapshot currentSnapshot = table.currentSnapshot(); + if (currentSnapshot == null) { + return EMPTY_RESULT; + } + + List<ManifestFile> repairedManifests = Lists.newArrayList(); + List<ManifestFile> newManifests = Lists.newArrayList(); + long repairedCount = 0L; + + for (ManifestContent content : ManifestContent.values()) { + RepairedManifests repaired = repairTable(content, currentSnapshot); + repairedManifests.addAll(repaired.repairedManifests()); + newManifests.addAll(repaired.newManifests()); + repairedCount += repaired.repairedCount(); + } + + if (repairedManifests.isEmpty()) { + return EMPTY_RESULT; + } + + if (dryRun) { + // the new manifests were written to determine what the repair would produce + deleteFiles(Iterables.transform(newManifests, ManifestFile::path)); + } else { + replaceManifests(repairedManifests, newManifests); + } + + LOG.info( + "Repaired the stats of {} manifest entries, rewriting {} manifests as {} (dryRun={})", + repairedCount, + repairedManifests.size(), + newManifests.size(), + dryRun); + + return ImmutableRepairTable.Result.builder() + .repairedManifests(repairedManifests) + .repairedEntryCount(repairedCount) + .build(); + } + + private RepairedManifests repairTable(ManifestContent content, Snapshot snapshot) { + List<ManifestFile> manifests = loadManifests(content, snapshot); + if (manifests.isEmpty()) { + return RepairedManifests.empty(); + } + + Dataset<Row> entryDF = buildManifestEntryDF(manifests); + + return withReusableDS( + entryDF, + df -> { + // find the entries whose stats disagree with the files they refer to + List<EntryVerdict> verdicts = + df.mapPartitions(newCheckStatsFunc(content), Encoders.bean(EntryVerdict.class)) + .collectAsList(); + + if (verdicts.isEmpty()) { + return RepairedManifests.empty(); + } + + long repairedCount = verdicts.size(); + + Set<String> manifestsToRewrite = + verdicts.stream().map(EntryVerdict::getManifest).collect(Collectors.toSet()); + List<ManifestFile> rewritten = + manifests.stream() + .filter(manifest -> manifestsToRewrite.contains(manifest.path())) + .collect(Collectors.toList()); + + Set<String> repairedPaths = + verdicts.stream().map(EntryVerdict::getPath).collect(Collectors.toSet()); + + // rewrite every entry of the affected manifests, repairing the incorrect ones + Dataset<Row> entriesToRewrite = + df.filter(df.col("manifest").isin(manifestsToRewrite.toArray())); + List<ManifestFile> written = + writeManifests(content, entriesToRewrite, rewritten.size(), repairedPaths); + + return RepairedManifests.of(rewritten, written, repairedCount); + }); + } + + /** + * Loads the live entries of the given manifests, keeping the manifest each entry was read from so + * that only the manifests containing an incorrect entry are rewritten. + */ + private Dataset<Row> buildManifestEntryDF(List<ManifestFile> manifests) { + Dataset<Row> manifestDF = + spark() + .createDataset(Lists.transform(manifests, ManifestFile::path), Encoders.STRING()) + .toDF("manifest"); + + Dataset<Row> entryDF = + loadMetadataTable(table, ENTRIES) + .filter("status < 2") // select only live entries + .selectExpr( + "input_file_name() as manifest", + "snapshot_id", + "sequence_number", + "file_sequence_number", + "data_file"); + + return entryDF.join( + manifestDF, manifestDF.col("manifest").equalTo(entryDF.col("manifest")), "left_semi"); + } + + private List<ManifestFile> writeManifests( + ManifestContent content, Dataset<Row> entryDF, int numManifests, Set<String> repairedPaths) { + StructType sparkType = (StructType) entryDF.schema().apply("data_file").dataType(); + Types.StructType combinedFileType = DataFile.getType(Partitioning.partitionType(table)); + ManifestWriterFactory writers = manifestWriters(); + Broadcast<Set<String>> repaired = sparkContext().broadcast(repairedPaths); + RepairContext context = newRepairContext(content); + + WriteManifests<?> writeFunc = + content == ManifestContent.DATA + ? new WriteDataManifests(writers, combinedFileType, sparkType, repaired, context) + : new WriteDeleteManifests(writers, combinedFileType, sparkType, repaired, context); + + // preserve the entry order of the manifests being rewritten + return writeFunc.apply(entryDF.repartition(numManifests)).collectAsList(); + } + + private CheckStats newCheckStatsFunc(ManifestContent content) { + return new CheckStats(newRepairContext(content)); + } + + private RepairContext newRepairContext(ManifestContent content) { + boolean repairColumnMetrics = + PropertyUtil.propertyAsBoolean( + options(), REPAIR_COLUMN_METRICS, REPAIR_COLUMN_METRICS_DEFAULT); + return new RepairContext( + sparkContext().broadcast(SerializableTableWithSize.copyOf(table)), + content, + repairColumnMetrics); + } + + private List<ManifestFile> loadManifests(ManifestContent content, Snapshot snapshot) { + switch (content) { + case DATA: + return snapshot.dataManifests(table.io()); + case DELETES: + return snapshot.deleteManifests(table.io()); + default: + throw new IllegalArgumentException("Unknown manifest content: " + content); + } + } + + private void replaceManifests( + Iterable<ManifestFile> deletedManifests, Iterable<ManifestFile> addedManifests) { + try { + org.apache.iceberg.RewriteManifests rewriteManifests = table.rewriteManifests(); + deletedManifests.forEach(rewriteManifests::deleteManifest); + addedManifests.forEach(rewriteManifests::addManifest); + commit(rewriteManifests); + + if (shouldStageManifests) { + // delete new manifests as they were rewritten before the commit + deleteFiles(Iterables.transform(addedManifests, ManifestFile::path)); + } + } catch (CommitStateUnknownException e) { + // don't clean up added manifest files, because they may have been successfully committed + throw e; + } catch (Exception e) { + if (e instanceof CleanableFailure) { + deleteFiles(Iterables.transform(addedManifests, ManifestFile::path)); + } + + throw e; + } + } + + private void deleteFiles(Iterable<String> locations) { + Iterable<FileInfo> files = + Iterables.transform(locations, location -> new FileInfo(location, MANIFEST)); + if (table.io() instanceof SupportsBulkOperations) { + deleteFiles((SupportsBulkOperations) table.io(), files.iterator()); + } else { + deleteFiles( + ThreadPools.getWorkerPool(), file -> table.io().deleteFile(file), files.iterator()); + } + } + + private ManifestWriterFactory manifestWriters() { + return new ManifestWriterFactory( + sparkContext().broadcast(SerializableTableWithSize.copyOf(table)), + formatVersion, + table.spec().specId(), Review Comment: Very good catch and thanks for the repro, handled it in https://github.com/apache/iceberg/pull/17622/changes/063328fa11eed44a44a58dacb7ee4f73c47bc71a -- 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] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
