lucene 2.4 变化修炼之 IndexerReader

大家用的最多的就是IndexReader上的open,其实在lucene里有DirectoryIndexerReader,关系如下:

 

abstract class DirectoryIndexReader extends IndexReader{}

 

其中定义的几个方法,请注意2.4和2.3.2的open方法的差别:

 

2.3.2:

static DirectoryIndexReader open(final Directory directory, final boolean closeDirectory, final IndexDeletionPolicy deletionPolicy) throws CorruptIndexException, IOException

 

2.4.0:

static DirectoryIndexReader open(final Directory directory, final boolean closeDirectory, final IndexDeletionPolicy deletionPolicy, final IndexCommit commit, final boolean readOnly) throws CorruptIndexException, IOException

 

实际上在2.4上赋予了Directory一个io核心的概念,我们可以看到所有的io操作最终都会转换成directory上的操作,因此才会在directory上赋予了commit和只读概念.

 

我们来看两个函数的变化

 

1 init:

在2.4中增加了如下的代码:

    if (!readOnly && segmentInfos != null) {
      // We assume that this segments_N was previously
      // properly sync'd:
      for(int i=0;i<segmentInfos.size();i++) {
        final SegmentInfo info = segmentInfos.info(i);
        List files = info.files();
        for(int j=0;j<files.size();j++)
          synced.add(files.get(j));
      }
    }

这段新加的代码在directory上增加了区段信息,区段信息可以做两件事情:

1 索引版本

2 是否优化

 

因此在reopen中2.4的代码得到了很多的改进,我曾经说过2.3里的reopen是个垃圾,因为它把所有的判断扔还给了开发人员,这是不合理的,下面看看reopen的新变化:

 

  public final synchronized IndexReader reopen() throws CorruptIndexException, IOException {
    ensureOpen();

    if (this.hasChanges || this.isCurrent()) {
      // this has changes, therefore we have the lock and don't need to reopen

这个应该就是针对changes中的:

LUCENE-1228: IndexWriter.commit() was not updating the index version and as result IndexReader.reopen() failed to sense index changes.
(Doron Cohen) 
      // OR: the index in the directory hasn't changed - nothing to do here
      return this;
    }

    return (DirectoryIndexReader) new SegmentInfos.FindSegmentsFile(directory) {

      protected Object doBody(String segmentFileName) throws CorruptIndexException, IOException {
        SegmentInfos infos = new SegmentInfos();
        infos.read(directory, segmentFileName);

        DirectoryIndexReader newReader = doReopen(infos);
       
        if (DirectoryIndexReader.this != newReader) {
          newReader.init(directory, infos, closeDirectory, readOnly);
          newReader.deletionPolicy = deletionPolicy;
        }

        return newReader;
      }
    }.run();
  }

 

最后看看open:

    SegmentInfos.FindSegmentsFile finder = new SegmentInfos.FindSegmentsFile(directory) {

      protected Object doBody(String segmentFileName) throws CorruptIndexException, IOException {

        SegmentInfos infos = new SegmentInfos();
        infos.read(directory, segmentFileName);

        DirectoryIndexReader reader;

        if (infos.size() == 1) {          // index is optimized
          reader = SegmentReader.get(readOnly, infos, infos.info(0), closeDirectory);
        } else if (readOnly) {
          reader = new ReadOnlyMultiSegmentReader(directory, infos, closeDirectory);
        } else {

原来代码中一旦没有优化就会使用多区段reader但是现在增加了ReadOnly版本,应该是commit策略有关的玩意,稍后我们来看看这东西好了.          reader = new MultiSegmentReader(directory, infos, closeDirectory, false);
        }
        reader.setDeletionPolicy(deletionPolicy);
        return reader;
      }
    };

 

    if (commit == null)
      return (DirectoryIndexReader) finder.run();
    else {
      if (directory != commit.getDirectory())
        throw new IOException("the specified commit does not match the specified Directory");
      // This can & will directly throw IOException if the
      // specified commit point has been deleted:
      return (DirectoryIndexReader) finder.doBody(commit.getSegmentsFileName());
    }

 

commit上 2.4中已经废止了原来的commitPoint类,改成了IndexCommit,作用类似主要方法定义如下:

 

   * Returns all index files referenced by this commit point.
   */
  public abstract Collection getFileNames() throws IOException;

  /**
   * Returns the {@link Directory} for the index.
   */
  public abstract Directory getDirectory();

 

增加如上的两个成员之后使得对于文件控制直接挂载到了directory,正像我说的2.4开始选择Directory为核心io处理了,呵呵.

 

最后来看看我们的readonlyMutiSegmentReader

 

class ReadOnlyMultiSegmentReader extends MultiSegmentReader {
  ReadOnlyMultiSegmentReader(Directory directory, SegmentInfos sis, boolean closeDirectory) throws IOException {
    super(directory, sis, closeDirectory, true);
  }

  ReadOnlyMultiSegmentReader(Directory directory, SegmentInfos infos, boolean closeDirectory, SegmentReader[] oldReaders, int[] oldStarts, Map oldNormsCache) throws IOException {
    super(directory, infos, closeDirectory, oldReaders, oldStarts, oldNormsCache, true);
  }

  protected void acquireWriteLock() {
    ReadOnlySegmentReader.noWrite();
  }
}

 

看来花头不多,只是改变了默认参数后需要增加这样一个类加以判断而已.今天就这些吧,下次继续...

 

分享到：

JAVA学习笔记之泛型接口 | lucene 2.4 变化修炼之前言

• 2008-11-10 17:57
• 浏览 2512
• 评论(0)
• 查看更多