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

Example 26 with SparkRDDWriteClient

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

the class TestAsyncCompaction method testRollbackForInflightCompaction.

@Test
public void testRollbackForInflightCompaction() throws Exception {
    // Rollback inflight compaction
    HoodieWriteConfig cfg = getConfig(false);
    try (SparkRDDWriteClient client = getHoodieWriteClient(cfg)) {
        HoodieReadClient readClient = getHoodieReadClient(cfg.getBasePath());
        String firstInstantTime = "001";
        String secondInstantTime = "004";
        String compactionInstantTime = "005";
        int numRecs = 2000;
        List<HoodieRecord> records = dataGen.generateInserts(firstInstantTime, numRecs);
        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();
        HoodieInstant pendingCompactionInstant = metaClient.getActiveTimeline().filterPendingCompactionTimeline().firstInstant().get();
        assertEquals(compactionInstantTime, pendingCompactionInstant.getTimestamp(), "Pending Compaction instant has expected instant time");
        assertEquals(State.REQUESTED, pendingCompactionInstant.getState(), "Pending Compaction instant has expected state");
        moveCompactionFromRequestedToInflight(compactionInstantTime, cfg);
        // Reload and rollback inflight compaction
        metaClient = HoodieTableMetaClient.builder().setConf(hadoopConf).setBasePath(cfg.getBasePath()).build();
        HoodieTable hoodieTable = HoodieSparkTable.create(cfg, context, metaClient);
        hoodieTable.rollbackInflightCompaction(new HoodieInstant(State.INFLIGHT, HoodieTimeline.COMPACTION_ACTION, compactionInstantTime));
        metaClient = HoodieTableMetaClient.builder().setConf(hadoopConf).setBasePath(cfg.getBasePath()).build();
        pendingCompactionInstant = metaClient.getCommitsAndCompactionTimeline().filterPendingCompactionTimeline().getInstants().findFirst().get();
        assertEquals("compaction", pendingCompactionInstant.getAction());
        assertEquals(State.REQUESTED, pendingCompactionInstant.getState());
        assertEquals(compactionInstantTime, pendingCompactionInstant.getTimestamp());
        // We indirectly test for the race condition where a inflight instant was first deleted then created new. Every
        // time this happens, the pending compaction instant file in Hoodie Meta path becomes an empty file (Note: Hoodie
        // reads compaction plan from aux path which is untouched). TO test for regression, we simply get file status
        // and look at the file size
        FileStatus fstatus = metaClient.getFs().getFileStatus(new Path(metaClient.getMetaPath(), pendingCompactionInstant.getFileName()));
        assertTrue(fstatus.getLen() > 0);
    }
}

Example 27 with SparkRDDWriteClient

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

the class TestAsyncCompaction method testInflightCompaction.

@Test
public void testInflightCompaction() throws Exception {
    // There is inflight compaction. Subsequent compaction run must work correctly
    HoodieWriteConfig cfg = getConfig(true);
    try (SparkRDDWriteClient client = getHoodieWriteClient(cfg)) {
        HoodieReadClient readClient = getHoodieReadClient(cfg.getBasePath());
        String firstInstantTime = "001";
        String secondInstantTime = "004";
        String compactionInstantTime = "005";
        String thirdInstantTime = "006";
        String fourthInstantTime = "007";
        int numRecs = 2000;
        List<HoodieRecord> records = dataGen.generateInserts(firstInstantTime, numRecs);
        records = runNextDeltaCommits(client, readClient, Arrays.asList(firstInstantTime, secondInstantTime), records, cfg, true, new ArrayList<>());
        // Schedule and mark compaction instant as inflight
        HoodieTableMetaClient metaClient = HoodieTableMetaClient.builder().setConf(hadoopConf).setBasePath(cfg.getBasePath()).build();
        HoodieTable hoodieTable = getHoodieTable(metaClient, cfg);
        scheduleCompaction(compactionInstantTime, client, cfg);
        moveCompactionFromRequestedToInflight(compactionInstantTime, cfg);
        // Complete ingestions
        runNextDeltaCommits(client, readClient, Arrays.asList(thirdInstantTime, fourthInstantTime), records, cfg, false, Arrays.asList(compactionInstantTime));
        // execute inflight compaction
        executeCompaction(compactionInstantTime, client, hoodieTable, cfg, numRecs, true);
    }
}

Example 28 with SparkRDDWriteClient

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

the class TestAsyncCompaction method testCompactionOnReplacedFiles.

