nutch1.2 DeleteDuplicates IndexMerger 详解



job 1 
    map: 默认Mapper ,
    输出为key:Text url value :IndexDoc

    job.setInputFormat(InputFormat.class);  关键在于这个Format会把lucene的doc转换成IndexDoc 代码如下:


public boolean next(Text key, IndexDoc indexDoc)
        throws IOException {
       
        // skip empty indexes
        if (indexReader == null || maxDoc <= 0)
          return false;

        // skip deleted documents
        while (doc < maxDoc && indexReader.isDeleted(doc)) doc++;
        if (doc >= maxDoc)
          return false;

        Document document = indexReader.document(doc);

        // fill in key
        key.set(document.get("url"));
        // fill in value
        indexDoc.keep = true;
        indexDoc.url.set(document.get("url"));
        indexDoc.hash.setDigest(document.get("digest"));
        indexDoc.score = Float.parseFloat(document.get("boost"));
        try {
          indexDoc.time = DateTools.stringToTime(document.get("tstamp"));
        } catch (Exception e) {
          // try to figure out the time from segment name
          try {
            String segname = document.get("segment");
            indexDoc.time = new SimpleDateFormat("yyyyMMddHHmmss").parse(segname).getTime();
            // make it unique
            indexDoc.time += doc;
          } catch (Exception e1) {
            // use current time
            indexDoc.time = System.currentTimeMillis();
          }
        }
        indexDoc.index = index;
        indexDoc.doc = doc;

        doc++;

        return true;
      }


reduce :UrlsReducer  为index的数据去重复准备数据,一个url只能有一个值
   
        1 对同一个key:如果有多个value,比较每个value的time,取最大的time,小于的数据设置 latest.keep = false; 写入,最后写入 latest.keep = true;的数据
          2 key:Text 是签名,value:IndexDoc


job2  map: 默认Mapper 
    map out的数据    key:MD5Hash  value:IndexDoc
       reduce:HashReducer   写入的数据都是要删除的
      1 对数据如果IndexDoc.keep=false 写入
       2 如果IndexDoc.keep=ture的多个值,如果dedup.keep.highest.score为true说明按照score比较,否则安装url的长度比较。如果按照score比较,score分数小的写入,否则删除url的长度长的


job 3
   map : DeleteDuplicates
        1 对keep不为true的数据写入
           key: 对应的目录 value:lucene docId


   reduce : DeleteDuplicates
    使用indexreader做删除操作

    代码如下

Path index = new Path(key.toString());
    IndexReader reader = IndexReader.open(new FsDirectory(fs, index, false, getConf()), false);
    try {
      while (values.hasNext()) {
        IntWritable value = values.next();
        LOG.debug("-delete " + index + " doc=" + value);
        reader.deleteDocument(value.get());
      }
    } finally {
      reader.close();
    }


merger索引IndexMerger
用构建crawl/indexes 下面的所有Directory
Directory[] dirs = new Directory[indexes.length];
    for (int i = 0; i < indexes.length; i++) {
      if (LOG.isInfoEnabled()) { LOG.info("Adding " + indexes[i]); }
      dirs[i] = new FsDirectory(fs, indexes[i], false, getConf());
    }


merger 使用IndexWriter  的addIndexesNoOptimize 进行merger

          //
    // Merge indices
    //
    IndexWriter writer = new IndexWriter(
    FSDirectory.open(new File(localOutput.toString())), null, true,
    MaxFieldLength.UNLIMITED);
    writer.setMergeFactor(getConf().getInt("indexer.mergeFactor", LogMergePolicy.DEFAULT_MERGE_FACTOR));
    writer.setMaxBufferedDocs(getConf().getInt("indexer.minMergeDocs", IndexWriter.DEFAULT_MAX_BUFFERED_DOCS));
    writer.setMaxMergeDocs(getConf().getInt("indexer.maxMergeDocs", LogMergePolicy.DEFAULT_MAX_MERGE_DOCS));
    writer.setTermIndexInterval(getConf().getInt("indexer.termIndexInterval", IndexWriter.DEFAULT_TERM_INDEX_INTERVAL));
    writer.setInfoStream(LogUtil.getDebugStream(LOG));
    writer.setUseCompoundFile(false);
    writer.setSimilarity(new NutchSimilarity());
    writer.addIndexesNoOptimize(dirs);
    writer.optimize();
    writer.close();

最后使用
fs.completeLocalOutput(outputIndex, tmpLocalOutput);
生成crawl/index 至此nutch索引的部分全部完

猜你喜欢

转载自chengqianl.iteye.com/blog/1597787
今日推荐