Examples with HoodieClusteringConfig org.apache.hudi.config.HoodieClusteringConfig used on opensource projects

Example 11 with HoodieClusteringConfig

use of org.apache.hudi.config.HoodieClusteringConfig in project hudi by apache.

the class TestHoodieClientOnCopyOnWriteStorage method testClusteringWithFailingValidator.

@Test
public void testClusteringWithFailingValidator() throws Exception {
    // setup clustering config.
    HoodieClusteringConfig clusteringConfig = HoodieClusteringConfig.newBuilder().withClusteringMaxNumGroups(10).withClusteringSortColumns("_hoodie_record_key").withInlineClustering(true).withClusteringTargetPartitions(0).withInlineClusteringNumCommits(1).build();
    try {
        testInsertAndClustering(clusteringConfig, true, true, false, FailingPreCommitValidator.class.getName(), COUNT_SQL_QUERY_FOR_VALIDATION, "");
        fail("expected pre-commit clustering validation to fail");
    } catch (HoodieValidationException e) {
    // expected
    }
}

Example 12 with HoodieClusteringConfig

use of org.apache.hudi.config.HoodieClusteringConfig in project hudi by apache.

the class TestHoodieClientOnCopyOnWriteStorage method performClustering.

private HoodieWriteMetadata<JavaRDD<WriteStatus>> performClustering(HoodieClusteringConfig clusteringConfig, boolean populateMetaFields, boolean completeClustering, String validatorClasses, String sqlQueryForEqualityValidation, String sqlQueryForSingleResultValidation, Pair<List<HoodieRecord>, List<String>> allRecords) throws IOException {
    HoodiePreCommitValidatorConfig validatorConfig = HoodiePreCommitValidatorConfig.newBuilder().withPreCommitValidator(StringUtils.nullToEmpty(validatorClasses)).withPrecommitValidatorEqualitySqlQueries(sqlQueryForEqualityValidation).withPrecommitValidatorSingleResultSqlQueries(sqlQueryForSingleResultValidation).build();
    HoodieWriteConfig config = getConfigBuilder().withAutoCommit(false).withPreCommitValidatorConfig(validatorConfig).withProps(populateMetaFields ? new Properties() : getPropertiesForKeyGen()).withClusteringConfig(clusteringConfig).build();
    // create client with new config.
    SparkRDDWriteClient client = getHoodieWriteClient(config);
    String clusteringCommitTime = client.scheduleClustering(Option.empty()).get().toString();
    HoodieWriteMetadata<JavaRDD<WriteStatus>> clusterMetadata = client.cluster(clusteringCommitTime, completeClustering);
    if (config.isPreserveHoodieCommitMetadataForClustering() && config.populateMetaFields()) {
        verifyRecordsWrittenWithPreservedMetadata(new HashSet<>(allRecords.getRight()), allRecords.getLeft(), clusterMetadata.getWriteStatuses().collect());
    } else {
        verifyRecordsWritten(clusteringCommitTime, populateMetaFields, allRecords.getLeft(), clusterMetadata.getWriteStatuses().collect(), config);
    }
    Set<HoodieFileGroupId> replacedFileIds = new HashSet<>();
    clusterMetadata.getPartitionToReplaceFileIds().entrySet().forEach(partitionFiles -> partitionFiles.getValue().stream().forEach(file -> replacedFileIds.add(new HoodieFileGroupId(partitionFiles.getKey(), file))));
    return clusterMetadata;
}

Example 13 with HoodieClusteringConfig

use of org.apache.hudi.config.HoodieClusteringConfig in project hudi by apache.

the class TestHoodieClientOnCopyOnWriteStorage method testInflightClusteringRollbackWhenUpdatesAllowed.

@ParameterizedTest
@ValueSource(booleans = { true, false })
public void testInflightClusteringRollbackWhenUpdatesAllowed(boolean rollbackPendingClustering) throws Exception {
    // setup clustering config with update strategy to allow updates during ingestion
    HoodieClusteringConfig clusteringConfig = HoodieClusteringConfig.newBuilder().withClusteringMaxNumGroups(10).withClusteringTargetPartitions(0).withClusteringUpdatesStrategy("org.apache.hudi.client.clustering.update.strategy.SparkAllowUpdateStrategy").withRollbackPendingClustering(rollbackPendingClustering).withInlineClustering(true).withInlineClusteringNumCommits(1).build();
    // start clustering, but don't commit keep it inflight
    List<HoodieRecord> allRecords = testInsertAndClustering(clusteringConfig, true, false);
    HoodieTableMetaClient metaClient = HoodieTableMetaClient.builder().setConf(hadoopConf).setBasePath(basePath).build();
    List<Pair<HoodieInstant, HoodieClusteringPlan>> pendingClusteringPlans = ClusteringUtils.getAllPendingClusteringPlans(metaClient).collect(Collectors.toList());
    assertEquals(1, pendingClusteringPlans.size());
    HoodieInstant pendingClusteringInstant = pendingClusteringPlans.get(0).getLeft();
    assertEquals(pendingClusteringInstant.getState(), INFLIGHT);
    // make an update to a filegroup within the partition that is pending clustering
    HoodieWriteConfig.Builder cfgBuilder = getConfigBuilder(EAGER);
    addConfigsForPopulateMetaFields(cfgBuilder, true);
    cfgBuilder.withClusteringConfig(clusteringConfig);
    HoodieWriteConfig config = cfgBuilder.build();
    SparkRDDWriteClient client = getHoodieWriteClient(config);
    String commitTime = HoodieActiveTimeline.createNewInstantTime();
    allRecords.addAll(dataGen.generateUpdates(commitTime, 200));
    writeAndVerifyBatch(client, allRecords, commitTime, true);
    // verify inflight clustering was rolled back
    metaClient.reloadActiveTimeline();
    pendingClusteringPlans = ClusteringUtils.getAllPendingClusteringPlans(metaClient).collect(Collectors.toList());
    assertEquals(config.isRollbackPendingClustering() ? 0 : 1, pendingClusteringPlans.size());
}

