Examples with DEFAULT_THIRD_PARTITION_PATH org.apache.hudi.common.testutils.HoodieTestDataGenerator.DEFAULT_THIRD_PARTITION_PATH used on opensource projects

Example 1 with DEFAULT_THIRD_PARTITION_PATH

use of org.apache.hudi.common.testutils.HoodieTestDataGenerator.DEFAULT_THIRD_PARTITION_PATH in project hudi by apache.

the class TestHoodieClientOnCopyOnWriteStorage method testUpsertsUpdatePartitionPath.

/**
 * This test ensures in a global bloom when update partition path is set to true in config, if an incoming record has mismatched partition
 * compared to whats in storage, then appropriate actions are taken. i.e. old record is deleted in old partition and new one is inserted
 * in the new partition.
 * test structure:
 * 1. insert 1 batch
 * 2. insert 2nd batch with larger no of records so that a new file group is created for partitions
 * 3. issue upserts to records from batch 1 with different partition path. This should ensure records from batch 1 are deleted and new
 * records are upserted to the new partition
 *
 * @param indexType index type to be tested for
 * @param config instance of {@link HoodieWriteConfig} to use
 * @param writeFn write function to be used for testing
 */
private void testUpsertsUpdatePartitionPath(IndexType indexType, HoodieWriteConfig config, Function3<JavaRDD<WriteStatus>, SparkRDDWriteClient, JavaRDD<HoodieRecord>, String> writeFn) throws Exception {
    // instantiate client
    HoodieWriteConfig hoodieWriteConfig = getConfigBuilder().withProps(config.getProps()).withCompactionConfig(HoodieCompactionConfig.newBuilder().compactionSmallFileSize(10000).build()).withIndexConfig(HoodieIndexConfig.newBuilder().withIndexType(indexType).withBloomIndexUpdatePartitionPath(true).withGlobalSimpleIndexUpdatePartitionPath(true).build()).withTimelineLayoutVersion(VERSION_0).build();
    HoodieTableMetaClient.withPropertyBuilder().fromMetaClient(metaClient).setTimelineLayoutVersion(VERSION_0).initTable(metaClient.getHadoopConf(), metaClient.getBasePath());
    // Set rollback to LAZY so no inflights are deleted
    hoodieWriteConfig.getProps().put(HoodieCompactionConfig.FAILED_WRITES_CLEANER_POLICY.key(), HoodieFailedWritesCleaningPolicy.LAZY.name());
    SparkRDDWriteClient client = getHoodieWriteClient(hoodieWriteConfig);
    // Write 1
    String newCommitTime = "001";
    int numRecords = 10;
    client.startCommitWithTime(newCommitTime);
    List<HoodieRecord> records = dataGen.generateInserts(newCommitTime, numRecords);
    Set<Pair<String, String>> expectedPartitionPathRecKeyPairs = new HashSet<>();
    // populate expected partition path and record keys
    for (HoodieRecord rec : records) {
        expectedPartitionPathRecKeyPairs.add(Pair.of(rec.getPartitionPath(), rec.getRecordKey()));
    }
    JavaRDD<HoodieRecord> writeRecords = jsc.parallelize(records, 1);
    JavaRDD<WriteStatus> result = writeFn.apply(client, writeRecords, newCommitTime);
    result.collect();
    // Check the entire dataset has all records
    String[] fullPartitionPaths = getFullPartitionPaths();
    assertPartitionPathRecordKeys(expectedPartitionPathRecKeyPairs, fullPartitionPaths);
    // verify one basefile per partition
    String[] fullExpectedPartitionPaths = getFullPartitionPaths(expectedPartitionPathRecKeyPairs.stream().map(Pair::getLeft).toArray(String[]::new));
    Map<String, Long> baseFileCounts = getBaseFileCountsForPaths(basePath, fs, fullExpectedPartitionPaths);
    for (Map.Entry<String, Long> entry : baseFileCounts.entrySet()) {
        assertEquals(1, entry.getValue());
    }
    assertTrue(baseFileCounts.entrySet().stream().allMatch(entry -> entry.getValue() == 1));
    // Write 2
    newCommitTime = "002";
    // so that a new file id is created
    numRecords = 20;
    client.startCommitWithTime(newCommitTime);
    List<HoodieRecord> recordsSecondBatch = dataGen.generateInserts(newCommitTime, numRecords);
    // populate expected partition path and record keys
    for (HoodieRecord rec : recordsSecondBatch) {
        expectedPartitionPathRecKeyPairs.add(Pair.of(rec.getPartitionPath(), rec.getRecordKey()));
    }
    writeRecords = jsc.parallelize(recordsSecondBatch, 1);
    result = writeFn.apply(client, writeRecords, newCommitTime);
    result.collect();
    // Check the entire dataset has all records
    fullPartitionPaths = getFullPartitionPaths();
    assertPartitionPathRecordKeys(expectedPartitionPathRecKeyPairs, fullPartitionPaths);
    // verify that there are more than 1 basefiles per partition
    // we can't guarantee randomness in partitions where records are distributed. So, verify atleast one partition has more than 1 basefile.
    baseFileCounts = getBaseFileCountsForPaths(basePath, fs, fullPartitionPaths);
    assertTrue(baseFileCounts.entrySet().stream().filter(entry -> entry.getValue() > 1).count() >= 1, "At least one partition should have more than 1 base file after 2nd batch of writes");
    // Write 3 (upserts to records from batch 1 with diff partition path)
    newCommitTime = "003";
    // update to diff partition paths
    List<HoodieRecord> recordsToUpsert = new ArrayList<>();
    for (HoodieRecord rec : records) {
        // remove older entry from expected partition path record key pairs
        expectedPartitionPathRecKeyPairs.remove(Pair.of(rec.getPartitionPath(), rec.getRecordKey()));
        String partitionPath = rec.getPartitionPath();
        String newPartitionPath = null;
        if (partitionPath.equalsIgnoreCase(DEFAULT_FIRST_PARTITION_PATH)) {
            newPartitionPath = DEFAULT_SECOND_PARTITION_PATH;
        } else if (partitionPath.equalsIgnoreCase(DEFAULT_SECOND_PARTITION_PATH)) {
            newPartitionPath = DEFAULT_THIRD_PARTITION_PATH;
        } else if (partitionPath.equalsIgnoreCase(DEFAULT_THIRD_PARTITION_PATH)) {
            newPartitionPath = DEFAULT_FIRST_PARTITION_PATH;
        } else {
            throw new IllegalStateException("Unknown partition path " + rec.getPartitionPath());
        }
        recordsToUpsert.add(new HoodieAvroRecord(new HoodieKey(rec.getRecordKey(), newPartitionPath), (HoodieRecordPayload) rec.getData()));
        // populate expected partition path and record keys
        expectedPartitionPathRecKeyPairs.add(Pair.of(newPartitionPath, rec.getRecordKey()));
    }
    writeRecords = jsc.parallelize(recordsToUpsert, 1);
    result = writeFn.apply(client, writeRecords, newCommitTime);
    result.collect();
    // Check the entire dataset has all records
    fullPartitionPaths = getFullPartitionPaths();
    assertPartitionPathRecordKeys(expectedPartitionPathRecKeyPairs, fullPartitionPaths);
}