Examples with HoodieBaseFile org.apache.hudi.common.model.HoodieBaseFile used on opensource projects

Example 36 with HoodieBaseFile

use of org.apache.hudi.common.model.HoodieBaseFile in project hudi by apache.

the class TestFileSliceMetricUtils method buildFileSlice.

private FileSlice buildFileSlice(long baseFileLen, List<Long> logFileLens) {
    final String baseFilePath = ".b5068208-e1a4-11e6-bf01-fe55135034f3_20170101134598.log.1";
    FileSlice slice = new FileSlice("partition_0", HoodieActiveTimeline.createNewInstantTime(), UUID.randomUUID().toString());
    HoodieBaseFile baseFile = new HoodieBaseFile(baseFilePath);
    baseFile.setFileLen(baseFileLen);
    slice.setBaseFile(baseFile);
    int logVersion = 1;
    for (long logFileLen : logFileLens) {
        String logFilePath = "." + UUID.randomUUID().toString() + "_20170101134598.log." + logVersion;
        HoodieLogFile logFile = new HoodieLogFile(logFilePath);
        logFile.setFileLen(logFileLen);
        slice.addLogFile(logFile);
        logVersion++;
    }
    return slice;
}

Example 37 with HoodieBaseFile

use of org.apache.hudi.common.model.HoodieBaseFile in project hudi by apache.

the class HoodieCompactor method generateCompactionPlan.

/**
 * Generate a new compaction plan for scheduling.
 *
 * @param context                               HoodieEngineContext
 * @param hoodieTable                           Hoodie Table
 * @param config                                Hoodie Write Configuration
 * @param compactionCommitTime                  scheduled compaction commit time
 * @param fgIdsInPendingCompactionAndClustering partition-fileId pairs for which compaction is pending
 * @return Compaction Plan
 * @throws IOException when encountering errors
 */
HoodieCompactionPlan generateCompactionPlan(HoodieEngineContext context, HoodieTable<T, I, K, O> hoodieTable, HoodieWriteConfig config, String compactionCommitTime, Set<HoodieFileGroupId> fgIdsInPendingCompactionAndClustering) throws IOException {
    // Accumulator to keep track of total log files for a table
    HoodieAccumulator totalLogFiles = context.newAccumulator();
    // Accumulator to keep track of total log file slices for a table
    HoodieAccumulator totalFileSlices = context.newAccumulator();
    ValidationUtils.checkArgument(hoodieTable.getMetaClient().getTableType() == HoodieTableType.MERGE_ON_READ, "Can only compact table of type " + HoodieTableType.MERGE_ON_READ + " and not " + hoodieTable.getMetaClient().getTableType().name());
    // TODO : check if maxMemory is not greater than JVM or executor memory
    // TODO - rollback any compactions in flight
    HoodieTableMetaClient metaClient = hoodieTable.getMetaClient();
    LOG.info("Compacting " + metaClient.getBasePath() + " with commit " + compactionCommitTime);
    List<String> partitionPaths = FSUtils.getAllPartitionPaths(context, config.getMetadataConfig(), metaClient.getBasePath());
    // filter the partition paths if needed to reduce list status
    partitionPaths = config.getCompactionStrategy().filterPartitionPaths(config, partitionPaths);
    if (partitionPaths.isEmpty()) {
        // In case no partitions could be picked, return no compaction plan
        return null;
    }
    SliceView fileSystemView = hoodieTable.getSliceView();
    LOG.info("Compaction looking for files to compact in " + partitionPaths + " partitions");
    context.setJobStatus(this.getClass().getSimpleName(), "Looking for files to compact");
    List<HoodieCompactionOperation> operations = context.flatMap(partitionPaths, partitionPath -> fileSystemView.getLatestFileSlices(partitionPath).filter(slice -> !fgIdsInPendingCompactionAndClustering.contains(slice.getFileGroupId())).map(s -> {
        List<HoodieLogFile> logFiles = s.getLogFiles().sorted(HoodieLogFile.getLogFileComparator()).collect(toList());
        totalLogFiles.add(logFiles.size());
        totalFileSlices.add(1L);
        // Avro generated classes are not inheriting Serializable. Using CompactionOperation POJO
        // for Map operations and collecting them finally in Avro generated classes for storing
        // into meta files.
        Option<HoodieBaseFile> dataFile = s.getBaseFile();
        return new CompactionOperation(dataFile, partitionPath, logFiles, config.getCompactionStrategy().captureMetrics(config, s));
    }).filter(c -> !c.getDeltaFileNames().isEmpty()), partitionPaths.size()).stream().map(CompactionUtils::buildHoodieCompactionOperation).collect(toList());
    LOG.info("Total of " + operations.size() + " compactions are retrieved");
    LOG.info("Total number of latest files slices " + totalFileSlices.value());
    LOG.info("Total number of log files " + totalLogFiles.value());
    LOG.info("Total number of file slices " + totalFileSlices.value());
    // Filter the compactions with the passed in filter. This lets us choose most effective
    // compactions only
    HoodieCompactionPlan compactionPlan = config.getCompactionStrategy().generateCompactionPlan(config, operations, CompactionUtils.getAllPendingCompactionPlans(metaClient).stream().map(Pair::getValue).collect(toList()));
    ValidationUtils.checkArgument(compactionPlan.getOperations().stream().noneMatch(op -> fgIdsInPendingCompactionAndClustering.contains(new HoodieFileGroupId(op.getPartitionPath(), op.getFileId()))), "Bad Compaction Plan. FileId MUST NOT have multiple pending compactions. " + "Please fix your strategy implementation. FileIdsWithPendingCompactions :" + fgIdsInPendingCompactionAndClustering + ", Selected workload :" + compactionPlan);
    if (compactionPlan.getOperations().isEmpty()) {
        LOG.warn("After filtering, Nothing to compact for " + metaClient.getBasePath());
    }
    return compactionPlan;
}

