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

Example 61 with HoodieWriteStat

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

the class TestMergeOnReadRollbackActionExecutor method testRollbackForCanIndexLogFile.

@Test
public void testRollbackForCanIndexLogFile() throws IOException {
    cleanupResources();
    setUpDFS();
    // 1. prepare data and assert data result
    // just generate one partitions
    dataGen = new HoodieTestDataGenerator(new String[] { DEFAULT_FIRST_PARTITION_PATH });
    HoodieWriteConfig cfg = HoodieWriteConfig.newBuilder().withPath(basePath).withSchema(HoodieTestDataGenerator.TRIP_EXAMPLE_SCHEMA).withParallelism(2, 2).withBulkInsertParallelism(2).withFinalizeWriteParallelism(2).withDeleteParallelism(2).withTimelineLayoutVersion(TimelineLayoutVersion.CURR_VERSION).withWriteStatusClass(MetadataMergeWriteStatus.class).withConsistencyGuardConfig(ConsistencyGuardConfig.newBuilder().withConsistencyCheckEnabled(true).build()).withCompactionConfig(HoodieCompactionConfig.newBuilder().compactionSmallFileSize(1024 * 1024).build()).withStorageConfig(HoodieStorageConfig.newBuilder().hfileMaxFileSize(1024 * 1024).parquetMaxFileSize(1024 * 1024).build()).forTable("test-trip-table").withIndexConfig(HoodieIndexConfig.newBuilder().withIndexType(HoodieIndex.IndexType.INMEMORY).build()).withEmbeddedTimelineServerEnabled(true).withFileSystemViewConfig(FileSystemViewStorageConfig.newBuilder().withEnableBackupForRemoteFileSystemView(// Fail test if problem connecting to timeline-server
    false).withStorageType(FileSystemViewStorageType.EMBEDDED_KV_STORE).build()).withRollbackUsingMarkers(false).withAutoCommit(false).build();
    // 1. prepare data
    new HoodieTestDataGenerator().writePartitionMetadata(fs, new String[] { DEFAULT_FIRST_PARTITION_PATH }, basePath);
    SparkRDDWriteClient client = getHoodieWriteClient(cfg);
    // Write 1 (only inserts)
    String newCommitTime = "001";
    client.startCommitWithTime(newCommitTime);
    List<HoodieRecord> records = dataGen.generateInsertsForPartition(newCommitTime, 2, DEFAULT_FIRST_PARTITION_PATH);
    JavaRDD<HoodieRecord> writeRecords = jsc.parallelize(records, 1);
    JavaRDD<WriteStatus> statuses = client.upsert(writeRecords, newCommitTime);
    org.apache.hudi.testutils.Assertions.assertNoWriteErrors(statuses.collect());
    client.commit(newCommitTime, statuses);
    // check fileSlice
    HoodieTable table = this.getHoodieTable(metaClient, cfg);
    SyncableFileSystemView fsView = getFileSystemViewWithUnCommittedSlices(table.getMetaClient());
    List<HoodieFileGroup> firstPartitionCommit2FileGroups = fsView.getAllFileGroups(DEFAULT_FIRST_PARTITION_PATH).collect(Collectors.toList());
    assertEquals(1, firstPartitionCommit2FileGroups.size());
    assertEquals(1, (int) firstPartitionCommit2FileGroups.get(0).getAllFileSlices().count());
    assertFalse(firstPartitionCommit2FileGroups.get(0).getAllFileSlices().findFirst().get().getBaseFile().isPresent());
    assertEquals(1, firstPartitionCommit2FileGroups.get(0).getAllFileSlices().findFirst().get().getLogFiles().count());
    String generatedFileID = firstPartitionCommit2FileGroups.get(0).getFileGroupId().getFileId();
    // check hoodieCommitMeta
    HoodieCommitMetadata commitMetadata = HoodieCommitMetadata.fromBytes(table.getMetaClient().getCommitTimeline().getInstantDetails(new HoodieInstant(true, HoodieTimeline.DELTA_COMMIT_ACTION, "001")).get(), HoodieCommitMetadata.class);
    List<HoodieWriteStat> firstPartitionWriteStat = commitMetadata.getPartitionToWriteStats().get(DEFAULT_FIRST_PARTITION_PATH);
    assertEquals(2, firstPartitionWriteStat.size());
    // we have an empty writeStat for all partition
    assert firstPartitionWriteStat.stream().anyMatch(wStat -> StringUtils.isNullOrEmpty(wStat.getFileId()));
    // we have one  non-empty writeStat which must contains update or insert
    assertEquals(1, firstPartitionWriteStat.stream().filter(wStat -> !StringUtils.isNullOrEmpty(wStat.getFileId())).count());
