Examples with HoodieCombineRealtimeFileSplit org.apache.hudi.hadoop.hive.HoodieCombineRealtimeFileSplit used on opensource projects

Example 1 with HoodieCombineRealtimeFileSplit

use of org.apache.hudi.hadoop.hive.HoodieCombineRealtimeFileSplit in project hudi by apache.

the class TestHoodieCombineHiveInputFormat method multiPartitionReadersRealtimeCombineHoodieInputFormat.

@Test
public void multiPartitionReadersRealtimeCombineHoodieInputFormat() throws Exception {
    // test for HUDI-1718
    Configuration conf = new Configuration();
    // initial commit
    Schema schema = HoodieAvroUtils.addMetadataFields(SchemaTestUtil.getEvolvedSchema());
    HoodieTestUtils.init(hadoopConf, tempDir.toAbsolutePath().toString(), HoodieTableType.MERGE_ON_READ);
    String commitTime = "100";
    final int numRecords = 1000;
    // Create 3 partitions, each partition holds one parquet file and 1000 records
    List<File> partitionDirs = InputFormatTestUtil.prepareMultiPartitionedParquetTable(tempDir, schema, 3, numRecords, commitTime, HoodieTableType.MERGE_ON_READ);
    HoodieCommitMetadata commitMetadata = CommitUtils.buildMetadata(Collections.emptyList(), Collections.emptyMap(), Option.empty(), WriteOperationType.UPSERT, schema.toString(), HoodieTimeline.COMMIT_ACTION);
    FileCreateUtils.createCommit(tempDir.toString(), commitTime, Option.of(commitMetadata));
    TableDesc tblDesc = Utilities.defaultTd;
    // Set the input format
    tblDesc.setInputFileFormatClass(HoodieParquetRealtimeInputFormat.class);
    LinkedHashMap<Path, PartitionDesc> pt = new LinkedHashMap<>();
    LinkedHashMap<Path, ArrayList<String>> talias = new LinkedHashMap<>();
    PartitionDesc partDesc = new PartitionDesc(tblDesc, null);
    pt.put(new Path(tempDir.toAbsolutePath().toString()), partDesc);
    ArrayList<String> arrayList = new ArrayList<>();
    arrayList.add(tempDir.toAbsolutePath().toString());
    talias.put(new Path(tempDir.toAbsolutePath().toString()), arrayList);
    MapredWork mrwork = new MapredWork();
    mrwork.getMapWork().setPathToPartitionInfo(pt);
    mrwork.getMapWork().setPathToAliases(talias);
    Path mapWorkPath = new Path(tempDir.toAbsolutePath().toString());
    Utilities.setMapRedWork(conf, mrwork, mapWorkPath);
    jobConf = new JobConf(conf);
    // Add three partition path to InputPaths
    Path[] partitionDirArray = new Path[partitionDirs.size()];
    partitionDirs.stream().map(p -> new Path(p.getPath())).collect(Collectors.toList()).toArray(partitionDirArray);
    FileInputFormat.setInputPaths(jobConf, partitionDirArray);
    jobConf.set(HAS_MAP_WORK, "true");
    // The following config tells Hive to choose ExecMapper to read the MAP_WORK
    jobConf.set(MAPRED_MAPPER_CLASS, ExecMapper.class.getName());
    // setting the split size to be 3 to create one split for 3 file groups
    jobConf.set(org.apache.hadoop.mapreduce.lib.input.FileInputFormat.SPLIT_MAXSIZE, "128000000");
    HoodieCombineHiveInputFormat combineHiveInputFormat = new HoodieCombineHiveInputFormat();
    String tripsHiveColumnTypes = "double,string,string,string,double,double,double,double,double";
    InputFormatTestUtil.setPropsForInputFormat(jobConf, schema, tripsHiveColumnTypes);
    InputSplit[] splits = combineHiveInputFormat.getSplits(jobConf, 1);
    // Since the SPLIT_SIZE is 3, we should create only 1 split with all 3 file groups
    assertEquals(1, splits.length);
    RecordReader<NullWritable, ArrayWritable> recordReader = combineHiveInputFormat.getRecordReader(splits[0], jobConf, null);
    NullWritable nullWritable = recordReader.createKey();
    ArrayWritable arrayWritable = recordReader.createValue();
    int counter = 0;
    HoodieCombineRealtimeHiveSplit hiveSplit = (HoodieCombineRealtimeHiveSplit) splits[0];
    HoodieCombineRealtimeFileSplit fileSplit = (HoodieCombineRealtimeFileSplit) hiveSplit.getInputSplitShim();
    List<FileSplit> realtimeFileSplits = fileSplit.getRealtimeFileSplits();
    while (recordReader.next(nullWritable, arrayWritable)) {
        // Hive use ioctx to extract partition info, when switch reader, ioctx should be updated.
        if (counter < 1000) {
            assertEquals(IOContextMap.get(jobConf).getInputPath().toString(), realtimeFileSplits.get(0).getPath().toString());
        } else if (counter < 2000) {
            assertEquals(IOContextMap.get(jobConf).getInputPath().toString(), realtimeFileSplits.get(1).getPath().toString());
        } else {
            assertEquals(IOContextMap.get(jobConf).getInputPath().toString(), realtimeFileSplits.get(2).getPath().toString());
        }
        counter++;
    }
    // should read out 3 splits, each for file0, file1, file2 containing 1000 records each
    assertEquals(3000, counter);
    recordReader.close();
}