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

Example 21 with SparkRDDWriteClient

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

the class TestHoodieSparkMergeOnReadTableRollback method testMORTableRestore.

@ParameterizedTest
@ValueSource(booleans = { true, false })
void testMORTableRestore(boolean restoreAfterCompaction) throws Exception {
    HoodieWriteConfig.Builder cfgBuilder = getConfigBuilder(false).withMarkersType(MarkerType.DIRECT.name());
    HoodieWriteConfig cfg = cfgBuilder.build();
    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);
    try (final SparkRDDWriteClient client = getHoodieWriteClient(cfg)) {
        HoodieTestDataGenerator dataGen = new HoodieTestDataGenerator();
        List<HoodieRecord> records = insertAndGetRecords("001", client, dataGen, 200);
        List<HoodieRecord> updates1 = updateAndGetRecords("002", client, dataGen, records);
        List<HoodieRecord> updates2 = updateAndGetRecords("003", client, dataGen, records);
        List<HoodieRecord> updates3 = updateAndGetRecords("004", client, dataGen, records);
        validateRecords(cfg, metaClient, updates3);
        if (!restoreAfterCompaction) {
            // restore to 002 and validate records.
            client.restoreToInstant("002");
            validateRecords(cfg, metaClient, updates1);
        } else {
            // trigger compaction and then trigger couple of upserts followed by restore.
            metaClient = HoodieTableMetaClient.reload(metaClient);
            String compactionInstantTime = "005";
            client.scheduleCompactionAtInstant(compactionInstantTime, Option.empty());
            HoodieWriteMetadata<JavaRDD<WriteStatus>> compactionMetadata = client.compact(compactionInstantTime);
            client.commitCompaction(compactionInstantTime, compactionMetadata.getCommitMetadata().get(), Option.empty());
            validateRecords(cfg, metaClient, updates3);
            List<HoodieRecord> updates4 = updateAndGetRecords("006", client, dataGen, records);
            List<HoodieRecord> updates5 = updateAndGetRecords("007", client, dataGen, records);
            validateRecords(cfg, metaClient, updates5);
            // restore to 003 and validate records.
            client.restoreToInstant("003");
            validateRecords(cfg, metaClient, updates2);
        }
    }
}

Example 22 with SparkRDDWriteClient

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

the class TestHoodieSparkMergeOnReadTableRollback method updateRecords.

private List<HoodieRecord> updateRecords(SparkRDDWriteClient client, HoodieTestDataGenerator dataGen, String commitTime, List<HoodieRecord> records, HoodieTableMetaClient metaClient, HoodieWriteConfig cfg, boolean assertLogFiles) throws IOException {
    client.startCommitWithTime(commitTime);
    records = dataGen.generateUpdates(commitTime, records);
    JavaRDD<HoodieRecord> writeRecords = jsc().parallelize(records, 1);
    List<WriteStatus> statuses = client.upsert(writeRecords, commitTime).collect();
    assertNoWriteErrors(statuses);
    if (assertLogFiles) {
        HoodieTable table = HoodieSparkTable.create(cfg, context(), metaClient);
        table.getHoodieView().sync();
        TableFileSystemView.SliceView tableRTFileSystemView = table.getSliceView();
        long numLogFiles = 0;
        for (String partitionPath : dataGen.getPartitionPaths()) {
            List<FileSlice> allSlices = tableRTFileSystemView.getLatestFileSlices(partitionPath).collect(Collectors.toList());
            assertEquals(1, allSlices.stream().filter(fileSlice -> fileSlice.getBaseFile().isPresent()).count());
            assertTrue(allSlices.stream().anyMatch(fileSlice -> fileSlice.getLogFiles().count() > 0));
            numLogFiles += allSlices.stream().filter(fileSlice -> fileSlice.getLogFiles().count() > 0).count();
        }
        assertTrue(numLogFiles > 0);
    }
    return records;
}

Example 23 with SparkRDDWriteClient

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

the class TestMarkerBasedRollbackStrategy method testCopyOnWriteRollback.

@ParameterizedTest(name = TEST_NAME_WITH_PARAMS)
@MethodSource("configParams")
public void testCopyOnWriteRollback(boolean useFileListingMetadata) throws Exception {
    HoodieWriteConfig writeConfig = getConfigBuilder().withRollbackUsingMarkers(true).withAutoCommit(false).withMetadataConfig(HoodieMetadataConfig.newBuilder().enable(useFileListingMetadata).build()).withPath(basePath).build();
    HoodieSparkEngineContext engineContext = new HoodieSparkEngineContext(jsc);
    try (SparkRDDWriteClient writeClient = new SparkRDDWriteClient(engineContext, writeConfig)) {
        // rollback 2nd commit and ensure stats reflect the info.
        List<HoodieRollbackStat> stats = testRun(useFileListingMetadata, writeConfig, writeClient);
        assertEquals(3, stats.size());
        for (HoodieRollbackStat stat : stats) {
            assertEquals(1, stat.getSuccessDeleteFiles().size());
            assertEquals(0, stat.getFailedDeleteFiles().size());
            assertEquals(0, stat.getCommandBlocksCount().size());
        }
    }
}