    firstPartitionWriteStat.stream().filter(wStat -> !StringUtils.isNullOrEmpty(wStat.getFileId())).forEach(wStat -> {
        assert wStat.getNumInserts() > 0;
    });
    // Write 2 (inserts)
    newCommitTime = "002";
    client.startCommitWithTime(newCommitTime);
    List<HoodieRecord> updateRecords = Collections.singletonList(dataGen.generateUpdateRecord(records.get(0).getKey(), newCommitTime));
    List<HoodieRecord> insertRecordsInSamePartition = dataGen.generateInsertsForPartition(newCommitTime, 2, DEFAULT_FIRST_PARTITION_PATH);
    List<HoodieRecord> insertRecordsInOtherPartition = dataGen.generateInsertsForPartition(newCommitTime, 2, DEFAULT_SECOND_PARTITION_PATH);
    List<HoodieRecord> recordsToBeWrite = Stream.concat(Stream.concat(updateRecords.stream(), insertRecordsInSamePartition.stream()), insertRecordsInOtherPartition.stream()).collect(Collectors.toList());
    writeRecords = jsc.parallelize(recordsToBeWrite, 1);
    statuses = client.upsert(writeRecords, newCommitTime);
    client.commit(newCommitTime, statuses);
    table = this.getHoodieTable(metaClient, cfg);
    commitMetadata = HoodieCommitMetadata.fromBytes(table.getMetaClient().getCommitTimeline().getInstantDetails(new HoodieInstant(false, HoodieTimeline.DELTA_COMMIT_ACTION, newCommitTime)).get(), HoodieCommitMetadata.class);
    assert commitMetadata.getPartitionToWriteStats().containsKey(DEFAULT_FIRST_PARTITION_PATH);
    assert commitMetadata.getPartitionToWriteStats().containsKey(DEFAULT_SECOND_PARTITION_PATH);
    List<HoodieWriteStat> hoodieWriteStatOptionList = commitMetadata.getPartitionToWriteStats().get(DEFAULT_FIRST_PARTITION_PATH);
    // Both update and insert record should enter same existing fileGroup due to small file handling
    assertEquals(1, hoodieWriteStatOptionList.size());
    assertEquals(generatedFileID, hoodieWriteStatOptionList.get(0).getFileId());
    // check insert and update numbers
    assertEquals(2, hoodieWriteStatOptionList.get(0).getNumInserts());
    assertEquals(1, hoodieWriteStatOptionList.get(0).getNumUpdateWrites());
    List<HoodieWriteStat> secondHoodieWriteStatOptionList = commitMetadata.getPartitionToWriteStats().get(DEFAULT_SECOND_PARTITION_PATH);
    // All insert should enter one fileGroup
    assertEquals(1, secondHoodieWriteStatOptionList.size());
    String fileIdInPartitionTwo = secondHoodieWriteStatOptionList.get(0).getFileId();
    assertEquals(2, hoodieWriteStatOptionList.get(0).getNumInserts());
    // Rollback
    HoodieInstant rollBackInstant = new HoodieInstant(HoodieInstant.State.INFLIGHT, HoodieTimeline.DELTA_COMMIT_ACTION, "002");
    BaseRollbackPlanActionExecutor mergeOnReadRollbackPlanActionExecutor = new BaseRollbackPlanActionExecutor(context, cfg, table, "003", rollBackInstant, false, cfg.shouldRollbackUsingMarkers());
    mergeOnReadRollbackPlanActionExecutor.execute().get();
    MergeOnReadRollbackActionExecutor mergeOnReadRollbackActionExecutor = new MergeOnReadRollbackActionExecutor(context, cfg, table, "003", rollBackInstant, true, false);
    // 3. assert the rollback stat
    Map<String, HoodieRollbackPartitionMetadata> rollbackMetadata = mergeOnReadRollbackActionExecutor.execute().getPartitionMetadata();
    assertEquals(2, rollbackMetadata.size());
    // 4. assert filegroup after rollback, and compare to the rollbackstat
    // assert the first partition data and log file size
    HoodieRollbackPartitionMetadata partitionMetadata = rollbackMetadata.get(DEFAULT_FIRST_PARTITION_PATH);
    assertTrue(partitionMetadata.getSuccessDeleteFiles().isEmpty());
    assertTrue(partitionMetadata.getFailedDeleteFiles().isEmpty());
    assertEquals(1, partitionMetadata.getRollbackLogFiles().size());
    // assert the second partition data and log file size
    partitionMetadata = rollbackMetadata.get(DEFAULT_SECOND_PARTITION_PATH);
    assertEquals(1, partitionMetadata.getSuccessDeleteFiles().size());
    assertTrue(partitionMetadata.getFailedDeleteFiles().isEmpty());
    assertTrue(partitionMetadata.getRollbackLogFiles().isEmpty());
    assertEquals(1, partitionMetadata.getSuccessDeleteFiles().size());
}

