beagled/Lucene.Net/upstream-changes/24_optimized-maybe-merge-segments.patch

   1 From: Joe Shaw <joeshaw@novell.com>
   2
   3 Optimize IndexWriter.MaybeMergeSegments() by caching the number of buffered
   4 documents rather than counting the number after each document addition.  This
   5 is a a backport from the Lucene devel 2.1 branch.
   6
   7 See https://issues.apache.org/jira/browse/LUCENE-388
   8
   9 Index: Index/IndexWriter.cs
  10 ===================================================================
  11 RCS file: /cvs/gnome/beagle/beagled/Lucene.Net/Index/IndexWriter.cs,v
  12 retrieving revision 1.4
  13 diff -u -p -u -r1.4 IndexWriter.cs
  14 --- Index/IndexWriter.cs        2 Oct 2006 17:08:56 -0000       1.4
  15 +++ Index/IndexWriter.cs        16 Oct 2006 20:41:55 -0000
  16 @@ -241,6 +241,8 @@ namespace Lucene.Net.Index
  17
  18                 private SegmentInfos segmentInfos = new SegmentInfos(); // the segments
  19                 private Directory ramDirectory = new RAMDirectory(); // for temp segs
  20 +
  21 +               private int singleDocSegmentsCount = 0; // for speeding decision on merge candidates
  22
  23                 private Lock writeLock;
  24
  25 @@ -618,6 +620,7 @@ namespace Lucene.Net.Index
  26                         lock (this)
  27                         {
  28                                 segmentInfos.Add(new SegmentInfo(segmentName, 1, ramDirectory));
  29 +                               singleDocSegmentsCount++;
  30                                 MaybeMergeSegments();
  31                         }
  32                 }
  33 @@ -815,8 +818,8 @@ namespace Lucene.Net.Index
  34                         while (targetMergeDocs <= maxMergeDocs)
  35                         {
  36                                 // find segments smaller than current target size
  37 -                               int minSegment = segmentInfos.Count;
  38 -                               int mergeDocs = 0;
  39 +                               int minSegment = segmentInfos.Count - singleDocSegmentsCount; // top 1-doc segments are taken for sure
  40 +                               int mergeDocs = singleDocSegmentsCount;
  41                                 while (--minSegment >= 0)
  42                                 {
  43                                         SegmentInfo si = segmentInfos.Info(minSegment);
  44 @@ -825,10 +828,11 @@ namespace Lucene.Net.Index
  45                                         mergeDocs += si.docCount;
  46                                 }
  47
  48 -                               if (mergeDocs >= targetMergeDocs)
  49 +                               if (mergeDocs >= targetMergeDocs) {
  50                                 // found a merge to do
  51                                         MergeSegments(minSegment + 1);
  52 -                               else
  53 +                                       singleDocSegmentsCount = 0;
  54 +                               } else
  55                                         break;
  56
  57                                 targetMergeDocs *= mergeFactor; // increase target size