Search in sources :

Example 1 with HiveFileIterator

use of io.trino.plugin.hive.util.HiveFileIterator in project trino by trinodb.

the class BackgroundHiveSplitLoader method buildManifestFileIterator.

@VisibleForTesting
Optional<Iterator<InternalHiveSplit>> buildManifestFileIterator(InputFormat<?, ?> targetInputFormat, String partitionName, Properties schema, List<HivePartitionKey> partitionKeys, TupleDomain<HiveColumnHandle> effectivePredicate, BooleanSupplier partitionMatchSupplier, boolean s3SelectPushdownEnabled, TableToPartitionMapping tableToPartitionMapping, Path parent, List<Path> paths, boolean splittable) throws IOException {
    FileSystem targetFilesystem = hdfsEnvironment.getFileSystem(hdfsContext, parent);
    Map<Path, LocatedFileStatus> fileStatuses = new HashMap<>();
    HiveFileIterator fileStatusIterator = new HiveFileIterator(table, parent, targetFilesystem, directoryLister, namenodeStats, IGNORED, false);
    fileStatusIterator.forEachRemaining(status -> fileStatuses.put(getPathWithoutSchemeAndAuthority(status.getPath()), status));
    List<LocatedFileStatus> locatedFileStatuses = new ArrayList<>();
    for (Path path : paths) {
        LocatedFileStatus status = fileStatuses.get(getPathWithoutSchemeAndAuthority(path));
        // Some files may not be listed by HiveFileIterator - if those are included in the manifest this check will fail as well.
        if (status == null) {
            return Optional.empty();
        }
        locatedFileStatuses.add(status);
    }
    InternalHiveSplitFactory splitFactory = new InternalHiveSplitFactory(targetFilesystem, partitionName, targetInputFormat, schema, partitionKeys, effectivePredicate, partitionMatchSupplier, tableToPartitionMapping, Optional.empty(), Optional.empty(), getMaxInitialSplitSize(session), isForceLocalScheduling(session), s3SelectPushdownEnabled, transaction, maxSplitFileSize);
    return Optional.of(locatedFileStatuses.stream().map(locatedFileStatus -> splitFactory.createInternalHiveSplit(locatedFileStatus, OptionalInt.empty(), splittable, Optional.empty())).filter(Optional::isPresent).map(Optional::get).iterator());
}
Also used : Path(org.apache.hadoop.fs.Path) InternalHiveSplitFactory(io.trino.plugin.hive.util.InternalHiveSplitFactory) Optional(java.util.Optional) HashMap(java.util.HashMap) FileSystem(org.apache.hadoop.fs.FileSystem) ArrayList(java.util.ArrayList) LocatedFileStatus(org.apache.hadoop.fs.LocatedFileStatus) HiveFileIterator(io.trino.plugin.hive.util.HiveFileIterator) VisibleForTesting(com.google.common.annotations.VisibleForTesting)

Example 2 with HiveFileIterator

use of io.trino.plugin.hive.util.HiveFileIterator in project trino by trinodb.

the class BackgroundHiveSplitLoader method loadPartition.