Example 62 with HoodieWriteStat

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

the class TestSparkHoodieHBaseIndex method testDelete.

@Test
public void testDelete() throws Exception {
    final String newCommitTime = "001";
    final int numRecords = 10;
    List<HoodieRecord> records = dataGen.generateInserts(newCommitTime, numRecords);
    JavaRDD<HoodieRecord> writeRecords = jsc().parallelize(records, 1);
    // Load to memory
    HoodieWriteConfig config = getConfig();
    SparkHoodieHBaseIndex index = new SparkHoodieHBaseIndex(config);
    try (SparkRDDWriteClient writeClient = getHoodieWriteClient(config)) {
        metaClient = HoodieTableMetaClient.reload(metaClient);
        HoodieTable hoodieTable = HoodieSparkTable.create(config, context, metaClient);
        // Test tagLocation without any entries in index
        JavaRDD<HoodieRecord> records1 = tagLocation(index, writeRecords, hoodieTable);
        assertEquals(0, records1.filter(record -> record.isCurrentLocationKnown()).count());
        // Insert records
        writeClient.startCommitWithTime(newCommitTime);
        JavaRDD<WriteStatus> writeStatues = writeClient.upsert(writeRecords, newCommitTime);
        assertNoWriteErrors(writeStatues.collect());
        writeClient.commit(newCommitTime, writeStatues);
        // Now tagLocation for these records, hbaseIndex should tag them correctly
        metaClient = HoodieTableMetaClient.reload(metaClient);
        hoodieTable = HoodieSparkTable.create(config, context, metaClient);
        List<HoodieRecord> records2 = tagLocation(index, writeRecords, hoodieTable).collect();
        assertEquals(numRecords, records2.stream().filter(record -> record.isCurrentLocationKnown()).count());
        assertEquals(numRecords, records2.stream().map(record -> record.getKey().getRecordKey()).distinct().count());
        assertEquals(numRecords, records2.stream().filter(record -> (record.getCurrentLocation() != null && record.getCurrentLocation().getInstantTime().equals(newCommitTime))).distinct().count());
        // Delete all records. This has to be done directly as deleting index entries
        // is not implemented via HoodieWriteClient
        JavaRDD<WriteStatus> deleteWriteStatues = writeStatues.map(w -> {
            WriteStatus newWriteStatus = new WriteStatus(true, 1.0);
            w.getWrittenRecords().forEach(r -> newWriteStatus.markSuccess(new HoodieAvroRecord(r.getKey(), null), Option.empty()));
            assertEquals(w.getTotalRecords(), newWriteStatus.getTotalRecords());
            newWriteStatus.setStat(new HoodieWriteStat());
            return newWriteStatus;
        });
        // if not for this caching, due to RDD chaining/lineage, first time update is called again when subsequent update is called.
        // So caching here to break the chain and so future update does not re-trigger update of older Rdd.
        deleteWriteStatues.cache();
        JavaRDD<WriteStatus> deleteStatus = updateLocation(index, deleteWriteStatues, hoodieTable);
        assertEquals(deleteStatus.count(), deleteWriteStatues.count());
        assertNoWriteErrors(deleteStatus.collect());
        // Ensure no records can be tagged
        List<HoodieRecord> records3 = tagLocation(index, writeRecords, hoodieTable).collect();
        assertEquals(0, records3.stream().filter(record -> record.isCurrentLocationKnown()).count());
        assertEquals(numRecords, records3.stream().map(record -> record.getKey().getRecordKey()).distinct().count());
        assertEquals(0, records3.stream().filter(record -> (record.getCurrentLocation() != null && record.getCurrentLocation().getInstantTime().equals(newCommitTime))).distinct().count());
    }
}

