Examples with BaseFileUtils org.apache.hudi.common.util.BaseFileUtils used on opensource projects

Example 6 with BaseFileUtils

use of org.apache.hudi.common.util.BaseFileUtils in project hudi by apache.

the class TestHoodieClientOnCopyOnWriteStorage method testSmallInsertHandlingForUpserts.

/**
 * Test scenario of new file-group getting added during upsert().
 */
@Test
public void testSmallInsertHandlingForUpserts() throws Exception {
    final String testPartitionPath = "2016/09/26";
    final int insertSplitLimit = 100;
    // setup the small file handling params
    // hold upto 200 records max
    HoodieWriteConfig config = getSmallInsertWriteConfig(insertSplitLimit, TRIP_EXAMPLE_SCHEMA, dataGen.getEstimatedFileSizeInBytes(150));
    dataGen = new HoodieTestDataGenerator(new String[] { testPartitionPath });
    SparkRDDWriteClient client = getHoodieWriteClient(config);
    BaseFileUtils fileUtils = BaseFileUtils.getInstance(metaClient);
    // Inserts => will write file1
    String commitTime1 = "001";
    client.startCommitWithTime(commitTime1);
    // this writes ~500kb
    List<HoodieRecord> inserts1 = dataGen.generateInserts(commitTime1, insertSplitLimit);
    Set<String> keys1 = recordsToRecordKeySet(inserts1);
    JavaRDD<HoodieRecord> insertRecordsRDD1 = jsc.parallelize(inserts1, 1);
    List<WriteStatus> statuses = client.upsert(insertRecordsRDD1, commitTime1).collect();
    assertNoWriteErrors(statuses);
    assertEquals(1, statuses.size(), "Just 1 file needs to be added.");
    String file1 = statuses.get(0).getFileId();
    assertEquals(100, fileUtils.readRowKeys(hadoopConf, new Path(basePath, statuses.get(0).getStat().getPath())).size(), "file should contain 100 records");
    // Update + Inserts such that they just expand file1
    String commitTime2 = "002";
    client.startCommitWithTime(commitTime2);
    List<HoodieRecord> inserts2 = dataGen.generateInserts(commitTime2, 40);
    Set<String> keys2 = recordsToRecordKeySet(inserts2);
    List<HoodieRecord> insertsAndUpdates2 = new ArrayList<>();
    insertsAndUpdates2.addAll(inserts2);
    insertsAndUpdates2.addAll(dataGen.generateUpdates(commitTime2, inserts1));
    JavaRDD<HoodieRecord> insertAndUpdatesRDD2 = jsc.parallelize(insertsAndUpdates2, 1);
    statuses = client.upsert(insertAndUpdatesRDD2, commitTime2).collect();
    assertNoWriteErrors(statuses);
    assertEquals(1, statuses.size(), "Just 1 file needs to be updated.");
    assertEquals(file1, statuses.get(0).getFileId(), "Existing file should be expanded");
    assertEquals(commitTime1, statuses.get(0).getStat().getPrevCommit(), "Existing file should be expanded");
    Path newFile = new Path(basePath, statuses.get(0).getStat().getPath());
    assertEquals(140, fileUtils.readRowKeys(hadoopConf, newFile).size(), "file should contain 140 records");
    List<GenericRecord> records = fileUtils.readAvroRecords(hadoopConf, newFile);
    for (GenericRecord record : records) {
        String recordKey = record.get(HoodieRecord.RECORD_KEY_METADATA_FIELD).toString();
        assertEquals(commitTime2, record.get(HoodieRecord.COMMIT_TIME_METADATA_FIELD).toString(), "only expect commit2");
