Is your index reader really atomic or maybe slow?
Since the first day Apache Lucene exposed the two fundamental concepts of reading and writing an index directly through IndexReader & IndexWriter. However, the API did not reflect reality; from the IndexWriter perspective this was desirable but when reading the index this caused several problems in the past. In reality a Lucene index is not a single index while logically treated as a such. This talk will introduce the new API classes AtomicReader and CompositeReader added in Lucene 4.0 as very general interfaces, and DirectoryReader, which most people know as the segment-based “Lucene index on disk”. The talk will also cover more changes and improvements to the search API like reader contexts that allow to convert local document ids to global ones from IndexSearcher. Lucene changed all IndexReaders to be read-only, so it’s no longer possible to modify indexes using those classes. Finally, Uwe Schindler will show migration paths from custom norm values to the various new ranking models that were added to Lucene; this includes using Similarity with Lucene 4.0’s DocValues as replacement for norms.
Watch the video of Uwe Schindler's talk here.