Example 63 with HoodieWriteStat

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

the class TestSparkHoodieHBaseIndex method getSampleWriteStatusWithFileId.

private WriteStatus getSampleWriteStatusWithFileId(final int numInserts, final int numUpdateWrites) {
    final WriteStatus writeStatus = new WriteStatus(false, 0.0);
    HoodieWriteStat hoodieWriteStat = new HoodieWriteStat();
    hoodieWriteStat.setNumInserts(numInserts);
    hoodieWriteStat.setNumUpdateWrites(numUpdateWrites);
    writeStatus.setStat(hoodieWriteStat);
    writeStatus.setFileId(UUID.randomUUID().toString());
    return writeStatus;
}

Example 64 with HoodieWriteStat

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

the class TestHoodieSparkMergeOnReadTableRollback method testInsertsGeneratedIntoLogFilesRollbackAfterCompaction.

@ParameterizedTest
@ValueSource(booleans = { true, false })
void testInsertsGeneratedIntoLogFilesRollbackAfterCompaction(boolean rollbackUsingMarkers) throws Exception {
    Properties properties = new Properties();
    properties.setProperty(HoodieTableConfig.BASE_FILE_FORMAT.key(), HoodieTableConfig.BASE_FILE_FORMAT.defaultValue().toString());
    HoodieTableMetaClient metaClient = getHoodieMetaClient(HoodieTableType.MERGE_ON_READ, properties);
    HoodieTestDataGenerator dataGen = new HoodieTestDataGenerator();
    // insert 100 records
    // Setting IndexType to be InMemory to simulate Global Index nature
    HoodieWriteConfig config = getConfigBuilder(false, rollbackUsingMarkers, HoodieIndex.IndexType.INMEMORY).build();
    try (SparkRDDWriteClient writeClient = getHoodieWriteClient(config)) {
        String newCommitTime = "100";
        writeClient.startCommitWithTime(newCommitTime);
        List<HoodieRecord> records = dataGen.generateInserts(newCommitTime, 100);
        JavaRDD<HoodieRecord> recordsRDD = jsc().parallelize(records, 1);
        JavaRDD<WriteStatus> statuses = writeClient.insert(recordsRDD, newCommitTime);
        writeClient.commit(newCommitTime, statuses);
        metaClient = HoodieTableMetaClient.reload(metaClient);
        HoodieTable table = HoodieSparkTable.create(config, context(), metaClient);
        table.getHoodieView().sync();
        TableFileSystemView.SliceView tableRTFileSystemView = table.getSliceView();
        long numLogFiles = 0;
        for (String partitionPath : dataGen.getPartitionPaths()) {
            assertTrue(tableRTFileSystemView.getLatestFileSlices(partitionPath).noneMatch(fileSlice -> fileSlice.getBaseFile().isPresent()));
            assertTrue(tableRTFileSystemView.getLatestFileSlices(partitionPath).anyMatch(fileSlice -> fileSlice.getLogFiles().count() > 0));
            numLogFiles += tableRTFileSystemView.getLatestFileSlices(partitionPath).filter(fileSlice -> fileSlice.getLogFiles().count() > 0).count();
        }
        assertTrue(numLogFiles > 0);
        // Do a compaction
        newCommitTime = writeClient.scheduleCompaction(Option.empty()).get().toString();
        HoodieWriteMetadata<JavaRDD<WriteStatus>> compactionMetadata = writeClient.compact(newCommitTime);
