IndexSort

最新推荐文章于 2024-02-20 14:32:49 发布

chuanyangwang

最新推荐文章于 2024-02-20 14:32:49 发布

阅读量142

点赞数

分类专栏： ES 文章标签： lucene apache 全文检索

本文链接：https://blog.csdn.net/chuanyangwang/article/details/121663585

版权

Lucene 排序函数文档比较器搜索优化早期终止

关键词由CSDN通过智能技术生成

ES 专栏收录该内容

50 篇文章 0 订阅

订阅专栏

排序函数获取

/** We cannot use the {@link FieldComparator} API because that API requires that you send it docIDs in order.  Note that this API
   *  allocates arrays[maxDoc] to hold the native values needed for comparison, but 1) they are transient (only alive while sorting this one
   *  segment), and 2) in the typical index sorting case, they are only used to sort newly flushed segments, which will be smaller than
   *  merged segments.  */
  static DocComparator getDocComparator(int maxDoc,
                                        SortField sortField,
                                        SortedDocValuesSupplier sortedProvider,
                                        NumericDocValuesSupplier numericProvider) throws IOException {

    final int reverseMul = sortField.getReverse() ? -1 : 1;
    final SortField.Type sortType = getSortFieldType(sortField);

    switch(sortType) {

      case STRING:
      {
        final SortedDocValues sorted = sortedProvider.get();
        final int missingOrd;
        if (sortField.getMissingValue() == SortField.STRING_LAST) {
          missingOrd = Integer.MAX_VALUE;
        } else {
          missingOrd = Integer.MIN_VALUE;
        }

        final int[] ords = new int[maxDoc];
        Arrays.fill(ords, missingOrd);
        int docID;
        while ((docID = sorted.nextDoc()) != NO_MORE_DOCS) {
          ords[docID] = sorted.ordValue();
        }

        return new DocComparator() {
          @Override
          public int compare(int docID1, int docID2) {
            return reverseMul * Integer.compare(ords[docID1], ords[docID2]);
          }
        };
      }

      case LONG:
      {
        final NumericDocValues dvs = numericProvider.get();
        long[] values = new long[maxDoc];
        if (sortField.getMissingValue() != null) {
          Arrays.fill(values, (Long) sortField.getMissingValue());
        }
        while (true) {
          int docID = dvs.nextDoc();
          if (docID == NO_MORE_DOCS) {
            break;
          }
          values[docID] = dvs.longValue();
        }

        return new DocComparator() {
          @Override
          public int compare(int docID1, int docID2) {
            return reverseMul * Long.compare(values[docID1], values[docID2]);
          }
        };
      }

      case INT:
      {
        final NumericDocValues dvs = numericProvider.get();
        int[] values = new int[maxDoc];
        if (sortField.getMissingValue() != null) {
          Arrays.fill(values, (Integer) sortField.getMissingValue());
        }

        while (true) {
          int docID = dvs.nextDoc();
          if (docID == NO_MORE_DOCS) {
            break;
          }
          values[docID] = (int) dvs.longValue();
        }

        return new DocComparator() {
          @Override
          public int compare(int docID1, int docID2) {
            return reverseMul * Integer.compare(values[docID1], values[docID2]);
          }
        };
      }

      case DOUBLE:
      {
        final NumericDocValues dvs = numericProvider.get();
        double[] values = new double[maxDoc];
        if (sortField.getMissingValue() != null) {
          Arrays.fill(values, (Double) sortField.getMissingValue());
        }
        while (true) {
          int docID = dvs.nextDoc();
          if (docID == NO_MORE_DOCS) {
            break;
          }
          values[docID] = Double.longBitsToDouble(dvs.longValue());
        }

        return new DocComparator() {
          @Override
          public int compare(int docID1, int docID2) {
            return reverseMul * Double.compare(values[docID1], values[docID2]);
          }
        };
      }

      case FLOAT:
      {
        final NumericDocValues dvs = numericProvider.get();
        float[] values = new float[maxDoc];
        if (sortField.getMissingValue() != null) {
          Arrays.fill(values, (Float) sortField.getMissingValue());
        }
        while (true) {
          int docID = dvs.nextDoc();
          if (docID == NO_MORE_DOCS) {
            break;
          }
          values[docID] = Float.intBitsToFloat((int) dvs.longValue());
        }

        return new DocComparator() {
          @Override
          public int compare(int docID1, int docID2) {
            return reverseMul * Float.compare(values[docID1], values[docID2]);
          }
        };
      }

      default:
        throw new IllegalArgumentException("unhandled SortField.getType()=" + sortField.getType());
    }
  }

是否可以快速结束。判断searchSort和indexSort是否是一致的。

org.apache.lucene.search.TopFieldCollector canEarlyTerminate

LeafMetaData 中包含了indexSort信息

this.metaData = new LeafMetaData(createdVersionMajor, si.info.getMinVersion(), si.info.getIndexSort());

  static boolean canEarlyTerminate(Sort searchSort, Sort indexSort) {
    final SortField[] fields1 = searchSort.getSort();
    final SortField[] fields2 = indexSort.getSort();
    // early termination is possible if fields1 is a prefix of fields2
    if (fields1.length > fields2.length) {
      return false;
    }
    return Arrays.asList(fields1).equals(Arrays.asList(fields2).subList(0, fields1.length));
  }