Ejemplo n.º 1
0
  @Override
  public void indexRow(GenericRow row) {
    // Find matching leaves in StarTree for row
    currentMatchingNodes.clear();
    StarTreeTableRow tableRow = extractValues(row);
    findMatchingLeaves(starTreeBuilder.getTree(), tableRow.getDimensions(), currentMatchingNodes);

    // Only write the raw value, maintaining sort order (we will write aggregates when sealing)
    for (StarTreeIndexNode node : currentMatchingNodes) {
      Map<Integer, Integer> pathValues = node.getPathValues();
      if (!pathValues.containsValue(StarTreeIndexNode.all())) {
        StarTreeTableRange range = starTreeBuilder.getDocumentIdRange(node.getNodeId());
        StarTreeTable subTable =
            starTreeBuilder.getTable().view(range.getStartDocumentId(), range.getDocumentCount());

        Integer nextMatchingDocumentId =
            starTreeBuilder.getNextDocumentId(tableRow.getDimensions());
        if (nextMatchingDocumentId == null) {
          throw new IllegalStateException("Could not assign document ID for row " + tableRow);
        }

        // Write using that document ID to all columns
        for (final String column : dictionaryCreatorMap.keySet()) {
          Object columnValueToIndex = row.getValue(column);
          if (schema.getFieldSpecFor(column).isSingleValueField()) {
            System.out.println(column + ": " + columnValueToIndex);
            int dictionaryIndex = dictionaryCreatorMap.get(column).indexOfSV(columnValueToIndex);
            ((SingleValueForwardIndexCreator) forwardIndexCreatorMap.get(column))
                .index(nextMatchingDocumentId, dictionaryIndex);
            if (config.createInvertedIndexEnabled()) {
              invertedIndexCreatorMap
                  .get(column)
                  .add(nextMatchingDocumentId, (Object) dictionaryIndex);
            }
          } else {
            int[] dictionaryIndex = dictionaryCreatorMap.get(column).indexOfMV(columnValueToIndex);
            ((MultiValueForwardIndexCreator) forwardIndexCreatorMap.get(column))
                .index(nextMatchingDocumentId, dictionaryIndex);
            if (config.createInvertedIndexEnabled()) {
              invertedIndexCreatorMap.get(column).add(nextMatchingDocumentId, dictionaryIndex);
            }
          }
        }
      }
    }
  }
Ejemplo n.º 2
0
  /**
   * Returns the unique metric values for each column.
   *
   * <p>The original unique values cannot be used because after aggregation, we almost certainly
   * have new values to encode that were not present in the original data set.
   */
  private Map<String, Set<Object>> computeUniqueMetricValues() {
    Map<String, Set<Object>> uniqueMetricValues = new HashMap<String, Set<Object>>();

    Iterator<StarTreeTableRow> tableIterator = starTreeBuilder.getTable().getAllCombinations();
    while (tableIterator.hasNext()) {
      StarTreeTableRow row = tableIterator.next();

      for (int i = 0; i < schema.getMetricNames().size(); i++) {
        String metricName = schema.getMetricNames().get(i);
        Object metricValue = row.getMetrics().get(i);
        Set<Object> uniqueValues = uniqueMetricValues.get(metricName);
        if (uniqueValues == null) {
          uniqueValues = new HashSet<Object>();
          uniqueMetricValues.put(metricName, uniqueValues);
        }
        uniqueValues.add(metricValue);
      }
    }

    return uniqueMetricValues;
  }
Ejemplo n.º 3
0
  @Override
  public void init(
      SegmentGeneratorConfig config,
      Map<String, ColumnIndexCreationInfo> columnInfo,
      Schema schema,
      int totalDocs,
      File outDir)
      throws Exception {
    // Member variables
    this.config = config;
    this.columnInfo = columnInfo;
    this.schema = schema;
    this.outDir = outDir;
    this.starTreeDimensionDictionary = new HashMap<String, Integer>();
    this.starTreeMetricDictionary = new HashMap<String, Integer>();