Example 14 with HoodieClusteringConfig

use of org.apache.hudi.config.HoodieClusteringConfig in project hudi by apache.

the class TestHoodieClientOnCopyOnWriteStorage method testPendingClusteringRollback.

@Test
public void testPendingClusteringRollback() throws Exception {
    boolean populateMetaFields = true;
    // setup clustering config.
    HoodieClusteringConfig clusteringConfig = HoodieClusteringConfig.newBuilder().withClusteringMaxNumGroups(10).withClusteringTargetPartitions(0).withInlineClusteringNumCommits(1).withInlineClustering(true).build();
    // start clustering, but don't commit
    List<HoodieRecord> allRecords = testInsertAndClustering(clusteringConfig, populateMetaFields, false);
    HoodieTableMetaClient metaClient = HoodieTableMetaClient.builder().setConf(hadoopConf).setBasePath(basePath).build();
    List<Pair<HoodieInstant, HoodieClusteringPlan>> pendingClusteringPlans = ClusteringUtils.getAllPendingClusteringPlans(metaClient).collect(Collectors.toList());
    assertEquals(1, pendingClusteringPlans.size());
    HoodieInstant pendingClusteringInstant = pendingClusteringPlans.get(0).getLeft();
    // complete another commit after pending clustering
    HoodieWriteConfig.Builder cfgBuilder = getConfigBuilder(EAGER);
    addConfigsForPopulateMetaFields(cfgBuilder, populateMetaFields);
    HoodieWriteConfig config = cfgBuilder.build();
    SparkRDDWriteClient client = getHoodieWriteClient(config);
    dataGen = new HoodieTestDataGenerator();
    String commitTime = HoodieActiveTimeline.createNewInstantTime();
    allRecords.addAll(dataGen.generateInserts(commitTime, 200));
    assertThrows(HoodieUpsertException.class, () -> writeAndVerifyBatch(client, allRecords, commitTime, populateMetaFields));
    // verify pending clustering can be rolled back (even though there is a completed commit greater than pending clustering)
    client.rollback(pendingClusteringInstant.getTimestamp());
    metaClient.reloadActiveTimeline();
    // verify there are no pending clustering instants
    assertEquals(0, ClusteringUtils.getAllPendingClusteringPlans(metaClient).count());
    // delete rollback.completed instant to mimic failed rollback of clustering. and then trigger rollback of clustering again. same rollback instant should be used.
    HoodieInstant rollbackInstant = metaClient.getActiveTimeline().getRollbackTimeline().lastInstant().get();
    FileCreateUtils.deleteRollbackCommit(metaClient.getBasePath(), rollbackInstant.getTimestamp());
    metaClient.reloadActiveTimeline();
    // create replace commit requested meta file so that rollback will not throw FileNotFoundException
    // create file slice with instantTime 001 and build clustering plan including this created 001 file slice.
    HoodieClusteringPlan clusteringPlan = ClusteringTestUtils.createClusteringPlan(metaClient, pendingClusteringInstant.getTimestamp(), "1");
    // create requested replace commit
    HoodieRequestedReplaceMetadata requestedReplaceMetadata = HoodieRequestedReplaceMetadata.newBuilder().setClusteringPlan(clusteringPlan).setOperationType(WriteOperationType.CLUSTER.name()).build();
    FileCreateUtils.createRequestedReplaceCommit(metaClient.getBasePath(), pendingClusteringInstant.getTimestamp(), Option.of(requestedReplaceMetadata));
    // trigger clustering again. no new rollback instants should be generated.
    try {
        client.cluster(pendingClusteringInstant.getTimestamp(), false);
    // new replace commit metadata generated is fake one. so, clustering will fail. but the intention of test is ot check for duplicate rollback instants.
    } catch (Exception e) {
    // ignore.
    }
    metaClient.reloadActiveTimeline();
    // verify that there is no new rollback instant generated
    HoodieInstant newRollbackInstant = metaClient.getActiveTimeline().getRollbackTimeline().lastInstant().get();
    assertEquals(rollbackInstant.getTimestamp(), newRollbackInstant.getTimestamp());
}