Examples with HoodieTimer org.apache.hudi.common.util.HoodieTimer used on opensource projects

Example 21 with HoodieTimer

use of org.apache.hudi.common.util.HoodieTimer in project hudi by apache.

the class RocksDBDAO method prefixSearch.

/**
 * Perform a prefix search and return stream of key-value pairs retrieved.
 *
 * @param columnFamilyName Column Family Name
 * @param prefix Prefix Key
 * @param <T> Type of value stored
 */
public <T extends Serializable> Stream<Pair<String, T>> prefixSearch(String columnFamilyName, String prefix) {
    ValidationUtils.checkArgument(!closed);
    final HoodieTimer timer = new HoodieTimer();
    timer.startTimer();
    long timeTakenMicro = 0;
    List<Pair<String, T>> results = new LinkedList<>();
    try (final RocksIterator it = getRocksDB().newIterator(managedHandlesMap.get(columnFamilyName))) {
        it.seek(prefix.getBytes());
        while (it.isValid() && new String(it.key()).startsWith(prefix)) {
            long beginTs = System.nanoTime();
            T val = SerializationUtils.deserialize(it.value());
            timeTakenMicro += ((System.nanoTime() - beginTs) / 1000);
            results.add(Pair.of(new String(it.key()), val));
            it.next();
        }
    }
    LOG.info("Prefix Search for (query=" + prefix + ") on " + columnFamilyName + ". Total Time Taken (msec)=" + timer.endTimer() + ". Serialization Time taken(micro)=" + timeTakenMicro + ", num entries=" + results.size());
    return results.stream();
}

Example 22 with HoodieTimer

use of org.apache.hudi.common.util.HoodieTimer in project hudi by apache.

the class MarkerDirState method flushMarkersToFile.

/**
 * Flushes markers to the underlying file.
 *
 * @param markerFileIndex  file index to use.
 */
private void flushMarkersToFile(int markerFileIndex) {
    LOG.debug("Write to " + markerDirPath + "/" + MARKERS_FILENAME_PREFIX + markerFileIndex);
    HoodieTimer timer = new HoodieTimer().startTimer();
    Path markersFilePath = new Path(markerDirPath, MARKERS_FILENAME_PREFIX + markerFileIndex);
    FSDataOutputStream fsDataOutputStream = null;
    BufferedWriter bufferedWriter = null;
    try {
        fsDataOutputStream = fileSystem.create(markersFilePath);
        bufferedWriter = new BufferedWriter(new OutputStreamWriter(fsDataOutputStream, StandardCharsets.UTF_8));
        bufferedWriter.write(fileMarkersMap.get(markerFileIndex).toString());
    } catch (IOException e) {
        throw new HoodieIOException("Failed to overwrite marker file " + markersFilePath, e);
    } finally {
        closeQuietly(bufferedWriter);
        closeQuietly(fsDataOutputStream);
    }
    LOG.debug(markersFilePath.toString() + " written in " + timer.endTimer() + " ms");
}

Example 23 with HoodieTimer

use of org.apache.hudi.common.util.HoodieTimer in project hudi by apache.

the class BatchedMarkerCreationRunnable method run.

@Override
public void run() {
    LOG.debug("Start processing create marker requests");
    HoodieTimer timer = new HoodieTimer().startTimer();
    for (BatchedMarkerCreationContext requestContext : requestContextList) {
        requestContext.getMarkerDirState().processMarkerCreationRequests(requestContext.getFutures(), requestContext.getFileIndex());
    }
    LOG.debug("Finish batch processing of create marker requests in " + timer.endTimer() + " ms");
}

Example 24 with HoodieTimer

use of org.apache.hudi.common.util.HoodieTimer in project hudi by apache.

the class TestHoodieBackedMetadata method validateMetadata.

