Examples with SparkTaskContextSupplier org.apache.hudi.client.SparkTaskContextSupplier used on opensource projects

Example 1 with SparkTaskContextSupplier

use of org.apache.hudi.client.SparkTaskContextSupplier in project hudi by apache.

the class MultipleSparkJobExecutionStrategy method readRecordsForGroupWithLogs.

/**
 * Read records from baseFiles, apply updates and convert to RDD.
 */
private HoodieData<HoodieRecord<T>> readRecordsForGroupWithLogs(JavaSparkContext jsc, List<ClusteringOperation> clusteringOps, String instantTime) {
    HoodieWriteConfig config = getWriteConfig();
    HoodieTable table = getHoodieTable();
    return HoodieJavaRDD.of(jsc.parallelize(clusteringOps, clusteringOps.size()).mapPartitions(clusteringOpsPartition -> {
        List<Iterator<HoodieRecord<T>>> recordIterators = new ArrayList<>();
        clusteringOpsPartition.forEachRemaining(clusteringOp -> {
            long maxMemoryPerCompaction = IOUtils.getMaxMemoryPerCompaction(new SparkTaskContextSupplier(), config);
            LOG.info("MaxMemoryPerCompaction run as part of clustering => " + maxMemoryPerCompaction);
            try {
                Schema readerSchema = HoodieAvroUtils.addMetadataFields(new Schema.Parser().parse(config.getSchema()));
                HoodieMergedLogRecordScanner scanner = HoodieMergedLogRecordScanner.newBuilder().withFileSystem(table.getMetaClient().getFs()).withBasePath(table.getMetaClient().getBasePath()).withLogFilePaths(clusteringOp.getDeltaFilePaths()).withReaderSchema(readerSchema).withLatestInstantTime(instantTime).withMaxMemorySizeInBytes(maxMemoryPerCompaction).withReadBlocksLazily(config.getCompactionLazyBlockReadEnabled()).withReverseReader(config.getCompactionReverseLogReadEnabled()).withBufferSize(config.getMaxDFSStreamBufferSize()).withSpillableMapBasePath(config.getSpillableMapBasePath()).withPartition(clusteringOp.getPartitionPath()).build();
                Option<HoodieFileReader> baseFileReader = StringUtils.isNullOrEmpty(clusteringOp.getDataFilePath()) ? Option.empty() : Option.of(HoodieFileReaderFactory.getFileReader(table.getHadoopConf(), new Path(clusteringOp.getDataFilePath())));
                HoodieTableConfig tableConfig = table.getMetaClient().getTableConfig();
                recordIterators.add(getFileSliceReader(baseFileReader, scanner, readerSchema, tableConfig.getPayloadClass(), tableConfig.getPreCombineField(), tableConfig.populateMetaFields() ? Option.empty() : Option.of(Pair.of(tableConfig.getRecordKeyFieldProp(), tableConfig.getPartitionFieldProp()))));
            } catch (IOException e) {
                throw new HoodieClusteringException("Error reading input data for " + clusteringOp.getDataFilePath() + " and " + clusteringOp.getDeltaFilePaths(), e);
            }
        });
        return new ConcatenatingIterator<>(recordIterators);
    }));
}

Example 2 with SparkTaskContextSupplier

use of org.apache.hudi.client.SparkTaskContextSupplier in project hudi by apache.

the class TestHoodieClientOnCopyOnWriteStorage method testUpsertsInternal.