private ListenableFuture<Void> loadPartition(HivePartitionMetadata partition) throws IOException {
    HivePartition hivePartition = partition.getHivePartition();
    String partitionName = hivePartition.getPartitionId();
    Properties schema = getPartitionSchema(table, partition.getPartition());
    List<HivePartitionKey> partitionKeys = getPartitionKeys(table, partition.getPartition());
    TupleDomain<HiveColumnHandle> effectivePredicate = compactEffectivePredicate.transformKeys(HiveColumnHandle.class::cast);
    BooleanSupplier partitionMatchSupplier = createPartitionMatchSupplier(dynamicFilter, hivePartition, getPartitionKeyColumnHandles(table, typeManager));
    if (!partitionMatchSupplier.getAsBoolean()) {
        // Avoid listing files and creating splits from a partition if it has been pruned due to dynamic filters
        return COMPLETED_FUTURE;
    }
    Path path = new Path(getPartitionLocation(table, partition.getPartition()));
    Configuration configuration = hdfsEnvironment.getConfiguration(hdfsContext, path);
    InputFormat<?, ?> inputFormat = getInputFormat(configuration, schema, false);
    FileSystem fs = hdfsEnvironment.getFileSystem(hdfsContext, path);
    boolean s3SelectPushdownEnabled = shouldEnablePushdownForTable(session, table, path.toString(), partition.getPartition());
    // S3 Select pushdown works at the granularity of individual S3 objects,
    // therefore we must not split files when it is enabled.
    // Skip header / footer lines are not splittable except for a special case when skip.header.line.count=1
    boolean splittable = !s3SelectPushdownEnabled && getFooterCount(schema) == 0 && getHeaderCount(schema) <= 1;
    if (inputFormat instanceof SymlinkTextInputFormat) {
        if (tableBucketInfo.isPresent()) {
            throw new TrinoException(NOT_SUPPORTED, "Bucketed table in SymlinkTextInputFormat is not yet supported");
        }
        InputFormat<?, ?> targetInputFormat = getInputFormat(configuration, schema, true);
        List<Path> targetPaths = hdfsEnvironment.doAs(hdfsContext.getIdentity(), () -> getTargetPathsFromSymlink(fs, path));
        Set<Path> parents = targetPaths.stream().map(Path::getParent).distinct().collect(toImmutableSet());
        if (optimizeSymlinkListing && parents.size() == 1 && !recursiveDirWalkerEnabled) {
            Optional<Iterator<InternalHiveSplit>> manifestFileIterator = buildManifestFileIterator(targetInputFormat, partitionName, schema, partitionKeys, effectivePredicate, partitionMatchSupplier, s3SelectPushdownEnabled, partition.getTableToPartitionMapping(), getOnlyElement(parents), targetPaths, splittable);
            if (manifestFileIterator.isPresent()) {
                fileIterators.addLast(manifestFileIterator.get());
                return COMPLETED_FUTURE;
            }
        }
        return createHiveSymlinkSplits(partitionName, targetInputFormat, schema, partitionKeys, effectivePredicate, partitionMatchSupplier, s3SelectPushdownEnabled, partition.getTableToPartitionMapping(), targetPaths);
    }
    Optional<BucketConversion> bucketConversion = Optional.empty();
    boolean bucketConversionRequiresWorkerParticipation = false;
    if (partition.getPartition().isPresent()) {
        Optional<HiveBucketProperty> partitionBucketProperty = partition.getPartition().get().getStorage().getBucketProperty();
        if (tableBucketInfo.isPresent() && partitionBucketProperty.isPresent()) {
            int readBucketCount = tableBucketInfo.get().getReadBucketCount();
            // TODO can partition's bucketing_version be different from table's?
            BucketingVersion bucketingVersion = partitionBucketProperty.get().getBucketingVersion();
            int partitionBucketCount = partitionBucketProperty.get().getBucketCount();
            // Here, it's just trying to see if its needs the BucketConversion.
            if (readBucketCount != partitionBucketCount) {
                bucketConversion = Optional.of(new BucketConversion(bucketingVersion, readBucketCount, partitionBucketCount, tableBucketInfo.get().getBucketColumns()));
                if (readBucketCount > partitionBucketCount) {
                    bucketConversionRequiresWorkerParticipation = true;
                }
            }
        }
    }
    Optional<BucketValidation> bucketValidation = Optional.empty();
    if (isValidateBucketing(session) && tableBucketInfo.isPresent()) {
        BucketSplitInfo info = tableBucketInfo.get();
        bucketValidation = Optional.of(new BucketValidation(info.getBucketingVersion(), info.getTableBucketCount(), info.getBucketColumns()));
    }
    InternalHiveSplitFactory splitFactory = new InternalHiveSplitFactory(fs, partitionName, inputFormat, schema, partitionKeys, effectivePredicate, partitionMatchSupplier, partition.getTableToPartitionMapping(), bucketConversionRequiresWorkerParticipation ? bucketConversion : Optional.empty(), bucketValidation, getMaxInitialSplitSize(session), isForceLocalScheduling(session), s3SelectPushdownEnabled, transaction, maxSplitFileSize);
    // on the input format to obtain file splits.
    if (shouldUseFileSplitsFromInputFormat(inputFormat)) {
        if (tableBucketInfo.isPresent()) {
            throw new TrinoException(NOT_SUPPORTED, "Trino cannot read bucketed partition in an input format with UseFileSplitsFromInputFormat annotation: " + inputFormat.getClass().getSimpleName());
        }
        if (AcidUtils.isTransactionalTable(table.getParameters())) {
            throw new TrinoException(NOT_SUPPORTED, "Hive transactional tables in an input format with UseFileSplitsFromInputFormat annotation are not supported: " + inputFormat.getClass().getSimpleName());
        }
        JobConf jobConf = toJobConf(configuration);
        FileInputFormat.setInputPaths(jobConf, path);
        // Pass SerDes and Table parameters into input format configuration
        fromProperties(schema).forEach(jobConf::set);
        InputSplit[] splits = hdfsEnvironment.doAs(hdfsContext.getIdentity(), () -> inputFormat.getSplits(jobConf, 0));
        return addSplitsToSource(splits, splitFactory);
    }
    List<Path> readPaths;
    List<HdfsFileStatusWithId> fileStatusOriginalFiles = ImmutableList.of();
    AcidInfo.Builder acidInfoBuilder = AcidInfo.builder(path);
    boolean isFullAcid = AcidUtils.isFullAcidTable(table.getParameters());
    if (AcidUtils.isTransactionalTable(table.getParameters())) {
        AcidUtils.Directory directory = hdfsEnvironment.doAs(hdfsContext.getIdentity(), () -> AcidUtils.getAcidState(path, configuration, validWriteIds.orElseThrow(() -> new IllegalStateException("No validWriteIds present")), false, true));
        if (isFullAcid) {