Example 38 with HoodieBaseFile

use of org.apache.hudi.common.model.HoodieBaseFile in project hudi by apache.

the class HoodieCompactor method compact.

/**
 * Execute a single compaction operation and report back status.
 */
public List<WriteStatus> compact(HoodieCompactionHandler compactionHandler, HoodieTableMetaClient metaClient, HoodieWriteConfig config, CompactionOperation operation, String instantTime, TaskContextSupplier taskContextSupplier) throws IOException {
    FileSystem fs = metaClient.getFs();
    Schema readerSchema = HoodieAvroUtils.addMetadataFields(new Schema.Parser().parse(config.getSchema()), config.allowOperationMetadataField());
    LOG.info("Compacting base " + operation.getDataFileName() + " with delta files " + operation.getDeltaFileNames() + " for commit " + instantTime);
    // TODO - FIX THIS
    // Reads the entire avro file. Always only specific blocks should be read from the avro file
    // (failure recover).
    // Load all the delta commits since the last compaction commit and get all the blocks to be
    // loaded and load it using CompositeAvroLogReader
    // Since a DeltaCommit is not defined yet, reading all the records. revisit this soon.
    String maxInstantTime = metaClient.getActiveTimeline().getTimelineOfActions(CollectionUtils.createSet(HoodieTimeline.COMMIT_ACTION, HoodieTimeline.ROLLBACK_ACTION, HoodieTimeline.DELTA_COMMIT_ACTION)).filterCompletedInstants().lastInstant().get().getTimestamp();
    long maxMemoryPerCompaction = IOUtils.getMaxMemoryPerCompaction(taskContextSupplier, config);
    LOG.info("MaxMemoryPerCompaction => " + maxMemoryPerCompaction);
    List<String> logFiles = operation.getDeltaFileNames().stream().map(p -> new Path(FSUtils.getPartitionPath(metaClient.getBasePath(), operation.getPartitionPath()), p).toString()).collect(toList());
    HoodieMergedLogRecordScanner scanner = HoodieMergedLogRecordScanner.newBuilder().withFileSystem(fs).withBasePath(metaClient.getBasePath()).withLogFilePaths(logFiles).withReaderSchema(readerSchema).withLatestInstantTime(maxInstantTime).withMaxMemorySizeInBytes(maxMemoryPerCompaction).withReadBlocksLazily(config.getCompactionLazyBlockReadEnabled()).withReverseReader(config.getCompactionReverseLogReadEnabled()).withBufferSize(config.getMaxDFSStreamBufferSize()).withSpillableMapBasePath(config.getSpillableMapBasePath()).withDiskMapType(config.getCommonConfig().getSpillableDiskMapType()).withBitCaskDiskMapCompressionEnabled(config.getCommonConfig().isBitCaskDiskMapCompressionEnabled()).withOperationField(config.allowOperationMetadataField()).withPartition(operation.getPartitionPath()).build();
    Option<HoodieBaseFile> oldDataFileOpt = operation.getBaseFile(metaClient.getBasePath(), operation.getPartitionPath());
