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

Example 66 with FileSlice

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

the class TestCopyOnWriteRollbackActionExecutor method performRollbackAndValidate.

private void performRollbackAndValidate(boolean isUsingMarkers, HoodieWriteConfig cfg, HoodieTable table, List<FileSlice> firstPartitionCommit2FileSlices, List<FileSlice> secondPartitionCommit2FileSlices) throws IOException {
    // 2. rollback
    HoodieInstant commitInstant;
    if (isUsingMarkers) {
        commitInstant = table.getActiveTimeline().getCommitTimeline().filterInflights().lastInstant().get();
    } else {
        commitInstant = table.getCompletedCommitTimeline().lastInstant().get();
    }
    BaseRollbackPlanActionExecutor copyOnWriteRollbackPlanActionExecutor = new BaseRollbackPlanActionExecutor(context, table.getConfig(), table, "003", commitInstant, false, table.getConfig().shouldRollbackUsingMarkers());
    HoodieRollbackPlan hoodieRollbackPlan = (HoodieRollbackPlan) copyOnWriteRollbackPlanActionExecutor.execute().get();
    CopyOnWriteRollbackActionExecutor copyOnWriteRollbackActionExecutor = new CopyOnWriteRollbackActionExecutor(context, cfg, table, "003", commitInstant, false, false);
    Map<String, HoodieRollbackPartitionMetadata> rollbackMetadata = copyOnWriteRollbackActionExecutor.execute().getPartitionMetadata();
    // 3. assert the rollback stat
    assertEquals(2, rollbackMetadata.size());
    for (Map.Entry<String, HoodieRollbackPartitionMetadata> entry : rollbackMetadata.entrySet()) {
        HoodieRollbackPartitionMetadata meta = entry.getValue();
        assertTrue(meta.getFailedDeleteFiles() == null || meta.getFailedDeleteFiles().size() == 0);
        assertTrue(meta.getSuccessDeleteFiles() == null || meta.getSuccessDeleteFiles().size() == 1);
    }
    // 4. assert filegroup after rollback, and compare to the rollbackstat
    // assert the first partition file group and file slice
    List<HoodieFileGroup> firstPartitionRollBack1FileGroups = table.getFileSystemView().getAllFileGroups(DEFAULT_FIRST_PARTITION_PATH).collect(Collectors.toList());
    assertEquals(1, firstPartitionRollBack1FileGroups.size());
    List<FileSlice> firstPartitionRollBack1FileSlices = firstPartitionRollBack1FileGroups.get(0).getAllFileSlices().collect(Collectors.toList());
    assertEquals(1, firstPartitionRollBack1FileSlices.size());
    firstPartitionCommit2FileSlices.removeAll(firstPartitionRollBack1FileSlices);
    assertEquals(1, firstPartitionCommit2FileSlices.size());
    assertEquals(firstPartitionCommit2FileSlices.get(0).getBaseFile().get().getPath(), this.fs.getScheme() + ":" + rollbackMetadata.get(DEFAULT_FIRST_PARTITION_PATH).getSuccessDeleteFiles().get(0));
    // assert the second partition file group and file slice
    List<HoodieFileGroup> secondPartitionRollBack1FileGroups = table.getFileSystemView().getAllFileGroups(DEFAULT_SECOND_PARTITION_PATH).collect(Collectors.toList());
    assertEquals(1, secondPartitionRollBack1FileGroups.size());
    List<FileSlice> secondPartitionRollBack1FileSlices = secondPartitionRollBack1FileGroups.get(0).getAllFileSlices().collect(Collectors.toList());
    assertEquals(1, secondPartitionRollBack1FileSlices.size());
    // assert the second partition rollback file is equals rollBack1SecondPartitionStat
    secondPartitionCommit2FileSlices.removeAll(secondPartitionRollBack1FileSlices);
    assertEquals(1, secondPartitionCommit2FileSlices.size());
    assertEquals(secondPartitionCommit2FileSlices.get(0).getBaseFile().get().getPath(), this.fs.getScheme() + ":" + rollbackMetadata.get(DEFAULT_SECOND_PARTITION_PATH).getSuccessDeleteFiles().get(0));
    assertFalse(WriteMarkersFactory.get(cfg.getMarkersType(), table, commitInstant.getTimestamp()).doesMarkerDirExist());
}

Example 67 with FileSlice

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

the class TestCopyOnWriteRollbackActionExecutor method testCopyOnWriteRollbackActionExecutor.