    // Dictionaries (will go in root segment)
    initializeAndBuildDictionaries(schema, columnInfo, outDir);

    // Compute dimension dictionary
    for (int i = 0; i < schema.getDimensionNames().size(); i++) {
      starTreeDimensionDictionary.put(schema.getDimensionNames().get(i), i);
    }
    LOG.info("StarTree dimension dictionary: {}", starTreeDimensionDictionary);

    // Compute the metric dictionary
    for (int i = 0; i < schema.getMetricNames().size(); i++) {
      starTreeMetricDictionary.put(schema.getMetricNames().get(i), i);
    }
    LOG.info("StarTree metric dictionary: {}", starTreeDimensionDictionary);

    // Compute StarTree split order
    splitOrder = computeSplitOrder(columnInfo);
    LOG.info("Computed split order {}", splitOrder);
    List<Integer> splitOrderIndexes = new ArrayList<Integer>();
    for (String dimensionName : splitOrder) {
      Integer dimensionId = starTreeDimensionDictionary.get(dimensionName);
      splitOrderIndexes.add(dimensionId);
    }
    Collections.reverse(splitOrderIndexes);

    // StarTree builder / table
    StarTreeTable table = new LinkedListStarTreeTable(); // TODO: ByteBuffer-based
    StarTreeDocumentIdMap documentIdMap =
        new HashMapStarTreeDocumentIdMap(); // TODO: ByteBuffer-based
    starTreeBuilder.init(
        splitOrderIndexes, starTreeIndexSpec.getMaxLeafRecords(), table, documentIdMap);

    // Build the StarTree structure and table
    LOG.info("Building StarTree table...");
    int count = 0;
    long startMillis = System.currentTimeMillis();
    recordReader.rewind();
    while (recordReader.hasNext()) {
      GenericRow row = recordReader.next();
      StarTreeTableRow starTreeTableRow = extractValues(row);
      starTreeBuilder.append(starTreeTableRow);
      count++;
    }
    long endMillis = System.currentTimeMillis();
    LOG.info(
        "Finished building StarTree table ({} documents, took {} ms)",
        count,
        endMillis - startMillis);

    LOG.info("Building StarTree (computing aggregates)...");
    startMillis = System.currentTimeMillis();
    starTreeBuilder.build();
    endMillis = System.currentTimeMillis();
    LOG.info("Finished building StarTree, took {} ms", endMillis - startMillis);

    // Re-compute the unique values for metrics including aggregates to allow for dictionary
    // encoding
    LOG.info("Re-computing unique metric values for dictionary encoding...");
    startMillis = System.currentTimeMillis();
    Map<String, Set<Object>> uniqueMetricValues = computeUniqueMetricValues();
    resetMetricDictionaries(uniqueMetricValues);
    endMillis = System.currentTimeMillis();
    LOG.info("Finished re-computing unique metric values (took {} ms)", endMillis - startMillis);

    // StarTree directory
    starTreeDir = new File(outDir, V1Constants.STARTREE_DIR);
    if (!starTreeDir.mkdir()) {
      throw new RuntimeException(
          "Could not create star tree directory " + starTreeDir.getAbsolutePath());
    }

    // For each column, build its dictionary and initialize a forwards and an inverted index for raw
    // / agg segment
    int totalAggDocs = starTreeBuilder.getTotalAggregateDocumentCount();
    int totalRawDocs = starTreeBuilder.getTotalRawDocumentCount();
    for (final String column : dictionaryCreatorMap.keySet()) {
      ColumnIndexCreationInfo indexCreationInfo = columnInfo.get(column);