    // But in this case, we need to give it a base file. Otherwise, it will lose base file in following fileSlice.
    if (!scanner.iterator().hasNext()) {
        if (!oldDataFileOpt.isPresent()) {
            scanner.close();
            return new ArrayList<>();
        } else {
        // TODO: we may directly rename original parquet file if there is not evolution/devolution of schema
        /*
        TaskContextSupplier taskContextSupplier = hoodieCopyOnWriteTable.getTaskContextSupplier();
        String newFileName = FSUtils.makeDataFileName(instantTime,
            FSUtils.makeWriteToken(taskContextSupplier.getPartitionIdSupplier().get(), taskContextSupplier.getStageIdSupplier().get(), taskContextSupplier.getAttemptIdSupplier().get()),
            operation.getFileId(), hoodieCopyOnWriteTable.getBaseFileExtension());
        Path oldFilePath = new Path(oldDataFileOpt.get().getPath());
        Path newFilePath = new Path(oldFilePath.getParent(), newFileName);
        FileUtil.copy(fs,oldFilePath, fs, newFilePath, false, fs.getConf());
        */
        }
    }
    // Compacting is very similar to applying updates to existing file
    Iterator<List<WriteStatus>> result;
    // If the dataFile is present, perform updates else perform inserts into a new base file.
    if (oldDataFileOpt.isPresent()) {
        result = compactionHandler.handleUpdate(instantTime, operation.getPartitionPath(), operation.getFileId(), scanner.getRecords(), oldDataFileOpt.get());
    } else {
        result = compactionHandler.handleInsert(instantTime, operation.getPartitionPath(), operation.getFileId(), scanner.getRecords());
    }
    scanner.close();
    Iterable<List<WriteStatus>> resultIterable = () -> result;
    return StreamSupport.stream(resultIterable.spliterator(), false).flatMap(Collection::stream).peek(s -> {
        s.getStat().setTotalUpdatedRecordsCompacted(scanner.getNumMergedRecordsInLog());
        s.getStat().setTotalLogFilesCompacted(scanner.getTotalLogFiles());
        s.getStat().setTotalLogRecords(scanner.getTotalLogRecords());
        s.getStat().setPartitionPath(operation.getPartitionPath());
        s.getStat().setTotalLogSizeCompacted(operation.getMetrics().get(CompactionStrategy.TOTAL_LOG_FILE_SIZE).longValue());
        s.getStat().setTotalLogBlocks(scanner.getTotalLogBlocks());
        s.getStat().setTotalCorruptLogBlock(scanner.getTotalCorruptBlocks());
        s.getStat().setTotalRollbackBlocks(scanner.getTotalRollbacks());
        RuntimeStats runtimeStats = new RuntimeStats();
        runtimeStats.setTotalScanTime(scanner.getTotalTimeTakenToReadAndMergeBlocks());
        s.getStat().setRuntimeStats(runtimeStats);
    }).collect(toList());
}

Example 39 with HoodieBaseFile

use of org.apache.hudi.common.model.HoodieBaseFile in project hudi by apache.

the class HoodieKeyLookupHandle method getLookupResult.