@ParameterizedTest
@ValueSource(booleans = { true, false })
public void testCopyOnWriteRollbackActionExecutor(boolean isUsingMarkers) throws IOException {
    // 1. prepare data and assert data result
    List<FileSlice> firstPartitionCommit2FileSlices = new ArrayList<>();
    List<FileSlice> secondPartitionCommit2FileSlices = new ArrayList<>();
    HoodieWriteConfig cfg = getConfigBuilder().withRollbackUsingMarkers(isUsingMarkers).withAutoCommit(false).build();
    this.twoUpsertCommitDataWithTwoPartitions(firstPartitionCommit2FileSlices, secondPartitionCommit2FileSlices, cfg, !isUsingMarkers);
    metaClient.reloadActiveTimeline();
    HoodieTable table = this.getHoodieTable(metaClient, cfg);
    performRollbackAndValidate(isUsingMarkers, cfg, table, firstPartitionCommit2FileSlices, secondPartitionCommit2FileSlices);
}

Example 68 with FileSlice

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

the class TestMergeOnReadRollbackActionExecutor method testMergeOnReadRollbackActionExecutor.

@ParameterizedTest
@ValueSource(booleans = { true, false })
public void testMergeOnReadRollbackActionExecutor(boolean isUsingMarkers) throws IOException {
    // 1. prepare data and assert data result
    List<FileSlice> firstPartitionCommit2FileSlices = new ArrayList<>();
    List<FileSlice> secondPartitionCommit2FileSlices = new ArrayList<>();
    HoodieWriteConfig cfg = getConfigBuilder().withRollbackUsingMarkers(isUsingMarkers).withAutoCommit(false).build();
    twoUpsertCommitDataWithTwoPartitions(firstPartitionCommit2FileSlices, secondPartitionCommit2FileSlices, cfg, !isUsingMarkers);
    List<HoodieLogFile> firstPartitionCommit2LogFiles = new ArrayList<>();
    List<HoodieLogFile> secondPartitionCommit2LogFiles = new ArrayList<>();
    firstPartitionCommit2FileSlices.get(0).getLogFiles().collect(Collectors.toList()).forEach(logFile -> firstPartitionCommit2LogFiles.add(logFile));
    assertEquals(1, firstPartitionCommit2LogFiles.size());
    secondPartitionCommit2FileSlices.get(0).getLogFiles().collect(Collectors.toList()).forEach(logFile -> secondPartitionCommit2LogFiles.add(logFile));
    assertEquals(1, secondPartitionCommit2LogFiles.size());
    HoodieTable table = this.getHoodieTable(metaClient, cfg);
    // 2. rollback
    HoodieInstant rollBackInstant = new HoodieInstant(isUsingMarkers, 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());
    for (Map.Entry<String, HoodieRollbackPartitionMetadata> entry : rollbackMetadata.entrySet()) {
        HoodieRollbackPartitionMetadata meta = entry.getValue();
        assertTrue(meta.getFailedDeleteFiles() == null || meta.getFailedDeleteFiles().size() == 0);
        assertTrue(meta.getSuccessDeleteFiles() == null || meta.getSuccessDeleteFiles().size() == 0);
    }
    // 4. assert file group after rollback, and compare to the rollbackstat
    // assert the first partition data and log file size
    List<HoodieFileGroup> firstPartitionRollBack1FileGroups = table.getFileSystemView().getAllFileGroups(DEFAULT_FIRST_PARTITION_PATH).collect(Collectors.toList());
    assertEquals(1, firstPartitionRollBack1FileGroups.size());
    List<FileSlice> firstPartitionRollBack1FileSlices = firstPartitionRollBack1FileGroups.get(0).getAllFileSlices().collect(Collectors.toList());
    assertEquals(1, firstPartitionRollBack1FileSlices.size());
    FileSlice firstPartitionRollBack1FileSlice = firstPartitionRollBack1FileSlices.get(0);
    List<HoodieLogFile> firstPartitionRollBackLogFiles = firstPartitionRollBack1FileSlice.getLogFiles().collect(Collectors.toList());
    assertEquals(2, firstPartitionRollBackLogFiles.size());
    firstPartitionRollBackLogFiles.removeAll(firstPartitionCommit2LogFiles);
    assertEquals(1, firstPartitionRollBackLogFiles.size());
    // assert the second partition data and log file size
    List<HoodieFileGroup> secondPartitionRollBack1FileGroups = table.getFileSystemView().getAllFileGroups(DEFAULT_SECOND_PARTITION_PATH).collect(Collectors.toList());
    assertEquals(1, secondPartitionRollBack1FileGroups.size());
    List<FileSlice> secondPartitionRollBack1FileSlices = secondPartitionRollBack1FileGroups.get(0).getAllFileSlices().collect(Collectors.toList());
    assertEquals(1, secondPartitionRollBack1FileSlices.size());
    FileSlice secondPartitionRollBack1FileSlice = secondPartitionRollBack1FileSlices.get(0);
    List<HoodieLogFile> secondPartitionRollBackLogFiles = secondPartitionRollBack1FileSlice.getLogFiles().collect(Collectors.toList());
    assertEquals(2, secondPartitionRollBackLogFiles.size());
    secondPartitionRollBackLogFiles.removeAll(secondPartitionCommit2LogFiles);
    assertEquals(1, secondPartitionRollBackLogFiles.size());
    assertFalse(WriteMarkersFactory.get(cfg.getMarkersType(), table, "002").doesMarkerDirExist());
}

Example 69 with FileSlice

use of org.apache.hudi.common.model.FileSlice 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 70 with FileSlice

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

the class TestCompactionAdminClient method validateUnScheduleFileId.