private void validateMetadata(SparkRDDWriteClient testClient) throws IOException {
    HoodieWriteConfig config = testClient.getConfig();
    SparkRDDWriteClient client;
    if (config.isEmbeddedTimelineServerEnabled()) {
        testClient.close();
        client = new SparkRDDWriteClient(testClient.getEngineContext(), testClient.getConfig());
    } else {
        client = testClient;
    }
    HoodieTableMetadata tableMetadata = metadata(client);
    assertNotNull(tableMetadata, "MetadataReader should have been initialized");
    if (!config.isMetadataTableEnabled()) {
        return;
    }
    HoodieTimer timer = new HoodieTimer().startTimer();
    HoodieSparkEngineContext engineContext = new HoodieSparkEngineContext(jsc);
    // Partitions should match
    FileSystemBackedTableMetadata fsBackedTableMetadata = new FileSystemBackedTableMetadata(engineContext, new SerializableConfiguration(hadoopConf), config.getBasePath(), config.shouldAssumeDatePartitioning());
    List<String> fsPartitions = fsBackedTableMetadata.getAllPartitionPaths();
    List<String> metadataPartitions = tableMetadata.getAllPartitionPaths();
    Collections.sort(fsPartitions);
    Collections.sort(metadataPartitions);
    assertEquals(fsPartitions.size(), metadataPartitions.size(), "Partitions should match");
    assertTrue(fsPartitions.equals(metadataPartitions), "Partitions should match");
    // Files within each partition should match
    metaClient = HoodieTableMetaClient.reload(metaClient);
    HoodieTable table = HoodieSparkTable.create(config, engineContext);
    TableFileSystemView tableView = table.getHoodieView();
    List<String> fullPartitionPaths = fsPartitions.stream().map(partition -> basePath + "/" + partition).collect(Collectors.toList());
    Map<String, FileStatus[]> partitionToFilesMap = tableMetadata.getAllFilesInPartitions(fullPartitionPaths);
    assertEquals(fsPartitions.size(), partitionToFilesMap.size());
    fsPartitions.forEach(partition -> {
        try {
            Path partitionPath;
            if (partition.equals("")) {
                // Should be the non-partitioned case
                partitionPath = new Path(basePath);
            } else {
                partitionPath = new Path(basePath, partition);
            }
            FileStatus[] fsStatuses = FSUtils.getAllDataFilesInPartition(fs, partitionPath);
            FileStatus[] metaStatuses = tableMetadata.getAllFilesInPartition(partitionPath);
            List<String> fsFileNames = Arrays.stream(fsStatuses).map(s -> s.getPath().getName()).collect(Collectors.toList());
            List<String> metadataFilenames = Arrays.stream(metaStatuses).map(s -> s.getPath().getName()).collect(Collectors.toList());
            Collections.sort(fsFileNames);
            Collections.sort(metadataFilenames);
            assertEquals(fsStatuses.length, partitionToFilesMap.get(partitionPath.toString()).length);
            // File sizes should be valid
            Arrays.stream(metaStatuses).forEach(s -> assertTrue(s.getLen() > 0));
            if ((fsFileNames.size() != metadataFilenames.size()) || (!fsFileNames.equals(metadataFilenames))) {
                LOG.info("*** File system listing = " + Arrays.toString(fsFileNames.toArray()));
                LOG.info("*** Metadata listing = " + Arrays.toString(metadataFilenames.toArray()));
                for (String fileName : fsFileNames) {
                    if (!metadataFilenames.contains(fileName)) {
                        LOG.error(partition + "FsFilename " + fileName + " not found in Meta data");
                    }
                }
                for (String fileName : metadataFilenames) {
                    if (!fsFileNames.contains(fileName)) {
                        LOG.error(partition + "Metadata file " + fileName + " not found in original FS");
                    }
                }
            }
            // Block sizes should be valid
            Arrays.stream(metaStatuses).forEach(s -> assertTrue(s.getBlockSize() > 0));
            List<Long> fsBlockSizes = Arrays.stream(fsStatuses).map(FileStatus::getBlockSize).collect(Collectors.toList());
            Collections.sort(fsBlockSizes);
            List<Long> metadataBlockSizes = Arrays.stream(metaStatuses).map(FileStatus::getBlockSize).collect(Collectors.toList());
            Collections.sort(metadataBlockSizes);
            assertEquals(fsBlockSizes, metadataBlockSizes);
            assertEquals(fsFileNames.size(), metadataFilenames.size(), "Files within partition " + partition + " should match");