@Test
public void testCompactionOnReplacedFiles() throws Exception {
    // Schedule a compaction. Replace those file groups and ensure compaction completes successfully.
    HoodieWriteConfig cfg = getConfig(true);
    try (SparkRDDWriteClient client = getHoodieWriteClient(cfg)) {
        HoodieReadClient readClient = getHoodieReadClient(cfg.getBasePath());
        String firstInstantTime = "001";
        String secondInstantTime = "004";
        String compactionInstantTime = "005";
        String replaceInstantTime = "006";
        String fourthInstantTime = "007";
        int numRecs = 2000;
        List<HoodieRecord> records = dataGen.generateInserts(firstInstantTime, numRecs);
        runNextDeltaCommits(client, readClient, Arrays.asList(firstInstantTime, secondInstantTime), records, cfg, true, new ArrayList<>());
        HoodieTableMetaClient metaClient = HoodieTableMetaClient.builder().setConf(hadoopConf).setBasePath(cfg.getBasePath()).build();
        HoodieTable hoodieTable = getHoodieTable(metaClient, cfg);
        scheduleCompaction(compactionInstantTime, client, cfg);
        metaClient.reloadActiveTimeline();
        HoodieInstant pendingCompactionInstant = metaClient.getActiveTimeline().filterPendingCompactionTimeline().firstInstant().get();
        assertEquals(compactionInstantTime, pendingCompactionInstant.getTimestamp(), "Pending Compaction instant has expected instant time");
        Set<HoodieFileGroupId> fileGroupsBeforeReplace = getAllFileGroups(hoodieTable, dataGen.getPartitionPaths());
        // replace by using insertOverwrite
        JavaRDD<HoodieRecord> replaceRecords = jsc.parallelize(dataGen.generateInserts(replaceInstantTime, numRecs), 1);
        client.startCommitWithTime(replaceInstantTime, HoodieTimeline.REPLACE_COMMIT_ACTION);
        client.insertOverwrite(replaceRecords, replaceInstantTime);
        metaClient.reloadActiveTimeline();
        hoodieTable = getHoodieTable(metaClient, cfg);
        Set<HoodieFileGroupId> newFileGroups = getAllFileGroups(hoodieTable, dataGen.getPartitionPaths());
        // make sure earlier file groups are not visible
        assertEquals(0, newFileGroups.stream().filter(fg -> fileGroupsBeforeReplace.contains(fg)).count());
        // compaction should run with associated file groups are replaced
        executeCompactionWithReplacedFiles(compactionInstantTime, client, hoodieTable, cfg, dataGen.getPartitionPaths(), fileGroupsBeforeReplace);
    }
}

Example 29 with SparkRDDWriteClient

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

the class TestAsyncCompaction method testScheduleCompactionAfterPendingIngestion.

@Test
public void testScheduleCompactionAfterPendingIngestion() throws Exception {
    // Case: Failure case. Earliest ingestion inflight instant time must be later than compaction time
    HoodieWriteConfig cfg = getConfig(false);
    SparkRDDWriteClient client = getHoodieWriteClient(cfg);
    HoodieReadClient readClient = getHoodieReadClient(cfg.getBasePath());
    String firstInstantTime = "001";
    String secondInstantTime = "004";
    String inflightInstantTime = "005";
    String compactionInstantTime = "006";
    int numRecs = 2000;
    List<HoodieRecord> records = dataGen.generateInserts(firstInstantTime, numRecs);
    records = runNextDeltaCommits(client, readClient, Arrays.asList(firstInstantTime, secondInstantTime), records, cfg, true, new ArrayList<>());
    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 inflightInstant = metaClient.getActiveTimeline().filterPendingExcludingCompaction().firstInstant().get();
    assertEquals(inflightInstantTime, inflightInstant.getTimestamp(), "inflight instant has expected instant time");
    assertThrows(IllegalArgumentException.class, () -> {
        // Schedule compaction but do not run them
        scheduleCompaction(compactionInstantTime, client, cfg);
    }, "Earliest ingestion inflight instant time must be later than compaction time");
}

Example 30 with SparkRDDWriteClient

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

the class TestHoodieCompactor method testWriteStatusContentsAfterCompaction.

@Test
public void testWriteStatusContentsAfterCompaction() throws Exception {
    // insert 100 records
    HoodieWriteConfig config = getConfigBuilder().withCompactionConfig(HoodieCompactionConfig.newBuilder().withMaxNumDeltaCommitsBeforeCompaction(1).build()).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);
        writeClient.insert(recordsRDD, newCommitTime).collect();
        // Update all the 100 records
        HoodieTable table = HoodieSparkTable.create(config, context);
        newCommitTime = "101";
        List<HoodieRecord> updatedRecords = dataGen.generateUpdates(newCommitTime, records);
        JavaRDD<HoodieRecord> updatedRecordsRDD = jsc.parallelize(updatedRecords, 1);
        HoodieIndex index = new HoodieBloomIndex(config, SparkHoodieBloomIndexHelper.getInstance());
        JavaRDD<HoodieRecord> updatedTaggedRecordsRDD = tagLocation(index, updatedRecordsRDD, table);
        writeClient.startCommitWithTime(newCommitTime);
        writeClient.upsertPreppedRecords(updatedTaggedRecordsRDD, newCommitTime).collect();
        metaClient.reloadActiveTimeline();
        // Verify that all data file has one log file
        table = HoodieSparkTable.create(config, context);
        for (String partitionPath : dataGen.getPartitionPaths()) {
            List<FileSlice> groupedLogFiles = table.getSliceView().getLatestFileSlices(partitionPath).collect(Collectors.toList());
            for (FileSlice fileSlice : groupedLogFiles) {
                assertEquals(1, fileSlice.getLogFiles().count(), "There should be 1 log file written for every data file");
            }
        }
        // Do a compaction
        table = HoodieSparkTable.create(config, context);
        String compactionInstantTime = "102";
        table.scheduleCompaction(context, compactionInstantTime, Option.empty());
        table.getMetaClient().reloadActiveTimeline();
        HoodieData<WriteStatus> result = (HoodieData<WriteStatus>) table.compact(context, compactionInstantTime).getWriteStatuses();
        // Verify that all partition paths are present in the WriteStatus result
        for (String partitionPath : dataGen.getPartitionPaths()) {
            List<WriteStatus> writeStatuses = result.collectAsList();
            assertTrue(writeStatuses.stream().filter(writeStatus -> writeStatus.getStat().getPartitionPath().contentEquals(partitionPath)).count() > 0);
        }
    }
}