            // From Hive version >= 3.0, delta/base files will always have file '_orc_acid_version' with value >= '2'.
            Path baseOrDeltaPath = directory.getBaseDirectory() != null ? directory.getBaseDirectory() : (directory.getCurrentDirectories().size() > 0 ? directory.getCurrentDirectories().get(0).getPath() : null);
            if (baseOrDeltaPath != null && AcidUtils.OrcAcidVersion.getAcidVersionFromMetaFile(baseOrDeltaPath, fs) >= 2) {
                // Trino cannot read ORC ACID tables with version < 2 (written by Hive older than 3.0)
                // See https://github.com/trinodb/trino/issues/2790#issuecomment-591901728 for more context
                // We perform initial version check based on _orc_acid_version file here.
                // If we cannot verify the version (the _orc_acid_version file may not exist),
                // we will do extra check based on ORC datafile metadata in OrcPageSourceFactory.
                acidInfoBuilder.setOrcAcidVersionValidated(true);
            }
        }
        readPaths = new ArrayList<>();
        // base
        if (directory.getBaseDirectory() != null) {
            readPaths.add(directory.getBaseDirectory());
        }
        // delta directories
        for (AcidUtils.ParsedDelta delta : directory.getCurrentDirectories()) {
            if (!delta.isDeleteDelta()) {
                readPaths.add(delta.getPath());
            }
        }
        // Create a registry of delete_delta directories for the partition
        for (AcidUtils.ParsedDelta delta : directory.getCurrentDirectories()) {
            if (delta.isDeleteDelta()) {
                if (!isFullAcid) {
                    throw new TrinoException(HIVE_BAD_DATA, format("Unexpected delete delta for a non full ACID table '%s'. Would be ignored by the reader: %s", table.getSchemaTableName(), delta.getPath()));
                }
                acidInfoBuilder.addDeleteDelta(delta.getPath());
            }
        }
        // initialize original files status list if present
        fileStatusOriginalFiles = directory.getOriginalFiles();
        for (HdfsFileStatusWithId hdfsFileStatusWithId : fileStatusOriginalFiles) {
            Path originalFilePath = hdfsFileStatusWithId.getFileStatus().getPath();
            long originalFileLength = hdfsFileStatusWithId.getFileStatus().getLen();
            if (originalFileLength == 0) {
                continue;
            }
            // Hive requires "original" files of transactional tables to conform to the bucketed tables naming pattern, to match them with delete deltas.
            int bucketId = getRequiredBucketNumber(originalFilePath);
            acidInfoBuilder.addOriginalFile(originalFilePath, originalFileLength, bucketId);
        }
    } else {
        // TODO https://github.com/trinodb/trino/issues/7603 - we should not referece acidInfoBuilder at allwhen we are not reading from non-ACID table
        // no ACID; no further validation needed
        acidInfoBuilder.setOrcAcidVersionValidated(true);
        readPaths = ImmutableList.of(path);
    }
    // Bucketed partitions are fully loaded immediately since all files must be loaded to determine the file to bucket mapping
    if (tableBucketInfo.isPresent()) {
        // TODO document in addToQueue() that it is sufficient to hold on to last returned future
        ListenableFuture<Void> lastResult = immediateVoidFuture();
        for (Path readPath : readPaths) {
            // list all files in the partition
            List<LocatedFileStatus> files = new ArrayList<>();
            try {
                Iterators.addAll(files, new HiveFileIterator(table, readPath, fs, directoryLister, namenodeStats, FAIL, ignoreAbsentPartitions));
            } catch (HiveFileIterator.NestedDirectoryNotAllowedException e) {
                // Fail here to be on the safe side. This seems to be the same as what Hive does
                throw new TrinoException(HIVE_INVALID_BUCKET_FILES, format("Hive table '%s' is corrupt. Found sub-directory '%s' in bucket directory for partition: %s", table.getSchemaTableName(), e.getNestedDirectoryPath(), splitFactory.getPartitionName()));
            }
            Optional<AcidInfo> acidInfo = isFullAcid ? acidInfoBuilder.build() : Optional.empty();
            lastResult = hiveSplitSource.addToQueue(getBucketedSplits(files, splitFactory, tableBucketInfo.get(), bucketConversion, splittable, acidInfo));
        }
        for (HdfsFileStatusWithId hdfsFileStatusWithId : fileStatusOriginalFiles) {
            List<LocatedFileStatus> locatedFileStatuses = ImmutableList.of((LocatedFileStatus) hdfsFileStatusWithId.getFileStatus());
            Optional<AcidInfo> acidInfo = isFullAcid ? Optional.of(acidInfoBuilder.buildWithRequiredOriginalFiles(getRequiredBucketNumber(hdfsFileStatusWithId.getFileStatus().getPath()))) : Optional.empty();
            lastResult = hiveSplitSource.addToQueue(getBucketedSplits(locatedFileStatuses, splitFactory, tableBucketInfo.get(), bucketConversion, splittable, acidInfo));
        }
        return lastResult;
    }
    for (Path readPath : readPaths) {
        Optional<AcidInfo> acidInfo = isFullAcid ? acidInfoBuilder.build() : Optional.empty();
        fileIterators.addLast(createInternalHiveSplitIterator(readPath, fs, splitFactory, splittable, acidInfo));
    }
    if (!fileStatusOriginalFiles.isEmpty()) {
        fileIterators.addLast(generateOriginalFilesSplits(splitFactory, fileStatusOriginalFiles, splittable, acidInfoBuilder, isFullAcid));
    }
    return COMPLETED_FUTURE;
}
Also used : Configuration(org.apache.hadoop.conf.Configuration) ArrayList(java.util.ArrayList) FileSystem(org.apache.hadoop.fs.FileSystem) HiveFileIterator(io.trino.plugin.hive.util.HiveFileIterator) ConfigurationUtils.toJobConf(io.trino.plugin.hive.util.ConfigurationUtils.toJobConf) JobConf(org.apache.hadoop.mapred.JobConf) InternalHiveSplitFactory(io.trino.plugin.hive.util.InternalHiveSplitFactory) AcidUtils(org.apache.hadoop.hive.ql.io.AcidUtils) HdfsFileStatusWithId(org.apache.hadoop.hive.shims.HadoopShims.HdfsFileStatusWithId) Properties(java.util.Properties) Maps.fromProperties(com.google.common.collect.Maps.fromProperties) SymlinkTextInputFormat(org.apache.hadoop.hive.ql.io.SymlinkTextInputFormat) BucketingVersion(io.trino.plugin.hive.util.HiveBucketing.BucketingVersion) HiveFileIterator(io.trino.plugin.hive.util.HiveFileIterator) Iterator(java.util.Iterator) BooleanSupplier(java.util.function.BooleanSupplier) InputSplit(org.apache.hadoop.mapred.InputSplit) Path(org.apache.hadoop.fs.Path) LocatedFileStatus(org.apache.hadoop.fs.LocatedFileStatus) BucketValidation(io.trino.plugin.hive.HiveSplit.BucketValidation) TrinoException(io.trino.spi.TrinoException) BucketConversion(io.trino.plugin.hive.HiveSplit.BucketConversion)

