http://issues.apache.org/jira/browse/LUCENENET-103
The story explains itself:
The objective for cloning was to make it more performant. The Directory copy approach was slower. For our purposes, the difference was thirty seconds for a manual RAMDirectory duplication using IndexWriter.AddIndexes(), 1299ms for Directory.Copy, versus 669ms for a deep clone, and 47.66ms for a shallow clone (and a LOT less RAM usage). We are going with a shallow clone because this is a multi-threaded server and there are thread locks all over the Lucene objects, but we don't modify a RAMDirectory once it is loaded. Rather, we rebuild the RAMDirectory in the equivalent of a cron job, then clone it across multiple threads.
Subscribe