I've found the problem. When Nhibernate.Search creates writer it does not load
settings for writer appropriately so that writer is always initialized as
following:
wr.SetMergeFactor(10);
wr.SetMaxMergeDocs(-2147483648);
wr.SetMaxBufferedDocs(10);
After launching sample program (2000 documents, 5000 updates) with these
parameters I get 250 segments (501 file in index, 7 minutes 30 seconds). Here
is a sample program. I wrote it as close as possible to the way
nhibernate.search uses lucene:
class Program
{
static string indexDirectory = @"d:\lucene test";
static string className = "myclass";
static Document CreateDocument(int id)
{
Document doc = new Document();
doc.Add(new Field("type", className,
Field.Store.YES, Field.Index.UN_TOKENIZED));
doc.Add(new Field("id", id.ToString(),
Field.Store.YES, Field.Index.UN_TOKENIZED));
doc.Add(new Field("field", "this is a not very
long, quite simple string", Field.Store.NO, Field.Index.TOKENIZED));
return doc;
}
static void Main(string[] args)
{
// create index
IndexWriter createWriter = new
IndexWriter(indexDirectory, new WhitespaceAnalyzer(), true);
createWriter.Close();
// adding 2000 documents and update one of them
5000 times
for (int i = 0; i < 5000; i++)
{
// removing document from index
Term term = new Term("id", "0");
IndexReader indexReader =
IndexReader.Open(indexDirectory);
TermDocs docs =
indexReader.TermDocs(term);
while (docs.Next())
{
int n =
docs.Doc();
if
(className.Equals(indexReader.Document(n).Get("type")))
indexReader.DeleteDocument(n);
}
docs.Close();
indexReader.Close();
// adding documents to index
IndexWriter wr = new
IndexWriter(indexDirectory, new WhitespaceAnalyzer(), false);
wr.AddDocument(CreateDocument(0));
if (i < 2000)
{
wr.AddDocument(CreateDocument(i));
}
wr.Close();
}
}
}
After setting MaxMergeDocs to 10000 I get 11 segments (23 files, 1 minute 12
seconds)with looks closer to the truth. But still should not there be 10
segments?