            assertTrue(fsFileNames.equals(metadataFilenames), "Files within partition " + partition + " should match");
            // FileSystemView should expose the same data
            List<HoodieFileGroup> fileGroups = tableView.getAllFileGroups(partition).collect(Collectors.toList());
            fileGroups.addAll(tableView.getAllReplacedFileGroups(partition).collect(Collectors.toList()));
            fileGroups.forEach(g -> LogManager.getLogger(TestHoodieBackedMetadata.class).info(g));
            fileGroups.forEach(g -> g.getAllBaseFiles().forEach(b -> LogManager.getLogger(TestHoodieBackedMetadata.class).info(b)));
            fileGroups.forEach(g -> g.getAllFileSlices().forEach(s -> LogManager.getLogger(TestHoodieBackedMetadata.class).info(s)));
            long numFiles = fileGroups.stream().mapToLong(g -> g.getAllBaseFiles().count() + g.getAllFileSlices().mapToLong(s -> s.getLogFiles().count()).sum()).sum();
            assertEquals(metadataFilenames.size(), numFiles);
        } catch (IOException e) {
            e.printStackTrace();
            assertTrue(false, "Exception should not be raised: " + e);
        }
    });
    HoodieBackedTableMetadataWriter metadataWriter = metadataWriter(client);
    assertNotNull(metadataWriter, "MetadataWriter should have been initialized");
    // Validate write config for metadata table
    HoodieWriteConfig metadataWriteConfig = metadataWriter.getWriteConfig();
    assertFalse(metadataWriteConfig.isMetadataTableEnabled(), "No metadata table for metadata table");
    // Metadata table should be in sync with the dataset
    HoodieTableMetaClient metadataMetaClient = HoodieTableMetaClient.builder().setConf(hadoopConf).setBasePath(metadataTableBasePath).build();
    // Metadata table is MOR
    assertEquals(metadataMetaClient.getTableType(), HoodieTableType.MERGE_ON_READ, "Metadata Table should be MOR");
    // Metadata table is HFile format
    assertEquals(metadataMetaClient.getTableConfig().getBaseFileFormat(), HoodieFileFormat.HFILE, "Metadata Table base file format should be HFile");
    // Metadata table has a fixed number of partitions
    // Cannot use FSUtils.getAllFoldersWithPartitionMetaFile for this as that function filters all directory
    // in the .hoodie folder.
    List<String> metadataTablePartitions = FSUtils.getAllPartitionPaths(engineContext, HoodieTableMetadata.getMetadataTableBasePath(basePath), false, false);
    assertEquals(metadataWriter.getEnabledPartitionTypes().size(), metadataTablePartitions.size());
    final Map<String, MetadataPartitionType> metadataEnabledPartitionTypes = new HashMap<>();
    metadataWriter.getEnabledPartitionTypes().forEach(e -> metadataEnabledPartitionTypes.put(e.getPartitionPath(), e));
    // Metadata table should automatically compact and clean
    // versions are +1 as autoclean / compaction happens end of commits
    int numFileVersions = metadataWriteConfig.getCleanerFileVersionsRetained() + 1;
    HoodieTableFileSystemView fsView = new HoodieTableFileSystemView(metadataMetaClient, metadataMetaClient.getActiveTimeline());
    metadataTablePartitions.forEach(partition -> {
        List<FileSlice> latestSlices = fsView.getLatestFileSlices(partition).collect(Collectors.toList());
        assertTrue(latestSlices.stream().map(FileSlice::getBaseFile).count() <= metadataEnabledPartitionTypes.get(partition).getFileGroupCount(), "Should have a single latest base file per file group");
        assertTrue(latestSlices.size() <= metadataEnabledPartitionTypes.get(partition).getFileGroupCount(), "Should have a single latest file slice per file group");
        assertTrue(latestSlices.size() <= (numFileVersions * metadataEnabledPartitionTypes.get(partition).getFileGroupCount()), "Should limit file slice to " + numFileVersions + " per file group, but was " + latestSlices.size());
    });
    LOG.info("Validation time=" + timer.endTimer());
}

Example 25 with HoodieTimer

use of org.apache.hudi.common.util.HoodieTimer in project hudi by apache.

the class HoodieWrapperFileSystem method executeFuncWithTimeMetrics.

protected static <R> R executeFuncWithTimeMetrics(String metricName, Path p, CheckedFunction<R> func) throws IOException {
    HoodieTimer timer = new HoodieTimer().startTimer();
    R res = func.get();
    Registry registry = getMetricRegistryForPath(p);
    if (registry != null) {
        registry.increment(metricName);
        registry.add(metricName + ".totalDuration", timer.endTimer());
    }
    return res;
}