        assertTrue(keys2.contains(recordKey) || keys1.contains(recordKey), "key expected to be part of commit2");
    }
    // update + inserts such that file1 is updated and expanded, a new file2 is created.
    String commitTime3 = "003";
    client.startCommitWithTime(commitTime3);
    List<HoodieRecord> insertsAndUpdates3 = dataGen.generateInserts(commitTime3, 200);
    Set<String> keys3 = recordsToRecordKeySet(insertsAndUpdates3);
    List<HoodieRecord> updates3 = dataGen.generateUpdates(commitTime3, inserts2);
    insertsAndUpdates3.addAll(updates3);
    JavaRDD<HoodieRecord> insertAndUpdatesRDD3 = jsc.parallelize(insertsAndUpdates3, 1);
    statuses = client.upsert(insertAndUpdatesRDD3, commitTime3).collect();
    assertNoWriteErrors(statuses);
    assertEquals(2, statuses.size(), "2 files needs to be committed.");
    HoodieTableMetaClient metadata = HoodieTableMetaClient.builder().setConf(hadoopConf).setBasePath(basePath).build();
    HoodieTable table = getHoodieTable(metadata, config);
    BaseFileOnlyView fileSystemView = table.getBaseFileOnlyView();
    List<HoodieBaseFile> files = fileSystemView.getLatestBaseFilesBeforeOrOn(testPartitionPath, commitTime3).collect(Collectors.toList());
    int numTotalInsertsInCommit3 = 0;
    int numTotalUpdatesInCommit3 = 0;
    for (HoodieBaseFile file : files) {
        if (file.getFileName().contains(file1)) {
            assertEquals(commitTime3, file.getCommitTime(), "Existing file should be expanded");
            records = fileUtils.readAvroRecords(hadoopConf, new Path(file.getPath()));
            for (GenericRecord record : records) {
                String recordKey = record.get(HoodieRecord.RECORD_KEY_METADATA_FIELD).toString();
                String recordCommitTime = record.get(HoodieRecord.COMMIT_TIME_METADATA_FIELD).toString();
                if (recordCommitTime.equals(commitTime3)) {
                    if (keys2.contains(recordKey)) {
                        keys2.remove(recordKey);
                        numTotalUpdatesInCommit3++;
                    } else {
                        numTotalInsertsInCommit3++;
                    }
                }
            }
            assertEquals(0, keys2.size(), "All keys added in commit 2 must be updated in commit3 correctly");
        } else {
            assertEquals(commitTime3, file.getCommitTime(), "New file must be written for commit 3");
            records = fileUtils.readAvroRecords(hadoopConf, new Path(file.getPath()));
            for (GenericRecord record : records) {
                String recordKey = record.get(HoodieRecord.RECORD_KEY_METADATA_FIELD).toString();
                assertEquals(commitTime3, record.get(HoodieRecord.COMMIT_TIME_METADATA_FIELD).toString(), "only expect commit3");
                assertTrue(keys3.contains(recordKey), "key expected to be part of commit3");
            }
            numTotalInsertsInCommit3 += records.size();
        }
    }
    assertEquals(numTotalUpdatesInCommit3, inserts2.size(), "Total updates in commit3 must add up");
    assertEquals(numTotalInsertsInCommit3, keys3.size(), "Total inserts in commit3 must add up");
}