/**
 * Of all the keys, that were added, return a list of keys that were actually found in the file group.
 */
public HoodieKeyLookupResult getLookupResult() {
    if (LOG.isDebugEnabled()) {
        LOG.debug("#The candidate row keys for " + partitionPathFileIDPair + " => " + candidateRecordKeys);
    }
    HoodieBaseFile dataFile = getLatestDataFile();
    List<String> matchingKeys = HoodieIndexUtils.filterKeysFromFile(new Path(dataFile.getPath()), candidateRecordKeys, hoodieTable.getHadoopConf());
    LOG.info(String.format("Total records (%d), bloom filter candidates (%d)/fp(%d), actual matches (%d)", totalKeysChecked, candidateRecordKeys.size(), candidateRecordKeys.size() - matchingKeys.size(), matchingKeys.size()));
    return new HoodieKeyLookupResult(partitionPathFileIDPair.getRight(), partitionPathFileIDPair.getLeft(), dataFile.getCommitTime(), matchingKeys);
}

Example 40 with HoodieBaseFile

use of org.apache.hudi.common.model.HoodieBaseFile in project hudi by apache.

the class IncrementalTimelineSyncFileSystemView method applyDeltaFileSlicesToPartitionView.

/**
 * Apply changes to partition file-system view. Base Implementation overwrites the entire partitions view assuming
 * some sort of map (in-mem/disk-based) is used. For View implementation which supports fine-granular updates (e:g
 * RocksDB), override this method.
 *
 * @param partition PartitionPath
 * @param deltaFileGroups Changed file-slices aggregated as file-groups
 * @param mode Delta Apply mode
 */
protected void applyDeltaFileSlicesToPartitionView(String partition, List<HoodieFileGroup> deltaFileGroups, DeltaApplyMode mode) {
    if (deltaFileGroups.isEmpty()) {
        LOG.info("No delta file groups for partition :" + partition);
        return;
    }
    List<HoodieFileGroup> fileGroups = fetchAllStoredFileGroups(partition).collect(Collectors.toList());
    /**
     * Note that while finding the new data/log files added/removed, the path stored in metadata will be missing the
     * base-path,scheme and authority. Ensure the matching process takes care of this discrepancy.
     */
    Map<String, HoodieBaseFile> viewDataFiles = fileGroups.stream().flatMap(HoodieFileGroup::getAllRawFileSlices).map(FileSlice::getBaseFile).filter(Option::isPresent).map(Option::get).map(df -> Pair.of(Path.getPathWithoutSchemeAndAuthority(new Path(df.getPath())).toString(), df)).collect(Collectors.toMap(Pair::getKey, Pair::getValue));
    // Note: Delta Log Files and Data FIles can be empty when adding/removing pending compactions
    Map<String, HoodieBaseFile> deltaDataFiles = deltaFileGroups.stream().flatMap(HoodieFileGroup::getAllRawFileSlices).map(FileSlice::getBaseFile).filter(Option::isPresent).map(Option::get).map(df -> Pair.of(Path.getPathWithoutSchemeAndAuthority(new Path(df.getPath())).toString(), df)).collect(Collectors.toMap(Pair::getKey, Pair::getValue));
    Map<String, HoodieLogFile> viewLogFiles = fileGroups.stream().flatMap(HoodieFileGroup::getAllRawFileSlices).flatMap(FileSlice::getLogFiles).map(lf -> Pair.of(Path.getPathWithoutSchemeAndAuthority(lf.getPath()).toString(), lf)).collect(Collectors.toMap(Pair::getKey, Pair::getValue));
    Map<String, HoodieLogFile> deltaLogFiles = deltaFileGroups.stream().flatMap(HoodieFileGroup::getAllRawFileSlices).flatMap(FileSlice::getLogFiles).map(lf -> Pair.of(Path.getPathWithoutSchemeAndAuthority(lf.getPath()).toString(), lf)).collect(Collectors.toMap(Pair::getKey, Pair::getValue));
    switch(mode) {
        case ADD:
            viewDataFiles.putAll(deltaDataFiles);
            viewLogFiles.putAll(deltaLogFiles);
            break;
        case REMOVE:
            deltaDataFiles.keySet().stream().forEach(p -> viewDataFiles.remove(p));
            deltaLogFiles.keySet().stream().forEach(p -> viewLogFiles.remove(p));
            break;
        default:
            throw new IllegalStateException("Unknown diff apply mode=" + mode);
    }
    HoodieTimeline timeline = deltaFileGroups.stream().map(df -> df.getTimeline()).findAny().get();
    List<HoodieFileGroup> fgs = buildFileGroups(viewDataFiles.values().stream(), viewLogFiles.values().stream(), timeline, true);
    storePartitionView(partition, fgs);
}