/**
 * Validate Unschedule operations.
 */
private void validateUnScheduleFileId(CompactionAdminClient client, String ingestionInstant, String compactionInstant, CompactionOperation op, int expNumRenames) throws Exception {
    ensureValidCompactionPlan(compactionInstant);
    // Check suggested rename operations
    List<Pair<HoodieLogFile, HoodieLogFile>> renameFiles = client.getRenamingActionsForUnschedulingCompactionOperation(metaClient, compactionInstant, op, Option.empty(), false);
    metaClient = HoodieTableMetaClient.builder().setConf(metaClient.getHadoopConf()).setBasePath(basePath).setLoadActiveTimelineOnLoad(true).build();
    // Log files belonging to file-slices created because of compaction request must be renamed
    Set<HoodieLogFile> gotLogFilesToBeRenamed = renameFiles.stream().map(Pair::getLeft).collect(Collectors.toSet());
    final HoodieTableFileSystemView fsView = new HoodieTableFileSystemView(metaClient, metaClient.getCommitsAndCompactionTimeline());
    Set<HoodieLogFile> expLogFilesToBeRenamed = fsView.getLatestFileSlices(HoodieTestUtils.DEFAULT_PARTITION_PATHS[0]).filter(fs -> fs.getBaseInstantTime().equals(compactionInstant)).filter(fs -> fs.getFileId().equals(op.getFileId())).flatMap(FileSlice::getLogFiles).collect(Collectors.toSet());
    assertEquals(expLogFilesToBeRenamed, gotLogFilesToBeRenamed, "Log files belonging to file-slices created because of compaction request must be renamed");
    validateRenameFiles(renameFiles, ingestionInstant, compactionInstant, fsView);
    Map<String, Long> fileIdToCountsBeforeRenaming = fsView.getLatestMergedFileSlicesBeforeOrOn(HoodieTestUtils.DEFAULT_PARTITION_PATHS[0], compactionInstant).filter(fs -> fs.getBaseInstantTime().equals(ingestionInstant)).filter(fs -> fs.getFileId().equals(op.getFileId())).map(fs -> Pair.of(fs.getFileId(), fs.getLogFiles().count())).collect(Collectors.toMap(Pair::getKey, Pair::getValue));
    // Call the main unschedule API
    client.unscheduleCompactionFileId(op.getFileGroupId(), false, false);
    metaClient = HoodieTableMetaClient.builder().setConf(metaClient.getHadoopConf()).setBasePath(basePath).setLoadActiveTimelineOnLoad(true).build();
    final HoodieTableFileSystemView newFsView = new HoodieTableFileSystemView(metaClient, metaClient.getCommitsAndCompactionTimeline());
    // Expect all file-slice whose base-commit is same as compaction commit to contain no new Log files
    newFsView.getLatestFileSlicesBeforeOrOn(HoodieTestUtils.DEFAULT_PARTITION_PATHS[0], compactionInstant, true).filter(fs -> fs.getBaseInstantTime().equals(compactionInstant)).filter(fs -> fs.getFileId().equals(op.getFileId())).forEach(fs -> {
        assertFalse(fs.getBaseFile().isPresent(), "No Data file must be present");
        assertEquals(0, fs.getLogFiles().count(), "No Log Files");
    });
    // Ensure same number of log-files before and after renaming per fileId
    Map<String, Long> fileIdToCountsAfterRenaming = newFsView.getAllFileGroups(HoodieTestUtils.DEFAULT_PARTITION_PATHS[0]).flatMap(HoodieFileGroup::getAllFileSlices).filter(fs -> fs.getBaseInstantTime().equals(ingestionInstant)).filter(fs -> fs.getFileId().equals(op.getFileId())).map(fs -> Pair.of(fs.getFileId(), fs.getLogFiles().count())).collect(Collectors.toMap(Pair::getKey, Pair::getValue));
    assertEquals(fileIdToCountsBeforeRenaming, fileIdToCountsAfterRenaming, "Each File Id has same number of log-files");
    assertEquals(1, fileIdToCountsAfterRenaming.size(), "Not Empty");
    assertEquals(expNumRenames, renameFiles.size(), "Expected number of renames");
}