John Leach
John Leach
It should be possible to have a small map of the possible different schemas in case of missing fields etc, without taking the scalability penalty of fetching for each segment.
Need to add a supplier on the Input and a custom close on the output. I will need this functionality for implementing a more complex field info.
LiveDocsFormat will perform a read of the bit vector in the case where we are doing a simple insert.
LiveDocsFormat needs to be lazy as well in cases where we have a large number of deletes. Need additional lazy semantics so single row insert does not perform reads of...
Significant performance issue.
This allows the two implementations to work on a per field basis and to not fetch blocks for positional elements unless they are needed to answer the query.
We always read block 0, even when not needed. This adds about 33% more reads to the simple insert record case.
Similar to other formats, make sure we do not perform a read during initialization.
Continuing the work where the IndexWriter does not need for example to pre-fetch data for StoredFields.
To significantly reduce I/O, make the readers initialize lazily.
The PointsReader does this for each segment open... `public Lucene86PointsReader(SegmentReadState readState) throws IOException { this.readState = readState; String metaFileName = IndexFileNames.segmentFileName(readState.segmentInfo.name, readState.segmentSuffix, Lucene86PointsFormat.META_EXTENSION); String indexFileName = IndexFileNames.segmentFileName(readState.segmentInfo.name, readState.segmentSuffix, Lucene86PointsFormat.INDEX_EXTENSION); String...