      Object[] uniqueValues = indexCreationInfo.getSortedUniqueElementsArray();
      if (schema.getMetricNames().contains(column)) {
        // Use the unique values including the new aggregate values
        uniqueValues = uniqueMetricValues.get(column).toArray();
      }

      if (schema.getFieldSpecFor(column).isSingleValueField()) {
        if (indexCreationInfo.isSorted()) {
          forwardIndexCreatorMap.put(
              column,
              new SingleValueSortedForwardIndexCreator(
                  outDir, uniqueValues.length, schema.getFieldSpecFor(column)));
          aggregateForwardIndexCreatorMap.put(
              column,
              new SingleValueSortedForwardIndexCreator(
                  starTreeDir, uniqueValues.length, schema.getFieldSpecFor(column)));
        } else {
          forwardIndexCreatorMap.put(
              column,
              new SingleValueUnsortedForwardIndexCreator(
                  schema.getFieldSpecFor(column),
                  outDir,
                  uniqueValues.length,
                  totalRawDocs,
                  indexCreationInfo.getTotalNumberOfEntries(),
                  indexCreationInfo.hasNulls()));
          aggregateForwardIndexCreatorMap.put(
              column,
              new SingleValueUnsortedForwardIndexCreator(
                  schema.getFieldSpecFor(column),
                  starTreeDir,
                  indexCreationInfo.getSortedUniqueElementsArray().length,
                  totalAggDocs,
                  indexCreationInfo.getTotalNumberOfEntries(),
                  indexCreationInfo.hasNulls()));
        }
      } else {
        forwardIndexCreatorMap.put(
            column,
            new MultiValueUnsortedForwardIndexCreator(
                schema.getFieldSpecFor(column),
                outDir,
                uniqueValues.length,
                totalRawDocs,
                indexCreationInfo.getTotalNumberOfEntries(),
                indexCreationInfo.hasNulls()));
        aggregateForwardIndexCreatorMap.put(
            column,
            new MultiValueUnsortedForwardIndexCreator(
                schema.getFieldSpecFor(column),
                starTreeDir,
                uniqueValues.length,
                totalAggDocs,
                indexCreationInfo.getTotalNumberOfEntries(),
                indexCreationInfo.hasNulls()));
      }

      if (config.createInvertedIndexEnabled()) {
        invertedIndexCreatorMap.put(
            column,
            new BitmapInvertedIndexCreator(
                outDir,
                indexCreationInfo.getSortedUniqueElementsArray().length,
                schema.getFieldSpecFor(column)));
        aggregateInvertedIndexCreatorMap.put(
            column,
            new BitmapInvertedIndexCreator(
                starTreeDir,
                indexCreationInfo.getSortedUniqueElementsArray().length,
                schema.getFieldSpecFor(column)));
      }
    }
  }
Ejemplo n.º 4
0
  @Override
  public void seal() throws ConfigurationException, IOException {
    // Write all the aggregate rows to the aggregate segment
    LOG.info("Writing aggregate segment...");
    long startMillis = System.currentTimeMillis();
    int currentAggregateDocumentId = 0;
    Iterator<StarTreeTableRow> itr = starTreeBuilder.getTable().getAllCombinations();
    while (itr.hasNext()) {
      StarTreeTableRow next = itr.next();
      if (next.getDimensions().contains(StarTreeIndexNode.all())) {
        // Write using that document ID to all columns
        for (final String column : dictionaryCreatorMap.keySet()) {
          Object dictionaryIndex = null; // TODO: Is this okay?