Example 7 with BaseFileUtils

use of org.apache.hudi.common.util.BaseFileUtils in project hudi by apache.

the class TestJavaCopyOnWriteActionExecutor method testUpdateRecords.

@Test
public void testUpdateRecords() throws Exception {
    // Prepare the AvroParquetIO
    HoodieWriteConfig config = makeHoodieClientConfig();
    int startInstant = 1;
    String firstCommitTime = makeNewCommitTime(startInstant++);
    HoodieJavaWriteClient writeClient = getHoodieWriteClient(config);
    writeClient.startCommitWithTime(firstCommitTime);
    metaClient = HoodieTableMetaClient.reload(metaClient);
    BaseFileUtils fileUtils = BaseFileUtils.getInstance(metaClient);
    String partitionPath = "2016/01/31";
    // Get some records belong to the same partition (2016/01/31)
    String recordStr1 = "{\"_row_key\":\"8eb5b87a-1feh-4edd-87b4-6ec96dc405a0\"," + "\"time\":\"2016-01-31T03:16:41.415Z\",\"number\":12}";
    String recordStr2 = "{\"_row_key\":\"8eb5b87b-1feu-4edd-87b4-6ec96dc405a0\"," + "\"time\":\"2016-01-31T03:20:41.415Z\",\"number\":100}";
    String recordStr3 = "{\"_row_key\":\"8eb5b87c-1fej-4edd-87b4-6ec96dc405a0\"," + "\"time\":\"2016-01-31T03:16:41.415Z\",\"number\":15}";
    String recordStr4 = "{\"_row_key\":\"8eb5b87d-1fej-4edd-87b4-6ec96dc405a0\"," + "\"time\":\"2016-01-31T03:16:41.415Z\",\"number\":51}";
    List<HoodieRecord> records = new ArrayList<>();
    RawTripTestPayload rowChange1 = new RawTripTestPayload(recordStr1);
    records.add(new HoodieAvroRecord(new HoodieKey(rowChange1.getRowKey(), rowChange1.getPartitionPath()), rowChange1));
    RawTripTestPayload rowChange2 = new RawTripTestPayload(recordStr2);
    records.add(new HoodieAvroRecord(new HoodieKey(rowChange2.getRowKey(), rowChange2.getPartitionPath()), rowChange2));
    RawTripTestPayload rowChange3 = new RawTripTestPayload(recordStr3);
    records.add(new HoodieAvroRecord(new HoodieKey(rowChange3.getRowKey(), rowChange3.getPartitionPath()), rowChange3));
    // Insert new records
    writeClient.insert(records, firstCommitTime);
    FileStatus[] allFiles = getIncrementalFiles(partitionPath, "0", -1);
    assertEquals(1, allFiles.length);
    // Read out the bloom filter and make sure filter can answer record exist or not
    Path filePath = allFiles[0].getPath();
    BloomFilter filter = fileUtils.readBloomFilterFromMetadata(hadoopConf, filePath);
    for (HoodieRecord record : records) {
        assertTrue(filter.mightContain(record.getRecordKey()));
    }
    // Read the base file, check the record content
    List<GenericRecord> fileRecords = fileUtils.readAvroRecords(hadoopConf, filePath);
    GenericRecord newRecord;
    int index = 0;
    for (GenericRecord record : fileRecords) {
        // System.out.println("Got :" + record.get("_row_key").toString() + ", Exp :" + records.get(index).getRecordKey());
        assertEquals(records.get(index).getRecordKey(), record.get("_row_key").toString());
        index++;
    }
    // We update the 1st record & add a new record
    String updateRecordStr1 = "{\"_row_key\":\"8eb5b87a-1feh-4edd-87b4-6ec96dc405a0\"," + "\"time\":\"2016-01-31T03:16:41.415Z\",\"number\":15}";
    RawTripTestPayload updateRowChanges1 = new RawTripTestPayload(updateRecordStr1);
    HoodieRecord updatedRecord1 = new HoodieAvroRecord(new HoodieKey(updateRowChanges1.getRowKey(), updateRowChanges1.getPartitionPath()), updateRowChanges1);
    RawTripTestPayload rowChange4 = new RawTripTestPayload(recordStr4);
    HoodieRecord insertedRecord1 = new HoodieAvroRecord(new HoodieKey(rowChange4.getRowKey(), rowChange4.getPartitionPath()), rowChange4);
    List<HoodieRecord> updatedRecords = Arrays.asList(updatedRecord1, insertedRecord1);
    String newCommitTime = makeNewCommitTime(startInstant++);
    metaClient = HoodieTableMetaClient.reload(metaClient);
    writeClient.startCommitWithTime(newCommitTime);
    List<WriteStatus> statuses = writeClient.upsert(updatedRecords, newCommitTime);
    allFiles = getIncrementalFiles(partitionPath, firstCommitTime, -1);
    assertEquals(1, allFiles.length);
    // verify new incremental file group is same as the previous one
    assertEquals(FSUtils.getFileId(filePath.getName()), FSUtils.getFileId(allFiles[0].getPath().getName()));
    // Check whether the record has been updated
    Path updatedFilePath = allFiles[0].getPath();
    BloomFilter updatedFilter = fileUtils.readBloomFilterFromMetadata(hadoopConf, updatedFilePath);
    for (HoodieRecord record : records) {
        // No change to the _row_key
        assertTrue(updatedFilter.mightContain(record.getRecordKey()));
    }
    assertTrue(updatedFilter.mightContain(insertedRecord1.getRecordKey()));
    // add this so it can further check below
    records.add(insertedRecord1);
    ParquetReader updatedReader = ParquetReader.builder(new AvroReadSupport<>(), updatedFilePath).build();
    index = 0;
    while ((newRecord = (GenericRecord) updatedReader.read()) != null) {
        assertEquals(newRecord.get("_row_key").toString(), records.get(index).getRecordKey());
        if (index == 0) {
            assertEquals("15", newRecord.get("number").toString());
        }
        index++;
    }
    updatedReader.close();
    // Also check the numRecordsWritten
    WriteStatus writeStatus = statuses.get(0);
    assertEquals(1, statuses.size(), "Should be only one file generated");
    // 3 rewritten records + 1 new record
    assertEquals(4, writeStatus.getStat().getNumWrites());
}