        statuses = compactionMetadata.getWriteStatuses();
        // Ensure all log files have been compacted into base files
        String extension = table.getBaseFileExtension();
        Collection<List<HoodieWriteStat>> stats = compactionMetadata.getCommitMetadata().get().getPartitionToWriteStats().values();
        assertEquals(numLogFiles, stats.stream().flatMap(Collection::stream).filter(state -> state.getPath().contains(extension)).count());
        assertEquals(numLogFiles, stats.stream().mapToLong(Collection::size).sum());
        // writeClient.commitCompaction(newCommitTime, statuses, Option.empty());
        // Trigger a rollback of compaction
        table.getActiveTimeline().reload();
        table.rollbackInflightCompaction(new HoodieInstant(HoodieInstant.State.INFLIGHT, HoodieTimeline.COMPACTION_ACTION, newCommitTime));
        metaClient = HoodieTableMetaClient.reload(metaClient);
        table = HoodieSparkTable.create(config, context(), metaClient);
        tableRTFileSystemView = table.getSliceView();
        ((SyncableFileSystemView) tableRTFileSystemView).reset();
        for (String partitionPath : dataGen.getPartitionPaths()) {
            List<FileSlice> fileSlices = getFileSystemViewWithUnCommittedSlices(metaClient).getAllFileSlices(partitionPath).filter(fs -> fs.getBaseInstantTime().equals("100")).collect(Collectors.toList());
            assertTrue(fileSlices.stream().noneMatch(fileSlice -> fileSlice.getBaseFile().isPresent()));
            assertTrue(fileSlices.stream().anyMatch(fileSlice -> fileSlice.getLogFiles().count() > 0));
        }
    }
}

Example 65 with HoodieWriteStat

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

the class TestTimelineUtils method getReplaceCommitMetadata.

private byte[] getReplaceCommitMetadata(String basePath, String commitTs, String replacePartition, int replaceCount, String newFilePartition, int newFileCount, Map<String, String> extraMetadata, WriteOperationType operationType) throws IOException {
    HoodieReplaceCommitMetadata commit = new HoodieReplaceCommitMetadata();
    commit.setOperationType(operationType);
    for (int i = 1; i <= newFileCount; i++) {
        HoodieWriteStat stat = new HoodieWriteStat();
        stat.setFileId(i + "");
        stat.setPartitionPath(Paths.get(basePath, newFilePartition).toString());
        stat.setPath(commitTs + "." + i + metaClient.getTableConfig().getBaseFileFormat().getFileExtension());
        commit.addWriteStat(newFilePartition, stat);
    }
    Map<String, List<String>> partitionToReplaceFileIds = new HashMap<>();
    if (replaceCount > 0) {
        partitionToReplaceFileIds.put(replacePartition, new ArrayList<>());
    }
    for (int i = 1; i <= replaceCount; i++) {
        partitionToReplaceFileIds.get(replacePartition).add(FSUtils.createNewFileIdPfx());
    }
    commit.setPartitionToReplaceFileIds(partitionToReplaceFileIds);
    for (Map.Entry<String, String> extraEntries : extraMetadata.entrySet()) {
        commit.addMetadata(extraEntries.getKey(), extraEntries.getValue());
    }
    return commit.toJsonString().getBytes(StandardCharsets.UTF_8);
}