          if (starTreeDimensionDictionary.containsKey(column)) {
            // Index the dimension value
            Integer dimensionId = starTreeDimensionDictionary.get(column);
            Integer dimensionValue = next.getDimensions().get(dimensionId);
            if (dimensionValue == StarTreeIndexNode.all()) {
              // Use all value
              Object allValue = StarTreeIndexNode.getAllValue(schema.getFieldSpecFor(column));
              if (schema.getFieldSpecFor(column).isSingleValueField()) {
                dictionaryIndex = dictionaryCreatorMap.get(column).indexOfSV(allValue);
              } else {
                dictionaryIndex = dictionaryCreatorMap.get(column).indexOfMV(allValue);
              }
            } else {
              dictionaryIndex = dimensionValue;
            }
          } else if (starTreeMetricDictionary.containsKey(column)) {
            // Index the aggregate metric
            Integer metricId = starTreeMetricDictionary.get(column);
            Object columnValueToIndex = next.getMetrics().get(metricId);
            if (schema.getFieldSpecFor(column).isSingleValueField()) {
              dictionaryIndex = dictionaryCreatorMap.get(column).indexOfSV(columnValueToIndex);
            } else {
              dictionaryIndex = dictionaryCreatorMap.get(column).indexOfMV(columnValueToIndex);
            }
          } else {
            // Just index the raw value
            Object columnValueToIndex =
                StarTreeIndexNode.getAllValue(schema.getFieldSpecFor(column));
            if (schema.getFieldSpecFor(column).isSingleValueField()) {
              dictionaryIndex = dictionaryCreatorMap.get(column).indexOfSV(columnValueToIndex);
            } else {
              dictionaryIndex = dictionaryCreatorMap.get(column).indexOfMV(columnValueToIndex);
            }
          }

          if (schema.getFieldSpecFor(column).isSingleValueField()) {
            ((SingleValueForwardIndexCreator) aggregateForwardIndexCreatorMap.get(column))
                .index(currentAggregateDocumentId, (Integer) dictionaryIndex);
          } else {
            ((MultiValueForwardIndexCreator) aggregateForwardIndexCreatorMap.get(column))
                .index(currentAggregateDocumentId, (int[]) dictionaryIndex);
          }

          if (config.createInvertedIndexEnabled()) {
            aggregateInvertedIndexCreatorMap
                .get(column)
                .add(currentAggregateDocumentId, dictionaryIndex);
          }
        }
        currentAggregateDocumentId++;
      }
    }
    long endMillis = System.currentTimeMillis();
    LOG.info("Done writing aggregate segment (took {} ms)", endMillis - startMillis);

    for (final String column : forwardIndexCreatorMap.keySet()) {
      forwardIndexCreatorMap.get(column).close();
      if (config.createInvertedIndexEnabled()) {
        invertedIndexCreatorMap.get(column).seal();
      }
      dictionaryCreatorMap.get(column).close();
    }

    for (final String column : aggregateForwardIndexCreatorMap.keySet()) {
      aggregateForwardIndexCreatorMap.get(column).close();
      if (config.createInvertedIndexEnabled()) {
        aggregateInvertedIndexCreatorMap.get(column).seal();
      }
      // n.b. The dictionary from raw data is used
    }

    writeMetadata(outDir, starTreeBuilder.getTotalRawDocumentCount());

    // Write star tree
    LOG.info("Writing " + V1Constants.STARTREE_FILE);
    startMillis = System.currentTimeMillis();
    File starTreeFile = new File(starTreeDir, V1Constants.STARTREE_FILE);
    OutputStream starTreeOutputStream = new FileOutputStream(starTreeFile);
    starTreeBuilder.getTree().writeTree(starTreeOutputStream);
    starTreeOutputStream.close();
    endMillis = System.currentTimeMillis();
    LOG.info("Wrote StarTree file (took {} ms)", endMillis - startMillis);

    // Copy the dictionary files into startree directory
    // n.b. this is done so the segment is as stand-alone as possible, though could be removed as an
    // optimization
    File[] dictionaryFiles =
        outDir.listFiles(
            new FilenameFilter() {
              @Override
              public boolean accept(File dir, String name) {
                return name.endsWith(V1Constants.Dict.FILE_EXTENTION);
              }
            });
    for (File dictionaryFile : dictionaryFiles) {
      FileUtils.copyFile(dictionaryFile, new File(starTreeDir, dictionaryFile.getName()));
    }

    // Write star tree metadata
    writeMetadata(starTreeDir, starTreeBuilder.getTotalAggregateDocumentCount());
  }