/**
 * Test one of HoodieWriteClient upsert(Prepped) APIs.
 *
 * @param config Write Config
 * @param writeFn One of Hoodie Write Function API
 * @throws Exception in case of error
 */
private void testUpsertsInternal(HoodieWriteConfig config, Function3<JavaRDD<WriteStatus>, SparkRDDWriteClient, JavaRDD<HoodieRecord>, String> writeFn, boolean isPrepped) throws Exception {
    // Force using older timeline layout
    HoodieWriteConfig hoodieWriteConfig = getConfigBuilder(HoodieFailedWritesCleaningPolicy.LAZY).withRollbackUsingMarkers(true).withProps(config.getProps()).withTimelineLayoutVersion(VERSION_0).build();
    HoodieTableMetaClient.withPropertyBuilder().fromMetaClient(metaClient).setTimelineLayoutVersion(VERSION_0).setPopulateMetaFields(config.populateMetaFields()).initTable(metaClient.getHadoopConf(), metaClient.getBasePath());
    SparkRDDWriteClient client = getHoodieWriteClient(hoodieWriteConfig);
    // Write 1 (only inserts)
    String newCommitTime = "001";
    String initCommitTime = "000";
    int numRecords = 200;
    insertFirstBatch(hoodieWriteConfig, client, newCommitTime, initCommitTime, numRecords, SparkRDDWriteClient::insert, isPrepped, true, numRecords, config.populateMetaFields());
    // Write 2 (updates)
    String prevCommitTime = newCommitTime;
    newCommitTime = "004";
    numRecords = 100;
    String commitTimeBetweenPrevAndNew = "002";
    updateBatch(hoodieWriteConfig, client, newCommitTime, prevCommitTime, Option.of(Arrays.asList(commitTimeBetweenPrevAndNew)), initCommitTime, numRecords, writeFn, isPrepped, true, numRecords, 200, 2, config.populateMetaFields());
    // Delete 1
    prevCommitTime = newCommitTime;
    newCommitTime = "005";
    numRecords = 50;
    deleteBatch(hoodieWriteConfig, client, newCommitTime, prevCommitTime, initCommitTime, numRecords, SparkRDDWriteClient::delete, isPrepped, true, 0, 150, config.populateMetaFields());
    // Now simulate an upgrade and perform a restore operation
    HoodieWriteConfig newConfig = getConfigBuilder().withProps(config.getProps()).withTimelineLayoutVersion(TimelineLayoutVersion.CURR_VERSION).build();
    client = getHoodieWriteClient(newConfig);
    client.savepoint("004", "user1", "comment1");
    client.restoreToInstant("004");
    assertFalse(metaClient.reloadActiveTimeline().getRollbackTimeline().lastInstant().isPresent());
    // Check the entire dataset has all records still
    String[] fullPartitionPaths = new String[dataGen.getPartitionPaths().length];
    for (int i = 0; i < fullPartitionPaths.length; i++) {
        fullPartitionPaths[i] = String.format("%s/%s/*", basePath, dataGen.getPartitionPaths()[i]);
    }
    assertEquals(200, HoodieClientTestUtils.read(jsc, basePath, sqlContext, fs, fullPartitionPaths).count(), "Must contain " + 200 + " records");
    // Perform Delete again on upgraded dataset.
    prevCommitTime = newCommitTime;
    newCommitTime = "006";
    numRecords = 50;
    deleteBatch(newConfig, client, newCommitTime, prevCommitTime, initCommitTime, numRecords, SparkRDDWriteClient::delete, isPrepped, true, 0, 150);
    HoodieActiveTimeline activeTimeline = new HoodieActiveTimeline(metaClient, false);
    List<HoodieInstant> instants = activeTimeline.getCommitTimeline().getInstants().collect(Collectors.toList());
    assertEquals(5, instants.size());
    assertEquals(new HoodieInstant(COMPLETED, COMMIT_ACTION, "001"), instants.get(0));
    assertEquals(new HoodieInstant(COMPLETED, COMMIT_ACTION, "004"), instants.get(1));
    // New Format should have all states of instants
    assertEquals(new HoodieInstant(REQUESTED, COMMIT_ACTION, "006"), instants.get(2));
    assertEquals(new HoodieInstant(INFLIGHT, COMMIT_ACTION, "006"), instants.get(3));
    assertEquals(new HoodieInstant(COMPLETED, COMMIT_ACTION, "006"), instants.get(4));
    final HoodieWriteConfig cfg = hoodieWriteConfig;
    final String instantTime = "007";
    HoodieTableMetaClient metaClient = HoodieTableMetaClient.builder().setConf(jsc.hadoopConfiguration()).setBasePath(basePath).build();
    String basePathStr = basePath;
    HoodieTable table = getHoodieTable(metaClient, cfg);
    String extension = metaClient.getTableConfig().getBaseFileFormat().getFileExtension();
    jsc.parallelize(Arrays.asList(1)).map(e -> {
        HoodieCommitMetadata commitMetadata = HoodieCommitMetadata.fromBytes(metaClient.getActiveTimeline().getInstantDetails(metaClient.getCommitsTimeline().filterCompletedInstants().lastInstant().get()).get(), HoodieCommitMetadata.class);
        String filePath = commitMetadata.getPartitionToWriteStats().values().stream().flatMap(w -> w.stream()).filter(s -> s.getPath().endsWith(extension)).findAny().map(ee -> ee.getPath()).orElse(null);
        String partitionPath = commitMetadata.getPartitionToWriteStats().values().stream().flatMap(w -> w.stream()).filter(s -> s.getPath().endsWith(extension)).findAny().map(ee -> ee.getPartitionPath()).orElse(null);
        Path baseFilePath = new Path(basePathStr, filePath);
        HoodieBaseFile baseFile = new HoodieBaseFile(baseFilePath.toString());
        try {
            HoodieMergeHandle handle = new HoodieMergeHandle(cfg, instantTime, table, new HashMap<>(), partitionPath, FSUtils.getFileId(baseFilePath.getName()), baseFile, new SparkTaskContextSupplier(), config.populateMetaFields() ? Option.empty() : Option.of((BaseKeyGenerator) HoodieSparkKeyGeneratorFactory.createKeyGenerator(new TypedProperties(config.getProps()))));
            WriteStatus writeStatus = new WriteStatus(false, 0.0);
            writeStatus.setStat(new HoodieWriteStat());
            writeStatus.getStat().setNumWrites(0);
            handle.performMergeDataValidationCheck(writeStatus);
        } catch (HoodieCorruptedDataException e1) {
            fail("Exception not expected because merge validation check is disabled");
        }
        try {
            final String newInstantTime = "006";
            cfg.getProps().setProperty("hoodie.merge.data.validation.enabled", "true");
            HoodieWriteConfig cfg2 = HoodieWriteConfig.newBuilder().withProps(cfg.getProps()).build();
            HoodieMergeHandle handle = new HoodieMergeHandle(cfg2, newInstantTime, table, new HashMap<>(), partitionPath, FSUtils.getFileId(baseFilePath.getName()), baseFile, new SparkTaskContextSupplier(), config.populateMetaFields() ? Option.empty() : Option.of((BaseKeyGenerator) HoodieSparkKeyGeneratorFactory.createKeyGenerator(new TypedProperties(config.getProps()))));
            WriteStatus writeStatus = new WriteStatus(false, 0.0);
            writeStatus.setStat(new HoodieWriteStat());
            writeStatus.getStat().setNumWrites(0);
            handle.performMergeDataValidationCheck(writeStatus);
            fail("The above line should have thrown an exception");
        } catch (HoodieCorruptedDataException e2) {
        // expected
        }
        return true;
    }).collect();
}