Example 24 with SparkRDDWriteClient

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

the class CompactionTestBase method executeCompactionWithReplacedFiles.

protected void executeCompactionWithReplacedFiles(String compactionInstantTime, SparkRDDWriteClient client, HoodieTable table, HoodieWriteConfig cfg, String[] partitions, Set<HoodieFileGroupId> replacedFileIds) throws IOException {
    client.compact(compactionInstantTime);
    List<FileSlice> fileSliceList = getCurrentLatestFileSlices(table);
    assertTrue(fileSliceList.stream().findAny().isPresent(), "Ensure latest file-slices are not empty");
    assertFalse(fileSliceList.stream().anyMatch(fs -> replacedFileIds.contains(fs.getFileGroupId())), "Compacted files should not show up in latest slices");
    // verify that there is a commit
    table = getHoodieTable(HoodieTableMetaClient.builder().setConf(hadoopConf).setBasePath(cfg.getBasePath()).setLoadActiveTimelineOnLoad(true).build(), cfg);
    HoodieTimeline timeline = table.getMetaClient().getCommitTimeline().filterCompletedInstants();
    // verify compaction commit is visible in timeline
    assertTrue(timeline.filterCompletedInstants().getInstants().filter(instant -> compactionInstantTime.equals(instant.getTimestamp())).findFirst().isPresent());
    for (String partition : partitions) {
        table.getSliceView().getLatestFileSlicesBeforeOrOn(partition, compactionInstantTime, true).forEach(fs -> {
            // verify that all log files are merged
            assertEquals(0, fs.getLogFiles().count());
            assertTrue(fs.getBaseFile().isPresent());
        });
    }
}

Example 25 with SparkRDDWriteClient

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

the class TestAsyncCompaction method testRollbackInflightIngestionWithPendingCompaction.

@Test
public void testRollbackInflightIngestionWithPendingCompaction() throws Exception {
    // Rollback inflight ingestion when there is pending compaction
    HoodieWriteConfig cfg = getConfig(false);
    String firstInstantTime = "001";
    String secondInstantTime = "004";
    String compactionInstantTime = "005";
    String inflightInstantTime = "006";
    String nextInflightInstantTime = "007";
    int numRecs = 2000;
    try (SparkRDDWriteClient client = getHoodieWriteClient(cfg)) {
        HoodieReadClient readClient = getHoodieReadClient(cfg.getBasePath());
        List<HoodieRecord> records = dataGen.generateInserts(firstInstantTime, numRecs);
        records = runNextDeltaCommits(client, readClient, Arrays.asList(firstInstantTime, secondInstantTime), records, cfg, true, new ArrayList<>());
        // Schedule compaction but do not run them
        scheduleCompaction(compactionInstantTime, client, cfg);
        HoodieTableMetaClient metaClient = HoodieTableMetaClient.builder().setConf(hadoopConf).setBasePath(cfg.getBasePath()).build();
        createNextDeltaCommit(inflightInstantTime, records, client, metaClient, cfg, true);
        metaClient = HoodieTableMetaClient.builder().setConf(hadoopConf).setBasePath(cfg.getBasePath()).build();
        HoodieInstant pendingCompactionInstant = metaClient.getActiveTimeline().filterPendingCompactionTimeline().firstInstant().get();
        assertEquals(compactionInstantTime, pendingCompactionInstant.getTimestamp(), "Pending Compaction instant has expected instant time");
        HoodieInstant inflightInstant = metaClient.getActiveTimeline().filterPendingExcludingCompaction().firstInstant().get();
        assertEquals(inflightInstantTime, inflightInstant.getTimestamp(), "inflight instant has expected instant time");
        // This should rollback
        client.startCommitWithTime(nextInflightInstantTime);
        // Validate
        metaClient = HoodieTableMetaClient.builder().setConf(hadoopConf).setBasePath(cfg.getBasePath()).build();
        inflightInstant = metaClient.getActiveTimeline().filterPendingExcludingCompaction().firstInstant().get();
        assertEquals(inflightInstant.getTimestamp(), nextInflightInstantTime, "inflight instant has expected instant time");
        assertEquals(1, metaClient.getActiveTimeline().filterPendingExcludingCompaction().getInstants().count(), "Expect only one inflight instant");
        // Expect pending Compaction to be present
        pendingCompactionInstant = metaClient.getActiveTimeline().filterPendingCompactionTimeline().firstInstant().get();
        assertEquals(compactionInstantTime, pendingCompactionInstant.getTimestamp(), "Pending Compaction instant has expected instant time");
    }
}