Aggregations

HiveFileIterator (io.trino.plugin.hive.util.HiveFileIterator)2 InternalHiveSplitFactory (io.trino.plugin.hive.util.InternalHiveSplitFactory)2 ArrayList (java.util.ArrayList)2 FileSystem (org.apache.hadoop.fs.FileSystem)2 LocatedFileStatus (org.apache.hadoop.fs.LocatedFileStatus)2 Path (org.apache.hadoop.fs.Path)2 VisibleForTesting (com.google.common.annotations.VisibleForTesting)1 Maps.fromProperties (com.google.common.collect.Maps.fromProperties)1 BucketConversion (io.trino.plugin.hive.HiveSplit.BucketConversion)1 BucketValidation (io.trino.plugin.hive.HiveSplit.BucketValidation)1 ConfigurationUtils.toJobConf (io.trino.plugin.hive.util.ConfigurationUtils.toJobConf)1 BucketingVersion (io.trino.plugin.hive.util.HiveBucketing.BucketingVersion)1 TrinoException (io.trino.spi.TrinoException)1 HashMap (java.util.HashMap)1 Iterator (java.util.Iterator)1 Optional (java.util.Optional)1 Properties (java.util.Properties)1 BooleanSupplier (java.util.function.BooleanSupplier)1 Configuration (org.apache.hadoop.conf.Configuration)1 AcidUtils (org.apache.hadoop.hive.ql.io.AcidUtils)1