Example 3 with SparkTaskContextSupplier

use of org.apache.hudi.client.SparkTaskContextSupplier in project hudi by apache.

the class TestHoodieFileWriterFactory method testGetFileWriter.

@Test
public void testGetFileWriter() throws IOException {
    // parquet file format.
    final String instantTime = "100";
    final Path parquetPath = new Path(basePath + "/partition/path/f1_1-0-1_000.parquet");
    final HoodieWriteConfig cfg = getConfig();
    HoodieTable table = HoodieSparkTable.create(cfg, context, metaClient);
    SparkTaskContextSupplier supplier = new SparkTaskContextSupplier();
    HoodieFileWriter<IndexedRecord> parquetWriter = HoodieFileWriterFactory.getFileWriter(instantTime, parquetPath, table, cfg, HoodieTestDataGenerator.AVRO_SCHEMA, supplier);
    assertTrue(parquetWriter instanceof HoodieParquetWriter);
    // hfile format.
    final Path hfilePath = new Path(basePath + "/partition/path/f1_1-0-1_000.hfile");
    HoodieFileWriter<IndexedRecord> hfileWriter = HoodieFileWriterFactory.getFileWriter(instantTime, hfilePath, table, cfg, HoodieTestDataGenerator.AVRO_SCHEMA, supplier);
    assertTrue(hfileWriter instanceof HoodieHFileWriter);
    // orc file format.
    final Path orcPath = new Path(basePath + "/partition/path/f1_1-0-1_000.orc");
    HoodieFileWriter<IndexedRecord> orcFileWriter = HoodieFileWriterFactory.getFileWriter(instantTime, orcPath, table, cfg, HoodieTestDataGenerator.AVRO_SCHEMA, supplier);
    assertTrue(orcFileWriter instanceof HoodieOrcWriter);
    // other file format exception.
    final Path logPath = new Path(basePath + "/partition/path/f.b51192a8-574b-4a85-b246-bcfec03ac8bf_100.log.2_1-0-1");
    final Throwable thrown = assertThrows(UnsupportedOperationException.class, () -> {
        HoodieFileWriter<IndexedRecord> logWriter = HoodieFileWriterFactory.getFileWriter(instantTime, logPath, table, cfg, HoodieTestDataGenerator.AVRO_SCHEMA, supplier);
    }, "should fail since log storage writer is not supported yet.");
    assertTrue(thrown.getMessage().contains("format not supported yet."));
}