Package org.apache.lucene.codecs.lucene60

Class Lucene60PointsFormat

  • public final class Lucene60PointsFormat
extends PointsFormat
    Lucene 6.0 point format, which encodes dimensional values in a block KD-tree structure for fast 1D range and N dimesional shape intersection filtering. See this paper for details.

    This data structure is written as a series of blocks on disk, with an in-memory perfectly balanced binary tree of split values referencing those blocks at the leaves.

    The .dim file has both blocks and the index split values, for each field. The file starts with CodecUtil.writeIndexHeader(org.apache.lucene.store.DataOutput, java.lang.String, int, byte[], java.lang.String).

    The blocks are written like this:

    • count (vInt)
    • delta-docID (vInt) count (delta coded docIDs, in sorted order)
    • packedValuecount (the byte[] value of each dimension packed into a single byte[])

    After all blocks for a field are written, then the index is written:

    • numDims (vInt)
    • maxPointsInLeafNode (vInt)
    • bytesPerDim (vInt)
    • count (vInt)
    • packed index (byte[])

    The packed index uses hierarchical delta and prefix coding to compactly encode the file pointer for all leaf blocks, once the tree is traversed, as well as the split dimension and split value for each inner node of the tree.

    After all fields blocks + index data are written, CodecUtil.writeFooter(org.apache.lucene.store.IndexOutput) writes the checksum.

    The .dii file records the file pointer in the .dim file where each field's index data was written. It starts with CodecUtil.writeIndexHeader(org.apache.lucene.store.DataOutput, java.lang.String, int, byte[], java.lang.String), then has:

    • fieldCount (vInt)
    • (fieldNumber (vInt), fieldFilePointer (vLong))fieldCount

    After all fields blocks + index data are written, CodecUtil.writeFooter(org.apache.lucene.store.IndexOutput) writes the checksum.

    WARNING: This API is experimental and might change in incompatible ways in the next release.

    • Constructor Detail

      • Lucene60PointsFormat

        public Lucene60PointsFormat()
        Sole constructor

    • Method Detail

      • fieldsReader

        public PointsReader fieldsReader​(SegmentReadState state)
                          throws IOException
        Description copied from class: PointsFormat
        Reads a segment. NOTE: by the time this call returns, it must hold open any files it will need to use; else, those files may be deleted. Additionally, required files may be deleted during the execution of this call before there is a chance to open them. Under these circumstances an IOException should be thrown by the implementation. IOExceptions are expected and will automatically cause a retry of the segment opening logic with the newly revised segments.
        Specified by:
        fieldsReader in class PointsFormat
        Throws:
        IOException